repo
stringlengths
2
99
file
stringlengths
13
225
code
stringlengths
0
18.3M
file_length
int64
0
18.3M
avg_line_length
float64
0
1.36M
max_line_length
int64
0
4.26M
extension_type
stringclasses
1 value
GalaxyDataset
GalaxyDataset-master/evaluationMetrics.py
0
0
0
py
GalaxyDataset
GalaxyDataset-master/preprocess.py
# -*- coding: utf-8 -*- import torch import torch.utils.data as Data import numpy as np from torchvision import datasets, transforms import argparse import os import random import yaml import downloadData def load_npy(path): # npy file: [[imgs, label], [imgs, label]...., [imgs, label]] # when allow_pickle=True, matrix needs same size if not os.path.isfile(path): print("files do not exists!!") return np_array = np.load(path, allow_pickle=True) imgs = [] label = [] for index in range(len(np_array)): imgs.append(np_array[index][0]) label.append(np_array[index][1]) torch_dataset = Data.TensorDataset(torch.from_numpy(np.array(imgs)), torch.from_numpy(np.array(label))) train_loader = Data.DataLoader( torch_dataset, batch_size=64, shuffle=True, num_workers=1 ) transform_test = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ]) test_loader = torch.utils.data.DataLoader( datasets.CIFAR10('data', train=False, transform=transform_test), batch_size=32, shuffle=True, num_workers=1 ) print("train_loader, test_loader generated succeed!") return train_loader, test_loader if __name__ == "__main__": dataloder = load_npy("./cifar10/splitByLabelsWithNormalAndErrorDataset/SplitByLabels_3666_truck.npy")
1,436
29.574468
107
py
GalaxyDataset
GalaxyDataset-master/digitfive/custom.py
from PIL import Image, ImageEnhance, ImageOps, ImageFilter import numpy as np import random class GaussianBlur(object): def __init__(self, sigma=None): if sigma is None: sigma = [.1, 2.] self.sigma = sigma def __call__(self, x): sigma = random.uniform(self.sigma[0], self.sigma[1]) x = x.filter(ImageFilter.GaussianBlur(radius=sigma)) return x class SubPolicy(object): def __init__(self, p1, operation1, magnitude_idx1, p2, operation2, magnitude_idx2, fillcolor=(128, 128, 128)): ranges = { "shearX": np.linspace(0, 0.3, 10), "shearY": np.linspace(0, 0.3, 10), "translateX": np.linspace(0, 150 / 331, 10), "translateY": np.linspace(0, 150 / 331, 10), "rotate": np.linspace(0, 30, 10), "color": np.linspace(0.0, 0.9, 10), "posterize": np.round(np.linspace(8, 4, 10), 0).astype(np.int), "solarize": np.linspace(256, 0, 10), "contrast": np.linspace(0.0, 0.9, 10), "sharpness": np.linspace(0.0, 0.9, 10), "brightness": np.linspace(0.0, 0.9, 10), "autocontrast": [0] * 10, "equalize": [0] * 10, "invert": [0] * 10 } # from https://stackoverflow.com/questions/5252170/specify-image-filling-color-when-rotating-in-python-with-pil-and-setting-expand def rotate_with_fill(img, magnitude): rot = img.convert("RGBA").rotate(magnitude) return Image.composite(rot, Image.new("RGBA", rot.size, (128,) * 4), rot).convert(img.mode) func = { "shearX": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, magnitude * random.choice([-1, 1]), 0, 0, 1, 0), Image.BICUBIC, fillcolor=fillcolor), "shearY": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, 0, magnitude * random.choice([-1, 1]), 1, 0), Image.BICUBIC, fillcolor=fillcolor), "translateX": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, magnitude * img.size[0] * random.choice([-1, 1]), 0, 1, 0), fillcolor=fillcolor), "translateY": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, 0, 0, 1, magnitude * img.size[1] * random.choice([-1, 1])), fillcolor=fillcolor), "rotate": lambda img, magnitude: rotate_with_fill(img, magnitude), "color": lambda img, magnitude: ImageEnhance.Color(img).enhance(1 + magnitude * random.choice([-1, 1])), "posterize": lambda img, magnitude: ImageOps.posterize(img, magnitude), "solarize": lambda img, magnitude: ImageOps.solarize(img, magnitude), "contrast": lambda img, magnitude: ImageEnhance.Contrast(img).enhance( 1 + magnitude * random.choice([-1, 1])), "sharpness": lambda img, magnitude: ImageEnhance.Sharpness(img).enhance( 1 + magnitude * random.choice([-1, 1])), "brightness": lambda img, magnitude: ImageEnhance.Brightness(img).enhance( 1 + magnitude * random.choice([-1, 1])), "autocontrast": lambda img, magnitude: ImageOps.autocontrast(img), "equalize": lambda img, magnitude: ImageOps.equalize(img), "invert": lambda img, magnitude: ImageOps.invert(img) } self.p1 = p1 self.operation1 = func[operation1] self.magnitude1 = ranges[operation1][magnitude_idx1] self.p2 = p2 self.operation2 = func[operation2] self.magnitude2 = ranges[operation2][magnitude_idx2] def __call__(self, img): if random.random() < self.p1: img = self.operation1(img, self.magnitude1) if random.random() < self.p2: img = self.operation2(img, self.magnitude2) return img class CIFAR10Policy(object): """ Randomly choose one of the best 25 Sub-policies on CIFAR10. Example: >>> policy = CIFAR10Policy() >>> transformed = policy(image) Example as a PyTorch Transform: >>> transform=transforms.Compose([ >>> transforms.Resize(256), >>> CIFAR10Policy(), >>> transforms.ToTensor()]) """ def __init__(self, fillcolor=(128, 128, 128)): self.policies = [ SubPolicy(0.1, "invert", 7, 0.2, "contrast", 6, fillcolor), SubPolicy(0.7, "rotate", 2, 0.3, "translateX", 9, fillcolor), SubPolicy(0.8, "sharpness", 1, 0.9, "sharpness", 3, fillcolor), SubPolicy(0.5, "shearY", 8, 0.7, "translateY", 9, fillcolor), SubPolicy(0.5, "autocontrast", 8, 0.9, "equalize", 2, fillcolor), SubPolicy(0.2, "shearY", 7, 0.3, "posterize", 7, fillcolor), SubPolicy(0.4, "color", 3, 0.6, "brightness", 7, fillcolor), SubPolicy(0.3, "sharpness", 9, 0.7, "brightness", 9, fillcolor), SubPolicy(0.6, "equalize", 5, 0.5, "equalize", 1, fillcolor), SubPolicy(0.6, "contrast", 7, 0.6, "sharpness", 5, fillcolor), SubPolicy(0.7, "color", 7, 0.5, "translateX", 8, fillcolor), SubPolicy(0.3, "equalize", 7, 0.4, "autocontrast", 8, fillcolor), SubPolicy(0.4, "translateY", 3, 0.2, "sharpness", 6, fillcolor), SubPolicy(0.9, "brightness", 6, 0.2, "color", 8, fillcolor), SubPolicy(0.5, "solarize", 2, 0.0, "invert", 3, fillcolor), SubPolicy(0.2, "equalize", 0, 0.6, "autocontrast", 0, fillcolor), SubPolicy(0.2, "equalize", 8, 0.6, "equalize", 4, fillcolor), SubPolicy(0.9, "color", 9, 0.6, "equalize", 6, fillcolor), SubPolicy(0.8, "autocontrast", 4, 0.2, "solarize", 8, fillcolor), SubPolicy(0.1, "brightness", 3, 0.7, "color", 0, fillcolor), SubPolicy(0.4, "solarize", 5, 0.9, "autocontrast", 3, fillcolor), SubPolicy(0.9, "translateY", 9, 0.7, "translateY", 9, fillcolor), SubPolicy(0.9, "autocontrast", 2, 0.8, "solarize", 3, fillcolor), SubPolicy(0.8, "equalize", 8, 0.1, "invert", 3, fillcolor), SubPolicy(0.7, "translateY", 9, 0.9, "autocontrast", 1, fillcolor) ] def __call__(self, img): policy_idx = random.randint(0, len(self.policies) - 1) return self.policies[policy_idx](img) def __repr__(self): return "AutoAugment CIFAR10 Policy"
6,520
48.030075
138
py
GalaxyDataset
GalaxyDataset-master/digitfive/femnist.py
from torchvision.datasets import MNIST, utils from PIL import Image import os.path import torch class FEMNIST(MNIST): """ This dataset is derived from the Leaf repository (https://github.com/TalwalkarLab/leaf) pre-processing of the Extended MNIST dataset, grouping examples by writer. Details about Leaf were published in "LEAF: A Benchmark for Federated Settings" https://arxiv.org/abs/1812.01097. """ resources = [ ('https://raw.githubusercontent.com/tao-shen/FEMNIST_pytorch/master/femnist.tar.gz', '59c65cec646fc57fe92d27d83afdf0ed')] def __init__(self, root, train=True, transform=None, target_transform=None, download=False): super(MNIST, self).__init__(root, transform=transform, target_transform=target_transform) self.train = train if download: self.download() if not self._check_exists(): raise RuntimeError('Dataset not found.' + ' You can use download=True to download it') if self.train: data_file = self.training_file else: data_file = self.test_file self.data, self.targets, self.users_index = torch.load(os.path.join(self.processed_folder, data_file)) # print(len(self.data)) # print(len(self.targets)) # print(len(self.users_index)) self.uid = [] i = 0 for n in self.users_index: for j in range(n): self.uid.append(i) i += 1 # print(len(self.uid)) # def __getitem__(self, index): # img, target = self.data[index], int(self.targets[index]) # img = Image.fromarray(img.numpy(), mode='F') # if self.transform is not None: # img = self.transform(img) # if self.target_transform is not None: # target = self.target_transform(target) # return img, target def __getitem__(self, index): img, target, uid = self.data[index], int(self.targets[index]), int(self.uid[index]) img = Image.fromarray(img.numpy(), mode='F') if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) return img, target def download(self): """Download the FEMNIST data if it doesn't exist in processed_folder already.""" import shutil if self._check_exists(): return utils.makedir_exist_ok(self.raw_folder) utils.makedir_exist_ok(self.processed_folder) # download files for url, md5 in self.resources: filename = url.rpartition('/')[2] utils.download_and_extract_archive(url, download_root=self.raw_folder, filename=filename, md5=md5) # process and save as torch files print('Processing...') shutil.move(os.path.join(self.raw_folder, self.training_file), self.processed_folder) shutil.move(os.path.join(self.raw_folder, self.test_file), self.processed_folder)
3,128
36.25
110
py
GalaxyDataset
GalaxyDataset-master/digitfive/usps.py
import scipy.io as scio import numpy as np from PIL import Image import os import os.path import torch import torchvision from torchvision import datasets, transforms from torchvision.datasets import MNIST, utils from torch.utils.data import DataLoader, Dataset # dataFile = 'usps_28x28.mat' # data = scio.loadmat(dataFile) # # for k in data.keys(): # # print(k) # # __header__ # # __version__ # # __globals__ # # dataset # dataset_training = data['dataset'][0] # dataset_test = data['dataset'][1] # # a = dataset_training[0] # data # # print(type(a)) # numpy # # print(len(a)) # 7438 # # print(len(a[0])) # 1 # # print(len(a[0][0])) # 28 # # print(len(a[0][0][0])) # 28 # # b = dataset_training[1] # targets # # print(len(b)) # 7438 # # print(len(b[0])) # 1 # training_data = [] # for img in dataset_training[0]: # img = img * 255 # img = img.tolist() # temp = img[0] # img.append(temp) # img.append(temp) # img = torch.Tensor(img) # img = img.permute(1, 2, 0) # # print(img.size()) # 28 28 3 # training_data.append(img) # # print(len(temp)) # # print(len(temp[0])) # # print(len(temp[0][0])) # training_targets = [] # for label in dataset_training[1]: # training_targets.append(label[0]) # # print(label[0]) # torch.save((training_data, training_targets), 'USPS/processed/training.pt') # test_data = [] # for img in dataset_test[0]: # img = img * 255 # img = img.tolist() # temp = img[0] # img.append(temp) # img.append(temp) # img = torch.Tensor(img) # img = img.permute(1, 2, 0) # # print(img.size()) # 28 28 3 # test_data.append(img) # # print(len(temp)) # # print(len(temp[0])) # # print(len(temp[0][0])) # test_targets = [] # for label in dataset_test[1]: # test_targets.append(label[0]) # # print(label[0]) # torch.save((test_data, test_targets), 'USPS/processed/test.pt') class USPS(MNIST): def __init__(self, *args, **kwargs): super(USPS, self).__init__(*args, **kwargs) def __getitem__(self, index): # print('type: ',type(self.data)) # print('len: ',len(self.data)) img, target = self.data[index], int(self.targets[index]) # print('type of img: ', type(torch.Tensor(img))) # print('type of img: ', type(img)) # print('img size', torch.Tensor(img).size()) # return a PIL Image img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # mode & permute # print('img: ', img) # print('img size', img.size()) if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) # print(img.size()) return img, target def digit_five_train_transforms(): all_transforms = transforms.Compose([ # transforms.RandomResizedCrop(28), # ToPILImage # transforms.RandomApply([transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8), # transforms.RandomGrayscale(p=0.2), # transforms.RandomAffine(degrees=15, # translate=[0.1, 0.1], # scale=[0.9, 1.1], # shear=15), transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms def digit_five_test_transforms(): all_transforms = transforms.Compose([ transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms class Loader(object): def __init__(self, dataset_ident, file_path='', download=False, batch_size=128, train_transform=digit_five_train_transforms(), test_transform=digit_five_test_transforms(), target_transform=None, use_cuda=False): kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {} loader_map = { # 'MNIST': MNIST, # 'MNISTM': MNISTM, # 'SVHN': SVHN, # 'SYN': SYN, 'USPS': USPS, # 'MNISTC': MNISTC, } num_class = { # 'MNIST': 10, # 'MNISTM': 10, # 'SVHN': 10, # 'SYN': 10, 'USPS': 10, # 'MNISTC': 10, } # Get the datasets self.train_dataset, self.test_dataset = self.get_dataset(loader_map[dataset_ident], file_path, download, train_transform, test_transform, target_transform) # Set the loaders self.train_loader = DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, **kwargs) self.test_loader = DataLoader(self.test_dataset, batch_size=batch_size, shuffle=False, **kwargs) tmp_batch = self.train_loader.__iter__().__next__()[0] self.img_shape = list(tmp_batch.size())[1:] self.num_class = num_class[dataset_ident] @staticmethod def get_dataset(dataset, file_path, download, train_transform, test_transform, target_transform): # Training and Validation datasets train_dataset = dataset(file_path, train=True, download=download, transform=train_transform, target_transform=target_transform) test_dataset = dataset(file_path, train=False, download=download, transform=test_transform, target_transform=target_transform) return train_dataset, test_dataset # loader = Loader('USPS') # dataset_train = loader.train_dataset # img = dataset_train[50][0] # print(dataset_train[50][1]) # img = img * 255 # print(img) # img = Image.fromarray(np.array(img.permute(1, 2, 0)).astype('uint8'), mode='RGB') # img.show()
5,890
30.502674
215
py
GalaxyDataset
GalaxyDataset-master/digitfive/svhn.py
import scipy.io as scio import numpy as np from PIL import Image import os import os.path import torch import torchvision from torchvision import datasets, transforms from torchvision.datasets import MNIST, utils from torch.utils.data import DataLoader, Dataset # dataFile = 'svhn_train_32x32.mat' # data = scio.loadmat(dataFile) # # for k in data.keys(): # # print(k) # # __header__ # # __version__ # # __globals__ # # X # # y # data['X'] = data['X'].transpose(3, 0, 1, 2) # a = data['X'] # print(len(a)) # 73257 # print(len(a[0])) # 32 # print(len(a[0][0])) # 32 # print(len(a[0][0][0])) # 3 # training_data = [] # for img in data['X']: # img = torch.from_numpy(img) # training_data.append(img) # training_targets = [] # for label in data['y']: # if label[0] == 10: # l = 0 # else: # l = label[0] # training_targets.append(l) # torch.save((training_data, training_targets), 'SVHN/processed/training.pt') # dataFile = 'svhn_test_32x32.mat' # data = scio.loadmat(dataFile) # # for k in data.keys(): # # print(k) # # __header__ # # __version__ # # __globals__ # # X # # y # data['X'] = data['X'].transpose(3, 0, 1, 2) # a = data['X'] # print(len(a)) # 26032 # print(len(a[0])) # 32 # print(len(a[0][0])) # 32 # print(len(a[0][0][0])) # 3 # test_data = [] # for img in data['X']: # img = torch.from_numpy(img) # test_data.append(img) # test_targets = [] # for label in data['y']: # if label[0] == 10: # l = 0 # else: # l = label[0] # test_targets.append(l) # torch.save((test_data, test_targets), 'SVHN/processed/test.pt') class SVHN(MNIST): def __init__(self, *args, **kwargs): super(SVHN, self).__init__(*args, **kwargs) def __getitem__(self, index): # print('type: ',type(self.data)) # print('len: ',len(self.data)) img, target = self.data[index], int(self.targets[index]) # print('type of img: ', type(torch.Tensor(img))) # print('img size', torch.Tensor(img).size()) # return a PIL Image img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # mode & permute # print('img: ', img) # print('img size', img.size()) if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) # print(img.size()) return img, target def digit_five_train_transforms(): all_transforms = transforms.Compose([ # transforms.RandomResizedCrop(28), # transforms.RandomApply([transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8), # transforms.RandomGrayscale(p=0.2), # transforms.RandomAffine(degrees=15, # translate=[0.1, 0.1], # scale=[0.9, 1.1], # shear=15), transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms def digit_five_test_transforms(): all_transforms = transforms.Compose([ transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms class Loader(object): def __init__(self, dataset_ident, file_path='', download=False, batch_size=128, train_transform=digit_five_train_transforms(), test_transform=digit_five_test_transforms(), target_transform=None, use_cuda=False): kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {} loader_map = { # 'MNIST': MNIST, # 'MNISTM': MNISTM, 'SVHN': SVHN, # 'SYN': SYN, # 'USPS': USPS, # 'MNISTC': MNISTC, } num_class = { # 'MNIST': 10, # 'MNISTM': 10, 'SVHN': 10, # 'SYN': 10, # 'USPS': 10, # 'MNISTC': 10, } # Get the datasets self.train_dataset, self.test_dataset = self.get_dataset(loader_map[dataset_ident], file_path, download, train_transform, test_transform, target_transform) # Set the loaders self.train_loader = DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, **kwargs) self.test_loader = DataLoader(self.test_dataset, batch_size=batch_size, shuffle=False, **kwargs) tmp_batch = self.train_loader.__iter__().__next__()[0] self.img_shape = list(tmp_batch.size())[1:] self.num_class = num_class[dataset_ident] @staticmethod def get_dataset(dataset, file_path, download, train_transform, test_transform, target_transform): # Training and Validation datasets train_dataset = dataset(file_path, train=True, download=download, transform=train_transform, target_transform=target_transform) test_dataset = dataset(file_path, train=False, download=download, transform=test_transform, target_transform=target_transform) return train_dataset, test_dataset # loader = Loader('SVHN') # dataset_train = loader.train_dataset # img = dataset_train[50][0] # print(dataset_train[50][1]) # img = img * 255 # img = Image.fromarray(np.array(img.permute(1, 2, 0)).astype('uint8'), mode='RGB') # img.show() # dataset_training = torch.load('MNISTM/processed/training.pt') # img = dataset_training[0][0] # print(img.size()) # img = Image.fromarray(np.array(img).astype('uint8'), mode='RGB') # img.show()
5,699
29.15873
215
py
GalaxyDataset
GalaxyDataset-master/digitfive/syn.py
import scipy.io as scio import numpy as np from PIL import Image import os import os.path import torch import torchvision from torchvision import datasets, transforms from torchvision.datasets import MNIST, utils from torch.utils.data import DataLoader, Dataset # dataFile = 'syn_number.mat' # data = scio.loadmat(dataFile) # for k in data.keys(): # print(k) # __header__ # __version__ # __globals__ # test_data # test_label # train_data # train_label # # data['train_data'] = data['train_data'].transpose(0, 3, 1, 2) # a = data['train_data'] # print(len(a)) # 25000 # print(len(a[0])) # 32 # print(len(a[0][0])) # 32 # print(len(a[0][0][0])) # 3 # training_data = [] # for img in data['train_data']: # img = torch.from_numpy(img).int() # training_data.append(img) # training_targets = [] # for label in data['train_label']: # l = label[0] # training_targets.append(l) # torch.save((training_data, training_targets), 'SYN/processed/training.pt') # a = data['test_data'] # print(len(a)) # 9000 # print(len(a[0])) # 3 # print(len(a[0][0])) # 32 # print(len(a[0][0][0])) # 32 # test_data = [] # for img in data['test_data']: # img = torch.from_numpy(img).int() # test_data.append(img) # test_targets = [] # for label in data['test_label']: # l = label[0] # test_targets.append(l) # torch.save((test_data, test_targets), 'SYN/processed/test.pt') class SYN(MNIST): def __init__(self, *args, **kwargs): super(SYN, self).__init__(*args, **kwargs) def __getitem__(self, index): # print('type: ',type(self.data)) # print('len: ',len(self.data)) img, target = self.data[index], int(self.targets[index]) # print('type of img: ', type(torch.Tensor(img))) # print('img size', torch.Tensor(img).size()) # return a PIL Image img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # mode & permute # print('img: ', img) # print('img size', img.size()) if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) # print(img.size()) return img, target def digit_five_train_transforms(): all_transforms = transforms.Compose([ # transforms.RandomResizedCrop(28), # transforms.RandomApply([transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8), # transforms.RandomGrayscale(p=0.2), # transforms.RandomAffine(degrees=15, # translate=[0.1, 0.1], # scale=[0.9, 1.1], # shear=15), transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms def digit_five_test_transforms(): all_transforms = transforms.Compose([ transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms class Loader(object): def __init__(self, dataset_ident, file_path='', download=False, batch_size=128, train_transform=digit_five_train_transforms(), test_transform=digit_five_test_transforms(), target_transform=None, use_cuda=False): kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {} loader_map = { # 'MNIST': MNIST, # 'MNISTM': MNISTM, # 'SVHN': SVHN, 'SYN': SYN, # 'USPS': USPS, # 'MNISTC': MNISTC, } num_class = { # 'MNIST': 10, # 'MNISTM': 10, # 'SVHN': 10, 'SYN': 10, # 'USPS': 10, # 'MNISTC': 10, } # Get the datasets self.train_dataset, self.test_dataset = self.get_dataset(loader_map[dataset_ident], file_path, download, train_transform, test_transform, target_transform) # Set the loaders self.train_loader = DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, **kwargs) self.test_loader = DataLoader(self.test_dataset, batch_size=batch_size, shuffle=False, **kwargs) tmp_batch = self.train_loader.__iter__().__next__()[0] self.img_shape = list(tmp_batch.size())[1:] self.num_class = num_class[dataset_ident] @staticmethod def get_dataset(dataset, file_path, download, train_transform, test_transform, target_transform): # Training and Validation datasets train_dataset = dataset(file_path, train=True, download=download, transform=train_transform, target_transform=target_transform) test_dataset = dataset(file_path, train=False, download=download, transform=test_transform, target_transform=target_transform) return train_dataset, test_dataset # loader = Loader('SYN') # dataset_train = loader.train_dataset # img = dataset_train[40][0] # print(dataset_train[40][1]) # img = img * 255 # img = Image.fromarray(np.array(img.permute(1, 2, 0)).astype('uint8'), mode='RGB') # img.show() # dataset_training = torch.load('MNISTM/processed/training.pt') # img = dataset_training[0][0] # print(img.size()) # img = Image.fromarray(np.array(img).astype('uint8'), mode='RGB') # img.show()
5,471
30.630058
215
py
GalaxyDataset
GalaxyDataset-master/digitfive/mnistm.py
import scipy.io as scio import numpy as np from PIL import Image import os import os.path import torch import torchvision from torchvision import datasets, transforms from torchvision.datasets import MNIST, utils from torch.utils.data import DataLoader, Dataset # dataFile = 'mnistm_with_label.mat' # data = scio.loadmat(dataFile) # for k in data.keys(): # print(k) # __header__ # __version__ # __globals__ # label_test # label_train # test # train # img = data['train'][4] # # img = Image.fromarray(img, mode='RGB') # # img.show() # img = torch.from_numpy(img) # print(img) # label = np.argmax(data['label_train'][4]) # print(label) # training_data = [] # for img in data['train']: # img = torch.from_numpy(img).int() # # print(img) # training_data.append(img) # img = training_data[0] # print(type(img)) # img = img.numpy().astype('uint8') # img = Image.fromarray(img, mode='RGB') # img.show() # training_targets = [] # for label in data['label_train']: # l = np.argmax(np.array(label)) # training_targets.append(l) # print(training_targets[0]) # torch.save((training_data, training_targets), 'MNISTM/processed/training.pt') # test_data = [] # for img in data['test']: # img = torch.from_numpy(img).int() # test_data.append(img) # test_targets = [] # for label in data['label_test']: # l = np.argmax(np.array(label)) # test_targets.append(l) # torch.save((test_data, test_targets), 'MNISTM/processed/test.pt') class MNISTM(MNIST): def __init__(self, *args, **kwargs): super(MNISTM, self).__init__(*args, **kwargs) def __getitem__(self, index): # print('type: ',type(self.data)) # print('len: ',len(self.data)) img, target = self.data[index], int(self.targets[index]) # print('type of img: ', type(torch.Tensor(img))) # print('img size', torch.Tensor(img).size()) # return a PIL Image img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # mode & permute # print('img: ', img) # print('img size', img.size()) if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) # print(img.size()) return img, target def digit_five_train_transforms(): all_transforms = transforms.Compose([ # transforms.RandomResizedCrop(28), # transforms.RandomApply([transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8), # transforms.RandomGrayscale(p=0.2), # transforms.RandomAffine(degrees=15, # translate=[0.1, 0.1], # scale=[0.9, 1.1], # shear=15), transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms def digit_five_test_transforms(): all_transforms = transforms.Compose([ transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms class Loader(object): def __init__(self, dataset_ident, file_path='', download=False, batch_size=128, train_transform=digit_five_train_transforms(), test_transform=digit_five_test_transforms(), target_transform=None, use_cuda=False): kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {} loader_map = { # 'MNIST': MNIST, 'MNISTM': MNISTM, # 'SVHN': SVHN, # 'SYN': SYN, # 'USPS': USPS, # 'MNISTC': MNISTC, } num_class = { # 'MNIST': 10, 'MNISTM': 10, # 'SVHN': 10, # 'SYN': 10, # 'USPS': 10, # 'MNISTC': 10, } # Get the datasets self.train_dataset, self.test_dataset = self.get_dataset(loader_map[dataset_ident], file_path, download, train_transform, test_transform, target_transform) # Set the loaders self.train_loader = DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, **kwargs) self.test_loader = DataLoader(self.test_dataset, batch_size=batch_size, shuffle=False, **kwargs) tmp_batch = self.train_loader.__iter__().__next__()[0] self.img_shape = list(tmp_batch.size())[1:] self.num_class = num_class[dataset_ident] @staticmethod def get_dataset(dataset, file_path, download, train_transform, test_transform, target_transform): # Training and Validation datasets train_dataset = dataset(file_path, train=True, download=download, transform=train_transform, target_transform=target_transform) test_dataset = dataset(file_path, train=False, download=download, transform=test_transform, target_transform=target_transform) return train_dataset, test_dataset # loader = Loader('MNISTM') # dataset_train = loader.train_dataset # img = dataset_train[40][0] # print(dataset_train[40][1]) # img = img * 255 # img = Image.fromarray(np.array(img.permute(1, 2, 0)).astype('uint8'), mode='RGB') # img.show() # dataset_training = torch.load('MNISTM/processed/training.pt') # img = dataset_training[0][0] # print(img.size()) # img = Image.fromarray(np.array(img).astype('uint8'), mode='RGB') # img.show()
5,512
29.97191
215
py
GalaxyDataset
GalaxyDataset-master/digitfive/SubPolicy.py
class SubPolicy(object): def __init__(self, p1, operation1, magnitude_idx1, p2, operation2, magnitude_idx2, fillcolor=(128, 128, 128)): ranges = { "shearX": np.linspace(0, 0.3, 10), "shearY": np.linspace(0, 0.3, 10), "translateX": np.linspace(0, 150 / 331, 10), "translateY": np.linspace(0, 150 / 331, 10), "rotate": np.linspace(0, 30, 10), "color": np.linspace(0.0, 0.9, 10), "posterize": np.round(np.linspace(8, 4, 10), 0).astype(np.int), "solarize": np.linspace(256, 0, 10), "contrast": np.linspace(0.0, 0.9, 10), "sharpness": np.linspace(0.0, 0.9, 10), "brightness": np.linspace(0.0, 0.9, 10), "autocontrast": [0] * 10, "equalize": [0] * 10, "invert": [0] * 10 } # from https://stackoverflow.com/questions/5252170/specify-image-filling-color-when-rotating-in-python-with-pil-and-setting-expand def rotate_with_fill(img, magnitude): rot = img.convert("RGBA").rotate(magnitude) return Image.composite(rot, Image.new("RGBA", rot.size, (128,) * 4), rot).convert(img.mode) func = { "shearX": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, magnitude * random.choice([-1, 1]), 0, 0, 1, 0), Image.BICUBIC, fillcolor=fillcolor), "shearY": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, 0, magnitude * random.choice([-1, 1]), 1, 0), Image.BICUBIC, fillcolor=fillcolor), "translateX": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, magnitude * img.size[0] * random.choice([-1, 1]), 0, 1, 0), fillcolor=fillcolor), "translateY": lambda img, magnitude: img.transform( img.size, Image.AFFINE, (1, 0, 0, 0, 1, magnitude * img.size[1] * random.choice([-1, 1])), fillcolor=fillcolor),
2,048
54.378378
138
py
GalaxyDataset
GalaxyDataset-master/digitfive/mnist.py
import scipy.io as scio import numpy as np from PIL import Image import os import os.path import torch import torchvision from torchvision import datasets, transforms from torchvision.datasets import MNIST, utils from torch.utils.data import DataLoader, Dataset # dataFile = 'mnist_data.mat' # data = scio.loadmat(dataFile) # for k in data.keys(): # print(k) # __header__ # __version__ # __globals__ # test_32 # test_28 # label_test # label_train # train_32 # train_28 # img = data['train_28'][0] # img = torch.from_numpy(img) # print(img) # torch.Size([28, 28, 1]) # training_data = [] # for img in data['train_28']: # img = img.transpose(2, 0, 1) # img = img.tolist() # temp = img[0] # img.append(temp) # img.append(temp) # img = torch.Tensor(img) # img = img.permute(1, 2, 0) # training_data.append(img) # img = training_data[0] # # img = Image.fromarray(img.numpy(), mode='RGB') # black # img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # astype('uint8') # img.show() # training_targets = [] # for label in data['label_train']: # l = np.argmax(np.array(label)) # training_targets.append(l) # torch.save((training_data, training_targets), 'MNIST/processed/training.pt') # test_data = [] # for img in data['test_28']: # img = img.transpose(2, 0, 1) # img = img.tolist() # temp = img[0] # img.append(temp) # img.append(temp) # img = torch.Tensor(img) # img = img.permute(1, 2, 0) # test_data.append(img) # test_targets = [] # for label in data['label_test']: # l = np.argmax(np.array(label)) # test_targets.append(l) # torch.save((test_data, test_targets), 'MNIST/processed/test.pt') # img = test_data[0] # # img = Image.fromarray(img.numpy(), mode='RGB') # black # img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # astype('uint8') # img.show() class MNIST(MNIST): def __init__(self, *args, **kwargs): super(MNIST, self).__init__(*args, **kwargs) def __getitem__(self, index): # print('type: ',type(self.data)) # print('len: ',len(self.data)) img, target = self.data[index], int(self.targets[index]) # print('type of img: ', type(torch.Tensor(img))) # print('img size', torch.Tensor(img).size()) # return a PIL Image img = Image.fromarray(img.numpy().astype('uint8'), mode='RGB') # mode & permute # print('img: ', img) # print('img size', img.size()) if self.transform is not None: img = self.transform(img) if self.target_transform is not None: target = self.target_transform(target) # print(img.size()) return img, target def digit_five_train_transforms(): all_transforms = transforms.Compose([ # transforms.RandomResizedCrop(28), # ToPILImage # transforms.RandomApply([transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8), # transforms.RandomGrayscale(p=0.2), # transforms.RandomAffine(degrees=15, # translate=[0.1, 0.1], # scale=[0.9, 1.1], # shear=15), transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms def digit_five_test_transforms(): all_transforms = transforms.Compose([ transforms.ToTensor(), # transforms.Normalize([0.4914, 0.4822, 0.4465], [0.2023, 0.1994, 0.2010]) ]) return all_transforms class Loader(object): def __init__(self, dataset_ident, file_path='', download=False, batch_size=128, train_transform=digit_five_train_transforms(), test_transform=digit_five_test_transforms(), target_transform=None, use_cuda=False): kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {} loader_map = { 'MNIST': MNIST, # 'MNISTM': MNISTM, # 'SVHN': SVHN, # 'SYN': SYN, # 'USPS': USPS, # 'MNISTC': MNISTC, } num_class = { 'MNIST': 10, # 'MNISTM': 10, # 'SVHN': 10, # 'SYN': 10, # 'USPS': 10, # 'MNISTC': 10, } # Get the datasets self.train_dataset, self.test_dataset = self.get_dataset(loader_map[dataset_ident], file_path, download, train_transform, test_transform, target_transform) # Set the loaders self.train_loader = DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, **kwargs) self.test_loader = DataLoader(self.test_dataset, batch_size=batch_size, shuffle=False, **kwargs) tmp_batch = self.train_loader.__iter__().__next__()[0] self.img_shape = list(tmp_batch.size())[1:] self.num_class = num_class[dataset_ident] @staticmethod def get_dataset(dataset, file_path, download, train_transform, test_transform, target_transform): # Training and Validation datasets train_dataset = dataset(file_path, train=True, download=download, transform=train_transform, target_transform=target_transform) test_dataset = dataset(file_path, train=False, download=download, transform=test_transform, target_transform=target_transform) return train_dataset, test_dataset # loader = Loader('MNIST') # dataset_train = loader.train_dataset # img = dataset_train[5][0] # print(dataset_train[5][1]) # print(img) # img = img * 255 # img = Image.fromarray(np.array(img.permute(1, 2, 0)).astype('uint8'), mode='RGB') # img.show()
5,753
30.966667
215
py
skimulator
skimulator-master/setup.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """Build and install the SKIM Simulator for Ocean Science package.""" import os import sys import errno import shutil import logging from setuptools import setup logger = logging.getLogger() handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) logger.addHandler(handler) logger.setLevel(logging.INFO) # Check Python version if not 3 == sys.version_info[0]: logger.error('This package is only available for Python 3.x') sys.exit(1) __package_name__ = 'skimulator' project_dir = os.path.dirname(__file__) git_exe = shutil.which('git') git_dir = os.path.join(project_dir, '.git') has_git = (git_exe is not None and os.path.isdir(git_dir)) readme_file = os.path.join(project_dir, 'README') package_dir = os.path.join(project_dir, __package_name__) init_file = os.path.join(package_dir, '__init__.py') share_dir = os.path.join(package_dir, 'share') version_file = os.path.join(share_dir, 'VERSION.txt') # Regenerate a version file from git history if has_git: import subprocess import datetime _githash = (git_exe, 'rev-parse', '--short', 'HEAD') githash = subprocess.check_output(_githash).decode('utf-8').strip() gitrev = (git_exe, 'rev-list', 'HEAD', '--count') commits = subprocess.check_output(gitrev).decode('utf-8').strip() # Note: this block could be replaced by # os.makedirs(share_dir, exist_ok=True) # if support for Python < 3.5 is dropped if not os.path.isdir(share_dir): try: os.makedirs(share_dir) except OSError: _, e, _ = sys.exc_info() if e.errno == errno.EEXIST: pass with open(version_file, 'wt') as f: f.write('{}\n'.format(commits)) f.write('{}\n'.format(githash)) f.write('{}\n'.format(datetime.datetime.utcnow().strftime('%Y-%m-%d'))) # - Read in the package version and author fields from the Python # main __init__.py file: # # IMPORTANT: this must be done AFTER trying to generate the VERSION.txt file # as the __version__ variable contained in the __init__.py will not only have # the major and minor, but also the commit number if VERSION.txt exists. metadata = {} with open(init_file, 'rt') as f: exec(f.read(), metadata) requirements = [] with open('requirements.txt', 'r') as f: lines = [x.strip() for x in f if 0 < len(x.strip())] requirements = [x for x in lines if x[0].isalpha()] with open(readme_file, 'rt') as f: long_description = f.read() optional_dependencies = {'plot': ['matplotlib', ], 'carto': ['matplotlib', 'cartopy'], 'numba': ['numba', ],} cmds = ['skimul2b = {}.cli:run_script'.format(__package_name__), 'skimul2c = {}.cli:run_l2c'.format(__package_name__), 'skimul2d = {}.cli:run_l2d'.format(__package_name__), 'interpl2d = {}.cli:interpolate_l2d'.format(__package_name__), ] # VERSION.txt must be added to the package if the file has been generated pkg_data = None pkg_data = {__package_name__: ['share/coeff.npy', 'share/coeffr.npy', 'share/Spline_128_64_TED_TAS_6_DEG.npy', 'share/Spline_128_64_TED_TAS_12_DEG.npy', 'share/Spline_128_64_TED_CB_12_DEG.npy', 'share/Spline_128_64_TED_CB_6_DEG.npy', 'share/noise_pdf_mss1d.npy', ]} if os.path.exists(version_file): pkg_data[__package_name__].append('share/VERSION.txt') setup(name=__package_name__, version=metadata['__version__'], description=metadata['__description__'], author=metadata['__author__'], author_email=metadata['__author_email__'], url=metadata['__url__'], license='COPYING', keywords=metadata['__keywords__'], long_description=long_description, packages=(__package_name__, f'{__package_name__}.error'), install_requires=requirements, setup_require=(), entry_points={'console_scripts': cmds}, extras_require=optional_dependencies, package_data=pkg_data, )
4,810
36.007692
79
py
skimulator
skimulator-master/test/projskim.py
import os import numpy import healpy as heal from numpy import linalg as LA from netCDF4 import Dataset import params as p import skimulator.const as const import skimulator.rw_data as rw import matplotlib.pyplot as plt theta1 = const.theta1 theta0 = const.theta0 gamma0 = const.gamma0 # - In parameter file ## TODO - # Number of pixel (resolution for healpix) nside = 256 # Number of diamonds for healpix ndiam = 12 ntotpixel = nside * nside * ndiam # Conditionning threshold thresh_cond = 10 # List of pass swathtab = [40] #,'p025','p042','p053','p055'] # cycle = 1 plot_res = 10 gamma = numpy.array(p.list_angle) ind_12 = numpy.where(gamma == 12) ind_6 = numpy.where(gamma == 6) gamma = numpy.deg2rad(gamma) n_ind_12 = len(ind_12) n_ind_6 = len(ind_6) path = os.path.join(p.outdatadir, '{}_c{:02d}'.format(p.config, cycle)) gpath = os.path.join(p.outdatadir, '{}_grid'.format(p.config)) #heal.mollview(np.arange(12*64*64)) #plt.show() # rebin variance to beam sig1 = 1E-2*numpy.loadtxt(p.rms_instr[ind_12[0][0]], usecols=(0, 1), unpack=True) sig2 = 1E-2*numpy.loadtxt(p.rms_instr[ind_6[0][0]], usecols=(0, 1), unpack=True) sig = numpy.zeros([len(gamma), 360]) for i in ind_12[0]: sig[i, : 180] = sig2[1, : 180] sig[i, 180 + numpy.arange(180)] = sig2[1, 180 - numpy.arange(180)] for i in ind_6[0]: sig[i, 0: 180] = sig1[1, 0: 180] sig[i, 180 + numpy.arange(180)] = sig1[1, 180 - numpy.arange(180)] #===== table to store image inversion im = numpy.zeros((ntotpixel, 3)) nim = numpy.zeros((ntotpixel)) cov = numpy.zeros((ntotpixel, 2, 2)) cov2 = numpy.zeros([ntotpixel, 2, 2]) vec = numpy.zeros([ntotpixel, 2]) vec2 = numpy.zeros([ntotpixel, 2]) vec3 = numpy.zeros([ntotpixel, 2]) vecdop = numpy.zeros([3, ntotpixel, 2]) for swath in swathtab: data = rw.Sat_SKIM(ifile='{}_p{:03d}.nc'.format(path, swath)) grid = rw.Sat_SKIM(ifile='{}_p{:03d}.nc'.format(gpath, swath)) data.load_data(p, ur_model=[], ur_obs=[], instr=[], u_model=[], v_model=[], lon_nadir=[], lat_nadir=[], time_nadir=[], lon=[], lat=[]) grid.load_swath(p, radial_angle=[]) ur = data.ur_model uro = data.ur_obs noise = data.instr u = data.u_model v = data.v_model rangle = numpy.mod(grid.radial_angle, numpy.pi *2) lon = data.lon lat = data.lat ndata, nbeam = ur.shape # COMPUTE SIGMA AT EACH SAMPLE sigma = numpy.zeros(ur.shape) arctan_rangle = numpy.arctan(numpy.sin(rangle + numpy.pi/2), numpy.cos(rangle + numpy.pi/2)) _tmp = numpy.rad2deg(numpy.mod(arctan_rangle + 2 * numpy.pi, 2 * numpy.pi)) for i in range(len(gamma)): sigma[:, i] = sig[i, (_tmp[:, i]).astype(int)] #COMPUTE ORBITAL SPEED FROM NADIR INFORMATION # TODO MANAGE BUG OF -180 to +180 TODOTODOTODO lat_nadir = data.lat_nadir lon_nadir = data.lon_nadir time_nadir = data.time_nadir time_nadir = time_nadir - time_nadir[0] latpar = numpy.polyfit(time_nadir, lat_nadir, 3) lonpar = numpy.polyfit(time_nadir, lon_nadir, 3) dlatorb = latpar[2] + 2*time_nadir*latpar[1] + 3*time_nadir**2*latpar[0] dlonorb = lonpar[2] + 2*time_nadir*lonpar[1] + 3*time_nadir**2*lonpar[0] #altiorb=793.0 alti_total = const.Rearth + const.sat_elev usat = (numpy.deg2rad(dlonorb) * numpy.cos(numpy.deg2rad(lat_nadir)) * alti_total) vsat = numpy.deg2rad(dlatorb) * alti_total dtheta = theta1 - gamma0 * numpy.sin(rangle - theta0) dgamma = gamma0 * numpy.cos(rangle - theta0) # nongeophysical doppler to add to measure gamma_mat = numpy.array([gamma,] * ndata) ddop = numpy.zeros(ur.shape) usat = numpy.array([usat, ]*len(gamma)).transpose() vsat = numpy.array([vsat, ]*len(gamma)).transpose() ddop1 = dtheta * (usat * numpy.sin(rangle) + vsat * numpy.cos(rangle)) ddop2 = dgamma * numpy.cos(gamma_mat) * (usat * numpy.cos(rangle) + vsat * numpy.sin(rangle)) ddop = ddop1 + ddop2 tdop=numpy.zeros((ndata, nbeam, 3)) tdop[:, :, 0] = usat * numpy.sin(rangle) + vsat * numpy.cos(rangle) tdop[:, :, 1] = (-numpy.sin(rangle) * (usat*numpy.sin(rangle) + vsat * numpy.cos(rangle)) + numpy.cos(rangle) * numpy.cos(gamma_mat) * (usat*numpy.cos(rangle) + vsat * numpy.sin(rangle))) tdop[:, :, 2] = (numpy.cos(rangle) * (usat * numpy.sin(rangle) + vsat * numpy.cos(rangle)) + numpy.sin(rangle) * numpy.cos(gamma_mat) * (usat * numpy.cos(rangle) + vsat * numpy.sin(rangle))) # DEBUG PLOT #plt.plot(ddop-(tdop[:,:,0]*theta1+tdop[:,:,1]*gamma0*np.cos(theta0)+tdop[:,:,2]*gamma0*np.sin(theta0))) #plt.show() uro = ddop + uro ww = 1 / sigma**2 ph = 2 * numpy.pi - numpy.deg2rad(lon) th = numpy.pi / 2 - numpy.deg2rad(lat) pidx = heal.ang2pix(nside, th, ph) for i in range(nbeam): for j in range(ndata): if ur[j, i] > -1E9: ip = pidx[j,i] # compute imulated model im[ip, 1] += u[j, i] im[ip, 2] += v[j,i] nim[ip] += 1 # compute covariance(s) model co = numpy.cos(rangle[j,i]) si = numpy.sin(rangle[j,i]) w = ww[j,i] cov[ip, 0, 0] += co * co cov[ip, 1, 0] += si * co cov[ip, 0, 1] += si * co cov[ip, 1, 1] += si * si cov2[ip, 0, 0] += w * co * co cov2[ip, 1, 0] += w * si * co cov2[ip, 0, 1] += w * si * co cov2[ip, 1, 1] += w * si * si # compute data vector model vec[ip, 0] += co * ur[j,i] vec[ip, 1] += si * ur[j,i] # compute data noise vector model vec2[ip, 0] += w* co * uro[j,i] vec2[ip, 1] += w * si * uro[j,i] # compute doppler projection for k in range(3): vecdop[k, ip, 0] += w * co * tdop[j,i,k] vecdop[k, ip, 1] += w * si * tdop[j,i,k] rim = 0 * im rim2 = 0 * im mask = numpy.zeros([ntotpixel]) rimdop = numpy.zeros([3, ntotpixel, 2]) for i in range(ntotpixel): if cov2[i, 0, 0] > 0: mask[i] = LA.cond(cov2[i, :, :]) if mask[i] < thresh_cond: rim[i, 1: 3] = LA.solve(cov[i, :, :],vec[i, :]) rim2[i, 1: 3] = LA.solve(cov2[i, :, :],vec2[i, :]) for k in range(3): rimdop[k, i, :] = LA.solve(cov2[i, :, :], vecdop[k, i, :]) im[i, 0] = numpy.sqrt((im[i, 1]/nim[i])**2 + (im[i, 2]/nim[i])**2) rim[i, 0] = numpy.sqrt(rim[i, 1]**2 + rim[i, 2]**2) rim2[i, 0] = numpy.sqrt(rim2[i, 1]**2 + rim2[i, 2]**2) mat=numpy.zeros([3,3]) vres=numpy.zeros([3]) for i in range(nbeam): for j in range(ndata): ip = pidx[j,i] if mask[ip] < thresh_cond and ur[j, i] > -1E9: co = numpy.cos(rangle[j, i]) si = numpy.sin(rangle[j, i]) w = ww[j, i] vv = uro[j, i] - rim2[ip, 1] * co - rim2[ip, 2] * si tvv = numpy.zeros((3,)) for k in range(3): tvv[k] = (tdop[j, i, k] - rimdop[k, ip, 0] * co - rimdop[k, ip, 1] * si) for k in range(3): for l in range(3): mat[k, l] += w * tvv[k] * tvv[l] for k in range(3): vres[k] += w * tvv[k] * vv rr = LA.solve(mat, vres) print(rr) #[yo, peach, roll] = rr #print(rr[0], numpy.arctan(rr[1], rr[2]), numpy.sqrt(rr[1]**2 + rr[2]**2)) rot=[numpy.rad2deg(numpy.mean(ph)), numpy.rad2deg(numpy.pi/2 - numpy.mean(th))] residu = (rr simulé - rr) . tdop[i, j, :] u_residu_alt = cos(azimuth) * residu v_residu_alt = sin(azimuth) * residu ur. heal.gnomview(mask,rot=rot,reso=plot_res * 3,title='CONDITIONING',max=thresh_cond) #plt.pcolormesh() plt.savefig('{}_conditioning.png'.format(p.config)) heal.gnomview(im[:,0],rot=rot, reso=plot_res, title='MODEL',xsize=600) plt.savefig('{}_model.png'.format(p.config)) heal.gnomview(rim[:,0],rot=rot,reso=plot_res,title='INVERSE',xsize=600) plt.savefig('{}_inverse.png'.format(p.config)) heal.gnomview(rim2[:,0],rot=rot,reso=plot_res,title='INVERSE WITH NOISE+DOPPLER',xsize=600) plt.savefig('{}_inverse_noise.png'.format(p.config))
8,525
32.046512
108
py
skimulator
skimulator-master/test/params2.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") + '/src/' # ------ Directory that contains orbit file: dir_setup = os.path.join(home, 'skimulator', 'data') # ------ Directory that contains your own inputs: indatadir = os.path.join(home, 'skimulator', 'example', 'input_fields') indatadir = '/mnt/data/model/ww3_fram/' #netcdf3/' # ------ Directory that contains your outputs: outdatadir = os.path.join(home, 'skimulator', 'example', 'skim_output') outdatadir = os.path.join('/mnt/data/project/skim/', 'skim_output') # ------ Orbit file: #filesat = os.path.join(dir_setup,'orbs1a.txt') filesat = os.path.join(dir_setup,'orbits1_ifremer') # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Name of the configuration (to build output files names) config="WW3_FRAM_8b105az" proc_number = 4 # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) # ------ Force the computation of the satellite grid: makesgrid = True # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = [335,23,73,85.00024] #------- Rotation speed of the antenna (in tr/min) rotation_speed = 5.20833 #------- List of position of beams: list_pos = (0, 72*math.pi/180., 144*math.pi/180., 216*math.pi / 180., 288*math.pi/180., 0, math.pi) #------- List of angle of beams in degrees: list_angle = (12, 12, 12, 12, 12, 6, 6) #------- List of timeshift as regard to nadir for 12 degree beams: list_shift = (1, 2, 4, 5, 7, 3, 6) #------- Cycle duration cycle = 0.0096 # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, 'list_of_file.txt') # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other options are ROMS, NEMO and CLS to read Nemo, roms or CLS) model = 'WW3' # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify velocities variable: varu = 'ucur' varv = 'vcur' # ------ Specify factor to convert velocity values in m/s: vel_factor = 1. # ------ Specify longitude variable: lonu = 'longitude' lonv = 'longitude' # ------ Specify latitude variable: latu = 'latitude' latv = 'latitude' # ------ Specify number of time in file: dim_time = (720,) # ------ Time step between two model outputs (in days): timestep = 1/24. # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 719. # ------ Not a number value: model_nan = -32767. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'linear' # -----------------------# # SKIM error parameters # -----------------------# # ------ File containing random coefficients to compute and save # random error coefficients so that runs are reproducible: # If file_coeff is specified and does not exist, file is created # If you don't want runs to be reproducible, file_coeff is set to None file_coeff = None # outdatadir+os.sep+'Random_coeff.nc' # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Instrument white noise error instr = True # ------- Instrument white noise rms # rms_instr = [10 * 10**(-2), 10 * 10**(-2), 10 * 10**(-2), 10 * 10**(-2), # 20 * 10 ** (-2)] rms_instr = [os.path.join(dir_setup, 'instrumentnoise_12.dat'), os.path.join(dir_setup, 'instrumentnoise_12.dat'), os.path.join(dir_setup, 'instrumentnoise_12.dat'), os.path.join(dir_setup, 'instrumentnoise_12.dat'), os.path.join(dir_setup, 'instrumentnoise_12.dat'), os.path.join(dir_setup, 'instrumentnoise_06.dat'), os.path.join(dir_setup, 'instrumentnoise_06.dat')] # Multplication factor for instrument noise data to compensate for faster time # cycle rms_instr_factor = 2.5 # ------- Stoke drift velocity [beam 12, beam 6] uss = True input_uss = os.path.join(indatadir, 'list_file_uss.txt') G = [50, 50, 50, 50, 50, 50, 50] bias_std = 0.09 errdcos = None #[25.2006/20, 25.2747/20, 25.4763/20, 25.4271/20, 19.9728/20] footprint_std = 0 #400 formula = False ## -- Geophysical error ## ---------------------- # ------ Wet tropo error (True to compute it): wet_tropo = True # ------ Beam print size (in km): # Gaussian footprint of sigma km sigma = 8.
5,776
37.771812
79
py
skimulator
skimulator-master/test/mod_diag.py
import numpy import netCDF4 import os import sys import glob import pickle import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import cartopy import skimulator.rw_data as rw # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() def diag_rms(listfile, modelbox, output, list_angle): nanmean12 = 0 nanmean06 = 0 nanmeaninstr12 = 0 nanmeaninstr06 = 0 nanmeanuwb12 = 0 nanmeanuwb06 = 0 nanmeanuwbc12 = 0 nanmeanuwbc06 = 0 nanmeandsig12 = 0 nanmeandsig06 = 0 nanmeangsig12 = 0 nanmeangsig06 = 0 nanstd12 = 0 nanstd06 = 0 nanstd12norm = 0 nanstd06norm = 0 i06 = 0 i12 = 0 for ifile in listfile: print(ifile) data = netCDF4.Dataset(ifile, 'r') ind0 = 100 ind1 = -100 if 'sigma0' in data.variables.keys(): sig = data.variables['sigma0'][ind0:ind1, :] uwnd = data.variables['uwnd'][ind0:ind1, :] vwnd = data.variables['vwnd'][ind0:ind1, :] wnd = numpy.sqrt(uwnd**2 + vwnd**2) lon = data.variables['lon'][ind0:ind1, :] lat = data.variables['lat'][ind0:ind1, :] lon = numpy.mod(lon + 180.0, 360.0) - 180.0 lon_nadir = data.variables['lon_nadir'][ind0:ind1] lon_nadir = numpy.mod(lon_nadir + 180.0, 360) - 180.0 lat_nadir = data.variables['lat_nadir'][ind0:ind1] vartrue = data.variables['ur_true'][ind0:ind1, :] varobs = data.variables['ur_obs'][ind0:ind1, :] if len(lon_nadir) < 700: print('track too short') continue instr = False if 'instr' in data.variables.keys(): instr = True varinstr = data.variables['instr'][ind0:ind1, :] varinstr[numpy.where(abs(varinstr) > 1)] = numpy.nan uwb = False if 'uwd' in data.variables.keys(): uwb = True varuwb = data.variables['uwd'][ind0:ind1, :] uwbc = False if 'uwd_est' in data.variables.keys(): uwbc = True varuwbc = data.variables['uwd_est'][ind0:ind1, :] varuwbc = varuwbc - varuwb dsigma = False if 'dsigma' in data.variables.keys(): dsigma = True vardsig = data.variables['dsigma'][ind0:ind1, :] dsigma = False gsig_atm = False if 'gsig_atm_err' in data.variables.keys(): gsig_atm = False vargsig = data.variables['gsig_atm_err'][ind0:ind1, :] mask = ((abs(vartrue)>100) | (abs(varobs)>100) | numpy.isnan(varinstr)) vartrue[mask] = numpy.nan varobs[mask] = numpy.nan beam_angle = list_angle #data['beam_angle'][:] data.close() # Remove borders if numpy.shape(lon)[0]<100: continue for i in range(numpy.shape(lon)[1]): if varobs[:, i].mask.all(): continue #if varuwbc[:, i].mask.all(): # continue _ind = numpy.where((wnd[:, i]>3) & (wnd[:, i]<20)) #_ind = numpy.where((sig[:, i]>10**-5) & (wnd[:, i]>3)) try: _tmp = numpy.nanmean(((varobs[_ind, i] - vartrue[_ind,i])/vartrue[_ind,i])**2) _tmp = numpy.nanmean(((varobs[_ind, i] - vartrue[_ind,i]))**2) _tmp2 = numpy.nanmean(abs(vartrue[_ind, i])**2) except: continue if instr is True: _tmpinstr = numpy.nanmean(abs(varinstr[_ind, i])**2) if uwb is True: try: _tmpuwb = numpy.nanmean(abs(varuwb[_ind, i])**2) except: continue if uwbc is True: _tmpuwbc = numpy.nanmean(abs(varuwbc[_ind, i])**2) if dsigma is True: _tmpdsig = numpy.nanmean(abs(vardsig[_ind, i])**2) if gsig_atm is True: _tmpgsig = numpy.nanmean(abs(vargsig[_ind, i])**2) if beam_angle[i] == 6: nanstd06norm += _tmp2 nanstd06 += _tmp if instr is True: nanmeaninstr06 += _tmpinstr if uwb is True: nanmeanuwb06 += _tmpuwb if uwbc is True: nanmeanuwbc06 += _tmpuwbc if dsigma is True: nanmeandsig06 += _tmpdsig if gsig_atm is True: nanmeangsig06 += _tmpgsig i06 += 1 else: nanstd12 += _tmp nanstd12norm += _tmp2 if instr is True: nanmeaninstr12 += _tmpinstr if uwb is True: nanmeanuwb12 += _tmpuwb if uwbc is True: nanmeanuwbc12 += _tmpuwbc if dsigma is True: nanmeandsig12 += _tmpdsig if gsig_atm is True: nanmeangsig12 += _tmpgsig i12 += 1 if i06 == 0: i06 = numpy.nan if i12 == 0: i12 = numpy.nan nanstd06 = numpy.sqrt(nanstd06/i06) nanstd12 = numpy.sqrt(nanstd12/i12) # nanstd06 = numpy.sqrt(nanstd06/i06) # nanstd12 = numpy.sqrt(nanstd12/i12) nanstd06norm = numpy.sqrt(nanstd06norm/i06) nanstd12norm = numpy.sqrt(nanstd12norm/i12) if instr is True: nanmeaninstr06 = numpy.sqrt(nanmeaninstr06/i06) nanmeaninstr12 = numpy.sqrt(nanmeaninstr12/i12) if uwb is True: nanmeanuwb06 = numpy.sqrt(nanmeanuwb06/i06) nanmeanuwb12 = numpy.sqrt(nanmeanuwb12/i12) if uwbc is True: nanmeanuwbc06 = numpy.sqrt(nanmeanuwbc06/i06) nanmeanuwbc12 = numpy.sqrt(nanmeanuwbc12/i12) if dsigma is True: nanmeandsig06 = numpy.sqrt(nanmeandsig06/i06) nanmeandsig12 = numpy.sqrt(nanmeandsig12/i12) if gsig_atm is True: nanmeangsig06 = numpy.sqrt(nanmeangsig06/i06) nanmeangsig12 = numpy.sqrt(nanmeangsig12/i12) print('RMSE 06: {}, RMSE 12: {}'.format(nanstd06, nanstd12)) print('NRMSE 06: {}, NRMSE 12: {}'.format(nanstd06norm, nanstd12norm)) if instr is True: print('instr rms 06: {}, instr rms 12: {}'.format(nanmeaninstr06, nanmeaninstr12)) if uwb is True: print('uwb rms 06: {}, uwb rms 12: {}'.format(nanmeanuwb06, nanmeanuwb12)) if uwbc is True: print('uwbc rms 06: {}, uwbc rms 12: {}'.format(nanmeanuwbc06, nanmeanuwbc12)) if dsigma is True: print('dsigma rms 06: {}, dsigma rms 12: {}'.format(nanmeandsig06, nanmeandsig12)) if gsig_atm is True: print('gsig rms 06: {}, gsig rms 12: {}'.format(nanmeangsig06, nanmeangsig12)) return nanstd06, nanstd12, nanstd06norm, nanstd12norm def diag_azimuth_rms(listfile, modelbox, output, list_angle): nanmean12 = 0 nanmean06 = 0 nanmeaninstr12 = 0 nanmeaninstr06 = 0 nanmeanuwb12 = 0 nanmeanuwb06 = 0 nanmeanuwbc12 = 0 nanmeanuwbc06 = 0 nanstd12 = 0 nanstd06 = 0 nanstd12norm = 0 nanstd06norm = 0 i06 = 0 i12 = 0 for ifile in listfile: print(ifile) data = netCDF4.Dataset(ifile, 'r') lon = data.variables['lon'][:] lat = data.variables['lat'][:] lon = numpy.mod(lon + 180.0, 360.0) - 180.0 lon_nadir = data.variables['lon_nadir'][:] lon_nadir = numpy.mod(lon_nadir + 180.0, 360) - 180.0 lat_nadir = data.variables['lat_nadir'][:] vartrue = data.variables['ur_true'][:] varobs = data.variables['ur_obs'][:] if 'instr' in data.variables.keys(): instr = True varinstr = data.variables['instr'][:] if 'uwd' in data.variables.keys(): uwb = True varuwb = data.variables['uwd'][:] if 'uwd_est' in data.variables.keys(): uwbc = True varuwbc = data.variables['uwd_est'][:] mask = ((abs(vartrue)>100) | (abs(varobs)>100)) #vartrue[mask] = numpy.nan #varobs[mask] = numpy.nan beam_angle = list_angle #data['beam_angle'][:] data.close() if numpy.shape(lon)[0]<100: continue for i in range(numpy.shape(lon)[1]): if varobs[:, i].mask.all(): continue if varuwbc[:, i].mask.all(): continue try: _tmp = numpy.nanmean(((varobs[:, i] - vartrue[:,i])/vartrue[:,i])**2) _tmp = numpy.nanmean(((varobs[:, i] - vartrue[:,i]))**2) _tmp2 = numpy.nanmean(abs(vartrue[:, i])**2) except: continue if instr is True: _tmpinstr = numpy.nanmean(abs(varinstr[:, i])**2) if uwb is True: try: _tmpuwb = numpy.nanmean(abs(varuwb[:, i])**2) except: continue if uwbc is True: _tmpuwbc = numpy.nanmean(abs(varuwbc[:, i])**2) if beam_angle[i] == 6: nanstd06norm += _tmp/_tmp2 nanstd06 += _tmp if instr is True: nanmeaninstr06 += _tmpinstr if uwb is True: nanmeanuwb06 += _tmpuwb if uwbc is True: nanmeanuwbc06 += _tmpuwbc i06 += 1 else: nanstd12 += _tmp nanstd12norm += _tmp/_tmp2 if instr is True: nanmeaninstr12 += _tmpinstr if uwb is True: nanmeanuwb12 += _tmpuwb if uwbc is True: nanmeanuwbc12 += _tmpuwbc i12 += 1 nanstd06 = numpy.sqrt(nanstd06/i06) nanstd12 = numpy.sqrt(nanstd12/i12) # nanstd06 = numpy.sqrt(nanstd06/i06) # nanstd12 = numpy.sqrt(nanstd12/i12) if instr is True: nanmeaninstr06 = numpy.sqrt(nanmeaninstr06/i06) nanmeaninstr12 = numpy.sqrt(nanmeaninstr12/i12) if uwb is True: nanmeanuwb06 = numpy.sqrt(nanmeanuwb06/i06) nanmeanuwb12 = numpy.sqrt(nanmeanuwb12/i12) if uwbc is True: nanmeanuwbc06 = numpy.sqrt(nanmeanuwbc06/i06) nanmeanuwbc12 = numpy.sqrt(nanmeanuwbc12/i12) print('RMSE 06: {}, RMSE 12: {}'.format(nanstd06, nanstd12)) print('NRMSE 06: {}, NRMSE 12: {}'.format(nanstd06norm, nanstd12norm)) if instr is True: print('instr rms 06: {}, instr rms 12: {}'.format(nanmeaninstr06, nanmeaninstr12)) if uwb is True: print('uwb rms 06: {}, uwb rms 12: {}'.format(nanmeanuwb06, nanmeanuwb12)) if uwbc is True: print('uwbc rms 06: {}, uwbc rms 12: {}'.format(nanmeanuwbc06, nanmeanuwbc12)) return nanstd06, nanstd12, nanstd06norm, nanstd12norm def bin_variables(listfile, listvar, bin_in, modelbox): lonp = numpy.arange(modelbox[0], modelbox[1] + modelbox[2], modelbox[2]) latp = numpy.arange(modelbox[3], modelbox[4] + modelbox[5], modelbox[5]) resol = numpy.sqrt((modelbox[2] * numpy.cos(numpy.deg2rad(latp))) **2 + modelbox[5]**2) dic_v = {} for ifile in listfile: data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:] lat = data['lat'][:] for j in range(len(lonp)): for i in range(len(latp)): ind_key = 10000 * int(i) + int(j) lon = numpy.mod(lon + 180, 360) - 180 dist = numpy.sqrt(((lonp[j] - lon) *numpy.cos(numpy.deg2rad(lat)))**2 + (latp[i] - lat)**2) iiobs = numpy.where(dist < resol[i]) if not iiobs: continue if ind_key not in dic_v.keys(): dic_v[ind_key] = {} for ivar in listvar: try: var = data[ivar][:] except: print(ifile, ivar) _mask = numpy.ma.getmaskarray(var) mask = _mask[iiobs] if ivar == 'ur_obs': ivar2 = 'diff_ur' var1 = data['ur_true'][:] var2 = numpy.array(abs(var.data[iiobs] - var1.data[iiobs])) if var2.any(): if ivar2 not in dic_v[ind_key].keys(): dic_v[ind_key][ivar2] = [] dic_v[ind_key][ivar2].append(var2[~mask]) ivar2 = 'cov_ur' cov = numpy.cov(var.data[iiobs], var1.data[iiobs]) var2 = (200*cov[0, 1] / (numpy.trace(cov))) if var2.any(): if ivar2 not in dic_v[ind_key].keys(): dic_v[ind_key][ivar2] = [] dic_v[ind_key][ivar2].append(var2) if ivar == 'uwd_est': ivar2 = 'diff_uwdr' var1 = data['uwd'][:] var2 = numpy.array(abs(var.data[iiobs] - var1.data[iiobs])) if var2.any(): if ivar2 not in dic_v[ind_key].keys(): dic_v[ind_key][ivar2] = [] dic_v[ind_key][ivar2].append(var2[~mask]) ivar2 = 'cov_uwdr' cov = numpy.cov(var.data[iiobs], var1.data[iiobs]) var2 = (200*cov[0, 1] / (numpy.trace(cov))) if var2.any(): if ivar2 not in dic_v[ind_key].keys(): dic_v[ind_key][ivar2] = [] dic_v[ind_key][ivar2].append(var2) var = numpy.array(var.data[iiobs]) if var.any(): if ivar not in dic_v[ind_key].keys(): dic_v[ind_key][ivar] = [] dic_v[ind_key][ivar].append(var[~mask]) data.close() with open(bin_in, 'wb') as f: pickle.dump(dic_v, f) def compute_rms(bin_in, bin_out, listvar, modelbox): with open(bin_in, 'rb') as f: dic_v = pickle.load(f) lonp = numpy.arange(modelbox[0], modelbox[1] + modelbox[2], modelbox[2]) latp = numpy.arange(modelbox[3], modelbox[4] + modelbox[5], modelbox[5]) rms = {} std = {} snr = {} cov = {} rms['lon'] = lonp rms['lat'] = latp std['lon'] = lonp std['lat'] = latp snr['lon'] = lonp snr['lat'] = latp if 'uwd_est' in listvar: listvar.append('diff_uwdr') listvar.append('diff_ur') listvar.append('cov_ur') listvar.append('cov_uwdr') for ivar in listvar: rms[ivar] = numpy.full((len(latp), len(lonp)), numpy.nan) std[ivar] = numpy.full((len(latp), len(lonp)), numpy.nan) snr[ivar] = numpy.full((len(latp), len(lonp)), numpy.nan) cov[ivar] = numpy.full((len(latp), len(lonp)), numpy.nan) for j in range(len(rms['lon'])): for i in range(len(rms['lat'])): ind_key = 10000 * int(i) + int(j) if ind_key not in dic_v.keys(): continue if 'ur_true' not in dic_v[ind_key].keys(): continue if 'uwd' not in dic_v[ind_key].keys(): continue var1 = numpy.concatenate(dic_v[ind_key]['ur_true']).ravel() if 'uwd' in dic_v[ind_key].keys(): var2 = numpy.concatenate(dic_v[ind_key]['uwd']).ravel() if not var1.any() or not var2.any(): continue rms_true = numpy.sqrt(numpy.nanmean(var1**2)) # import pdb ; pdb.set_trace() for ivar in listvar: if ivar not in dic_v[ind_key].keys(): continue if len(dic_v[ind_key][ivar]) == 0: continue print(len(dic_v[ind_key][ivar])) try: var = numpy.concatenate(dic_v[ind_key][ivar]).ravel() except: continue rms[ivar][i, j] = numpy.sqrt(numpy.nanmean(var**2)) std[ivar][i, j] = numpy.nanstd(var) if ivar == 'diff_ur': snr[ivar][i, j] = rms_true / std[ivar][i, j] elif ivar == 'instr' or ivar == 'diff_uwdr': snr[ivar][i, j] = rms_true / std[ivar][i, j] elif 'cov' in ivar: cov[ivar][i, j] = numpy.nanmean(var) with open('rms_{}'.format(bin_out), 'wb') as f: pickle.dump(rms, f) with open('std_{}'.format(bin_out), 'wb') as f: pickle.dump(std, f) with open('snr_{}'.format(bin_out), 'wb') as f: pickle.dump(snr, f) with open('cov_{}'.format(bin_out), 'wb') as f: pickle.dump(cov, f) def plot_rms(pfile, list_var, outfile, isrms=True, isstd=True, issnr=True, isvar=True): import mod_plot if isrms is True: with open('rms_{}'.format(pfile), 'rb') as f: rms = pickle.load(f) if isstd is True: with open('std_{}'.format(pfile), 'rb') as f: std = pickle.load(f) if issnr is True: with open('snr_{}'.format(pfile), 'rb') as f: snr = pickle.load(f) if isvar is True: with open('cov_{}'.format(pfile), 'rb') as f: cov = pickle.load(f) if 'uwd_est' in list_var: list_var.append('diff_uwdr') list_var.append('diff_ur') for ivar in list_var: if ivar in rms.keys() and isrms is True: lon = rms['lon'] lat = rms['lat'] var = rms[ivar] _outfile = 'rms_{}_{}.png'.format(ivar, outfile) vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') if ivar in std.keys() and isstd is True: lon = std['lon'] lat = std['lat'] var = std[ivar] vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) _outfile = 'std_{}_{}.png'.format(ivar, outfile) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') if ivar in snr.keys(): lon = snr['lon'] lat = snr['lat'] var = snr[ivar] vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) _outfile = 'snr_{}_{}.png'.format(ivar, outfile) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') for ivar in ['cov_uwdr', 'cov_ur']: if ivar in cov.keys(): lon = cov['lon'] lat = cov['lat'] var = cov[ivar] vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) _outfile = 'snr_{}_{}.png'.format(ivar, outfile) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') return None
19,253
38.374233
94
py
skimulator
skimulator-master/test/mod_plot.py
import numpy import netCDF4 import os import sys import glob import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import cartopy import skimulator.rw_data as rw # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() def plot_all(listfile, modelbox, output): # Initialize plot projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() fig = pyplot.figure(figsize=(12,12)) if modelbox is None: projection = cartopy.crs.Orthographic(0, 0) ax = pyplot.axes(projection=projection) if modelbox is not None: ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], crs=transform) norder = 6 else: ax.set_global() norder = 1 #ax.add_feature(cartopy.feature.OCEAN, zorder=norder) #ax.add_feature(cartopy.feature.LAND, zorder=norder, edgecolor='black') gl = ax.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False # Loop on files for ifile in listfile: data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:] lat = data['lat'][:] lon = numpy.mod(lon + 180.0, 360.0) - 180.0 lon_nadir = data['lon_nadir'][:] lon_nadir = numpy.mod(lon_nadir + 180.0, 360) - 180.0 lat_nadir = data['lat_nadir'][:] data.close() pyplot.plot(lon[:, 0], lat[:, 0], '.', color=listcolor[0], markersize=1, transform=transform) pyplot.plot(lon_nadir, lat_nadir, '.', color=listcolor[4], markersize=1, transform=transform) # Save figure pyplot.savefig(output) return fig def plot_variable(listfile, nvar, modelbox, output): # Initialize plot projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() fig = pyplot.figure(figsize=(12,12)) if modelbox is None: projection = cartopy.crs.Orthographic(0, 0) ax = pyplot.axes(projection=projection) if modelbox is not None: ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], crs=transform) norder = 6 else: ax.set_global() norder = 1 #ax.add_feature(cartopy.feature.OCEAN, zorder=norder) ax.add_feature(cartopy.feature.LAND, zorder=norder, edgecolor='black') gl = ax.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False # Loop on files for ifile in listfile: data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:] lat = data['lat'][:] lon = numpy.mod(lon + 180.0, 360.0) - 180.0 lon_nadir = data['lon_nadir'][:] lon_nadir = numpy.mod(lon_nadir + 180.0, 360) - 180.0 lat_nadir = data['lat_nadir'][:] var = data[nvar][:] radial_angle = data['radial_angle'][:] data.close() for i in range(numpy.shape(lon)[1]): if var[:, i].mask.all(): continue try: pyplot.scatter(lon[:, i], lat[:, i], c=var[:, i], cmap='jet', vmin=-1, vmax=1, s=1, #color=listcolor[0], markersize=1, transform=transform) except: continue pyplot.plot(lon_nadir, lat_nadir, '.', color=listcolor[4], markersize=1, transform=transform) try: pyplot.colorbar() except: pass # Save figure # pyplot.savefig(output) return fig def plot_vectors(listfile, nvar, modelbox, output, scale=20): # Initialize plot projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() fig = pyplot.figure(figsize=(12,12)) if modelbox is None: projection = cartopy.crs.Orthographic(0, 0) ax = pyplot.axes(projection=projection) if modelbox is not None: ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], crs=transform) norder = 6 else: ax.set_global() norder = 1 #ax.add_feature(cartopy.feature.OCEAN, zorder=norder) ax.add_feature(cartopy.feature.LAND, zorder=norder, edgecolor='black') gl = ax.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False # Loop on files for ifile in listfile: data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:] lat = data['lat'][:] lon = numpy.mod(lon + 180.0, 360.0) - 180.0 lon_nadir = data['lon_nadir'][:] lon_nadir = numpy.mod(lon_nadir + 180.0, 360) - 180.0 lat_nadir = data['lat_nadir'][:] var = data[nvar][:] radial_angle = data['radial_angle'][:] u_true = data['ucur'][:] v_true = data['vcur'][:] uvar = var * numpy.cos(radial_angle) vvar = var * numpy.sin(radial_angle) data.close() for i in range(numpy.shape(lon)[1]): pyplot.quiver(lon[:, i], lat[:, i], u_true[:, i], v_true[:, i], transform=transform, color='green', scale=scale) pyplot.quiver(lon[:, i], lat[:, i], uvar[:, i], vvar[:, i], transform=transform, color='red', scale=scale) pyplot.plot(lon_nadir, lat_nadir, '.', color=listcolor[4], markersize=1, transform=transform) # Save figure pyplot.savefig(output) def plot_diag(lon, lat, var, outfile, vmin=None, vmax=None, cmap='jet'): lon = numpy.mod(lon + 180.0, 360.0) - 180.0 if vmin is None: vmin = numpy.nanpercentile(var, 1) if vmax is None: vmax = numpy.nanpercentile(var, 99) projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() fig = pyplot.figure(figsize=(12,12)) modelbox = [numpy.min(lon), numpy.max(lon), numpy.min(lat), numpy.max(lat)] ax = pyplot.axes(projection=projection) if modelbox is not None: ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], crs=transform) norder = 6 else: ax.set_global() norder = 1 #ax.add_feature(cartopy.feature.OCEAN, zorder=norder) ax.add_feature(cartopy.feature.LAND, zorder=norder, edgecolor='black') gl = ax.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False c = ax.pcolormesh(lon, lat, var, vmin=vmin, vmax=vmax, cmap=cmap) pyplot.colorbar(c) pyplot.savefig(outfile) return None
6,884
35.428571
80
py
skimulator
skimulator-master/test/plot_grid.py
import mod_plot import params as p import numpy import glob import os # Initialize variables indatadir = p.outdatadir config = p.config modelbox = p.modelbox #modelbox = [-5, 5, 75, 85] #modelbox = [-65, 55, 40, 45] modelbox[0] = numpy.mod(modelbox[0] + 180.0, 360.0) - 180.0 modelbox[1] = numpy.mod(modelbox[1] + 180.0, 360.0) - 180.0 filesgrid = os.path.join(indatadir, '{}_grid'.format(config)) listfile = glob.glob('{}*.nc'.format(filesgrid)) output = os.path.join(indatadir, 'Grid_{}.png'.format(config)) # Plot all Grids: mod_plot.plot_all(listfile, modelbox, output) print('Plot saved in {}'.format(output)) # Plot first 6 passes" listfile = sorted(listfile) output = os.path.join(indatadir, 'Grid_6{}.png'.format(config)) mod_plot.plot_all(listfile[:6], modelbox, output) print('Plot saved in {}'.format(output))
828
26.633333
64
py
skimulator
skimulator-master/test/diag_l2b.py
import numpy import mod_diag import glob import os import matplotlib import json import sys if len(sys.argv) < 1: print('Provide json file for diagnostics') sys.exit(1) file_param = sys.argv[1] with open(file_param, 'r') as f: params = json.load(f) modelbox = params['l2b']['modelbox'] config = params['l2b']['config'] indatadir = params['l2b']['indatadir'] outdatadir = params['l2b']['outdatadir'] list_angle = params['l2b']['list_angle'] # indatadir = params['/tmp/key/data/skim_eq_output/{}'.format(config) filesgrid = os.path.join(indatadir, '{}_grid'.format(config)) listfile = glob.glob('{}*.nc'.format(filesgrid)) files = os.path.join(indatadir, '{}_c01'.format(config)) listfiles = glob.glob('{}*.nc'.format(files)) listfiles = sorted(listfiles) output = os.path.join(outdatadir, 'Grid_{}.png'.format(config)) output2 = os.path.join(outdatadir, 'c_{}.png'.format(config)) print(config) # listfiles.remove('/tmp/key/data/skim_eq_output/WW3_EQ_metop_2018_8a_c01_p022.nc') mod_diag.diag_rms(listfiles[:], modelbox, output, list_angle) listvar = ['ur_true', 'instr', 'ur_obs', 'uwd', 'uwd_est'] listvar = ['ur_true', 'instr', 'ur_obs', 'uwd'] #, 'uwd_est'] modelbox2 = params['l2b']['modelbox_bin'] bin_file = os.path.join(outdatadir, '{}.pyo'.format(config)) #mod_diag.bin_variables(listfiles[:], listvar, bin_file, modelbox2) bin_file2 = '{}.pyo'.format(config) #mod_diag.compute_rms(bin_file, bin_file2, listvar, modelbox2) #mod_diag.plot_rms(bin_file2, listvar, config)
1,499
32.333333
83
py
skimulator
skimulator-master/test/diags_l2d.py
import numpy from matplotlib import pyplot import netCDF4 import glob import os import scipy.signal from scipy.fftpack import fft def cpsd1d(hh1=None, hh2=None, dx=1.,tap=0.05, detrend=True): hh1 = hh1 - numpy.mean(hh1) hh2 = hh2 - numpy.mean(hh2) nx = numpy.shape(hh1)[0] if detrend: hh1 = scipy.signal.detrend(hh1) hh2 = scipy.signal.detrend(hh2) if tap>0: ntaper = numpy.int(tap * nx + 0.5) taper = numpy.zeros(nx)+1. taper[:ntaper] = numpy.cos(numpy.arange(ntaper) / (ntaper-1.) *numpy.pi/2 + 3*numpy.pi/2) taper[-ntaper:] = numpy.cos(-numpy.arange(-ntaper+1, 1)/(ntaper-1.) *numpy.pi/2 + 3*numpy.pi/2) hh1 = hh1 * taper hh2 = hh2 * taper ss1 = fft(hh1) ss2 = fft(hh2) ff = numpy.arange(1, nx/2 - 1) / (nx*dx) C = numpy.cos(numpy.angle(ss1[1: int(nx/2)-1]) - numpy.angle(ss2[1: int(nx/2)-1])) return ff, C def coherency_l2d(datadir_input, config, var, nal_min, posting, basename, outfile='coherency'): list_files_all = [] pyplot.figure() for indir, iconfig, ivar in zip(datadir_input, config, var): print(indir, iconfig, ivar) list_files = glob.glob(os.path.join(indir, '{}_l2d_*.nc'.format(iconfig))) fid = netCDF4.Dataset(list_files[0], 'r') varx = 'ux_{}'.format(var) vary = 'uy_{}'.format(var) _tmp = numpy.array(fid.variables[varx][:]) _tmp[_tmp < -10] = numpy.nan fid.close() nac = _tmp.shape[1] nal = nac dal = posting #km MCuac = [] countuac = 0 MCual = [] countual = 0 for ifile in list_files: ref = {} skim = {} fid = netCDF4.Dataset(ifile, 'r') skim[varx] = numpy.array(fid.variables[varx][:]) skim[varx][ref['ux_obs']<-10] = numpy.nan skim[varx] = numpy.ma.masked_invalid(ref[varx]) ref['ux_true'] = numpy.array(fid.variables['ux_true'][:]) ref['ux_true'][ref['ux_true']<-10] = numpy.nan ref['ux_true'] = numpy.ma.masked_invalid(ref['ux_true']) skim[vary] = numpy.array(fid.variables[vary][:]) skim[vary][skim[vary]<-10] = numpy.nan skim[vary] = numpy.ma.masked_invalid(skim[vary]) ref['uy_true'] = numpy.array(fid.variables['uy_true'][:]) ref['uy_true'][ref['uy_true']<-10] = numpy.nan ref['uy_true'] = numpy.ma.masked_invalid(ref['uy_true']) fid.close() if numpy.shape(skim[varx])[1] < nac: continue idate = ifile[-18:-9] print(idate) #mfile = '{}{}{}.nc'.format(basename[0], idate, basename[1]) #fidm = netCDF4.Dataset(mfile, 'r') #ref['ucur'] = numpy.array(fid.variables['ucur'][:]) for i in range(nac): checknanref = + ref['ux_true'][:, i] checknanobs = + skim[varx][:, i] merged_mask = (checknanref.mask | checknanobs.mask) indok = numpy.where(~merged_mask)[0] _ind = numpy.where(indok[1:] - indok[:-1] > 1) ensind = numpy.split(indok, numpy.cumsum(_ind) + 1) for chunk in range(len(ensind)): if len(ensind[chunk]) > nal_min: s1 = ref['ux_true'][ensind[chunk][:nal_min], i] s2 = skim[varx][ensind[chunk][:nal_min], i] ffac, C = cpsd1d(hh1=s1, hh2=s2, dx=dal, tap=0.5, detrend=True) countuac += 1 try: MCuac += C except: MCuac = +C s1 = ref['uy_true'][ensind[chunk][:nal_min], i] s2 = skim[vary][ensind[chunk][:nal_min], i] try: ffal, C = cpsd1d(hh1=s1, hh2=s2, dx=dal, tap=0.5, detrend=True) countual += 1 except: continue try: MCual += C except: MCual = +C MCuac = numpy.array(MCuac) / countuac MCual = numpy.array(MCual) / countual pyplot.semilogx(ffac, MCuac, label='{} {}'.format(ivar, iconfig)) # ,c='k') #pyplot.semilogx(ffal, MCual, label='along {} {}'.format(ivar, iconfig)) # ,c='k') pyplot.grid() pyplot.legend() pyplot.xlabel('/cy/km') pyplot.ylabel('coherency') pyplot.savefig('coherency_{}.png'.format(outfile)) def rms_l2d(datadir_input, config): datadir_output = './' glob_files = os.path.join(datadir_input, '{}_l2d_*.nc'.format(config)) list_files = glob.glob(glob_files) ref = {} skim = {} fid = netCDF4.Dataset(list_files[0], 'r') _tmp = numpy.array(fid.variables['ux_true'][:]) #_tmp = numpy.array(fid.variables['u_ac_true'][:]) _tmp[_tmp < -10] = numpy.nan fid.close() nac = _tmp.shape[1] nal = nac std_uac = numpy.zeros(nac) std_ual = numpy.zeros(nal) ntot_ac = numpy.zeros(nac) ntot_al = numpy.zeros(nal) std_uacm = numpy.zeros(nac) std_ualm = numpy.zeros(nal) ntot_acm = numpy.zeros(nac) ntot_alm = numpy.zeros(nal) for filev in list_files: fid = netCDF4.Dataset(filev, 'r') ipath = int(filev[-6:-3]) if ipath > 400: continue # if ipath%2==0: # continue ref['uac'] = numpy.array(fid.variables['uy_true'][:]) ref['uac'][ref['uac'] < -10] = numpy.nan ref['ual'] = numpy.array(fid.variables['ux_true'][:]) ref['ual'][ref['ual'] < -10] = numpy.nan skim['uacm'] = numpy.array(fid.variables['uy_noerr'][:]) skim['uacm'][skim['uacm'] < -10] = numpy.nan skim['ualm'] = numpy.array(fid.variables['ux_noerr'][:]) skim['ualm'][skim['ualm'] < -10] = numpy.nan skim['uac'] = numpy.array(fid.variables['uy_obs'][:]) skim['uac'][skim['uac'] < -10] = numpy.nan skim['ual'] = numpy.array(fid.variables['ux_obs'][:]) skim['ual'][skim['ual'] < -10] = numpy.nan nuac = ref['uac'].shape[1] fid.close() if nuac < 61: continue for i in range(nuac): it_ac = len(numpy.where(numpy.isnan(skim['uac'][:, i]) == False)[0]) if it_ac >= 62: std_uac[i] += numpy.nanstd(skim['uac'][:, i] - ref['uac'][:, i])*it_ac ntot_ac[i] += it_ac it_al = len(numpy.where(numpy.isnan(skim['ual'][:, i]) == False)[0]) if it_al >= 62: std_ual[i] += numpy.nanstd(skim['ual'][:, i] - ref['ual'][:, i])*it_al ntot_al[i] += it_al it_acm = len(numpy.where(numpy.isnan(skim['uacm'][:, i]) == False)[0]) if it_acm >= 62: std_uacm[i] += numpy.nanstd(skim['uacm'][:, i] - ref['uac'][:, i])*it_acm ntot_acm[i] += it_acm it_alm = len(numpy.where(numpy.isnan(skim['ualm'][:, i]) == False)[0]) if it_alm >= 62: std_ualm[i] += numpy.nanstd(skim['ualm'][:, i] - ref['ual'][:, i])*it_alm ntot_alm[i] += it_alm std_uac = std_uac/ntot_ac std_ual = std_ual/ntot_al std_uacm = std_uacm/ntot_acm std_ualm = std_ualm/ntot_alm f, (ax1, ax2) = pyplot.subplots(1, 2, sharey= True, figsize=(12,5 )) xac = numpy.arange(-(nac - 1) * p.posting/2, (nac + 1)* p.posting/2, p.posting) _ind = numpy.where(numpy.abs(xac)>40) print(config, 'uac', numpy.nanmean(std_uac[_ind])) print(config, 'ual', numpy.nanmean(std_ual[_ind])) print(config, 'uacm', numpy.nanmean(std_uacm[_ind])) print(config, 'ualm', numpy.nanmean(std_ualm[_ind])) ax1.plot(xac, std_uac, 'r', label='eastward') ax1.plot(xac, std_ual, 'b', label='northward') ax1.set_title('Observation {}'.format(config)) ax1.set_ylim([0.00, 0.18]) ax1.legend() ax2.plot(xac, std_uacm, 'r', label='eastward') ax2.plot(xac, std_ualm, 'b', label='northward') ax2.set_title('Error-free {}'.format(config)) ax2.legend() pyplot.savefig('std_{}.png'.format(config)) if '__main__' == __name__: import params as p # rms_l2c(p.outdatadir, p.config) length_al = 200 #coherency_l2c((p.outdatadir, p.outdatadir), (p.config, p.config), # ('obs','model'), length_al, # p.posting, outfile='{}_obs_model'.format(p.config)) list_config = ('WW3_AT_metop_2018_8a', 'WW3_AT_metop_2018_8b', 'WW3_AT_metop_2018_8c', 'WW3_AT_metop_2018_6a') list_config = ('WW3_EQ_metop_2018_8a', 'WW3_EQ_metop_2018_8b', 'WW3_EQ_metop_2018_8c', 'WW3_EQ_metop_2018_6a') #list_config = ('WW3_FR_metop_2018_8a', 'WW3_FR_metop_2018_8b', # 'WW3_FR_metop_2018_8c', 'WW3_FR_metop_2018_6a') list_dir = [] for iconfig in list_config: outdatadir = os.path.join('/tmp/key/data/skim_at_output', iconfig) outdatadir = os.path.join('/tmp/key/data/skim_eq_output', iconfig) # outdatadir = os.path.join('/tmp/key/data/skim_fr_output', iconfig) list_dir.append(outdatadir) rms_l2c(outdatadir, iconfig) coherency_l2c((outdatadir, outdatadir), (iconfig, iconfig), ('obs','model'), length_al, p.posting, outfile='{}_obs_model'.format(iconfig)) list_var = ('obs', 'obs', 'obs', 'obs') #, 'obs') nal_min = 200 print(list_dir, list_config) coherency_l2c(list_dir, list_config, list_var, nal_min, p.posting, outfile=list_config[0][:-8])
10,114
38.666667
90
py
skimulator
skimulator-master/test/diags_l2c.py
import numpy from matplotlib import pyplot import netCDF4 import glob import os import sys import json import scipy.signal from scipy.fftpack import fft def cpsd1d(hh1=None, hh2=None, dx=1.,tap=0.05, detrend=True): hh1 = hh1 - numpy.mean(hh1) hh2 = hh2 - numpy.mean(hh2) nx = numpy.shape(hh1)[0] if detrend: hh1 = scipy.signal.detrend(hh1) hh2 = scipy.signal.detrend(hh2) if tap>0: ntaper = numpy.int(tap * nx + 0.5) taper = numpy.zeros(nx)+1. taper[:ntaper] = numpy.cos(numpy.arange(ntaper) / (ntaper-1.) *numpy.pi/2 + 3*numpy.pi/2) taper[-ntaper:] = numpy.cos(-numpy.arange(-ntaper+1, 1)/(ntaper-1.) *numpy.pi/2 + 3*numpy.pi/2) hh1 = hh1 * taper hh2 = hh2 * taper ss1 = fft(hh1) ss2 = fft(hh2) ff = numpy.arange(1, nx/2 - 1) / (nx*dx) C = numpy.cos(numpy.angle(ss1[1: int(nx/2)-1]) - numpy.angle(ss2[1: int(nx/2)-1])) return ff, C def coherency_l2c(datadir_input, config, var, nal_min, posting, outfile='coherency', fsize=1): list_files_all = [] pyplot.figure() for indir, iconfig, ivar in zip(datadir_input, config, var): print(indir, iconfig, ivar) list_files = glob.glob(os.path.join(indir, '{}_l2c_c*p*.nc'.format(iconfig))) fid = netCDF4.Dataset(list_files[0], 'r') try: _tmp = numpy.array(fid.variables['u_ac_true'][:]) except: continue _tmp[_tmp < -10] = numpy.nan fid.close() nac = _tmp.shape[1] nac_mid = int(nac/2) nal = nac dal = posting #km MCuac = [] countuac = 0 MCual = [] countual = 0 for ifile in list_files: ref = {} skim = {} fid = netCDF4.Dataset(ifile, 'r') try: ref['uac'] = numpy.array(fid.variables['u_ac_true'][:]) ref['uac'] = scipy.ndimage.uniform_filter(ref['uac'], size=fisze) except: print(ifile) continue ref['uac'][ref['uac']<-10] = numpy.nan ref['uac'] = numpy.ma.masked_invalid(ref['uac']) ref['ual'] = numpy.array(fid.variables['u_al_true'][:]) ref['ual'] = scipy.ndimage.uniform_filter(ref['ual'], size=fsize) ref['ual'][ref['ual']<-10] = numpy.nan ref['ual'] = numpy.ma.masked_invalid(ref['ual']) skim['uac'] = numpy.array(fid.variables['u_ac_{}'.format(ivar)][:]) skim['uac'][numpy.abs(skim['uac']) > 10] = numpy.nan skim['uac'] = numpy.ma.masked_invalid(skim['uac']) skim['ual'] = numpy.array(fid.variables['u_al_{}'.format(ivar)][:]) skim['ual'][numpy.abs(skim['ual']) > 10] = numpy.nan skim['ual'] = numpy.ma.masked_invalid(skim['ual']) fid.close() if numpy.shape(skim['ual'])[1] < nac: continue for i in range(3, nac-3): checknanref = + ref['uac'][:, i] checknanobs = + skim['uac'][:, i] merged_mask = (checknanref.mask | checknanobs.mask) indok = numpy.where(~merged_mask)[0] _ind = numpy.where(indok[1:] - indok[:-1] > 1) ensind = numpy.split(indok, numpy.cumsum(_ind) + 1) for chunk in range(len(ensind)): if len(ensind[chunk]) > nal_min: s1 = ref['uac'][ensind[chunk][:nal_min], i] s2 = skim['uac'][ensind[chunk][:nal_min], i] ffac, C = cpsd1d(hh1=s1, hh2=s2, dx=dal, tap=0.5, detrend=True) countuac += 1 try: MCuac += C except: MCuac = +C s1 = ref['ual'][ensind[chunk][:nal_min], i] s2 = skim['ual'][ensind[chunk][:nal_min], i] try: ffal, C = cpsd1d(hh1=s1, hh2=s2, dx=dal, tap=0.5, detrend=True) countual += 1 except: continue try: MCual += C except: MCual = +C MCuac = numpy.array(MCuac) / countuac MCual = numpy.array(MCual) / countual pyplot.semilogx(ffac, MCuac, label='{} {}'.format(ivar, iconfig)) # ,c='k') #pyplot.semilogx(ffal, MCual, label='along {} {}'.format(ivar, iconfig)) # ,c='k') pyplot.grid() pyplot.legend() pyplot.xlabel('/cy/km') pyplot.ylabel('coherency') pyplot.savefig('{}.png'.format(outfile)) def rms_l2c(datadir_input, config, output, threshold=0.1, fsize=1): datadir_output = output glob_files = os.path.join(datadir_input, '{}_l2c_c01*p*.nc'.format(config)) list_files = glob.glob(glob_files) ref = {} skim = {} print(datadir_input, config) fid = netCDF4.Dataset(list_files[5], 'r') try: _tmp = numpy.array(fid.variables['u_ac_true'][:]) except: print(list_files[0]) #_tmp = numpy.array(fid.variables['u_ac_true'][:]) _tmp[_tmp < -10] = numpy.nan fid.close() nac = _tmp.shape[1] nal = nac std_uac = numpy.zeros(nac) std_ual = numpy.zeros(nal) ntot_ac = numpy.zeros(nac) ntot_al = numpy.zeros(nal) std_uacm = numpy.zeros(nac) std_ualm = numpy.zeros(nal) ntot_acm = numpy.zeros(nac) ntot_alm = numpy.zeros(nal) print(nac) list_key = [#'u_al_wd', 'u_ac_wd',# 'u_al_wdrem', 'u_ac_wdrem', 'u_ac_instr', 'u_al_instr',# 'u_ac_dsigma', 'u_al_dsigma', 'u_ac_true', 'u_al_true'] list_fkey = ['uwnd', 'vwnd'] #, 'rain', 'mssu', 'mssc'] std_err = {} ntot_err = {} for ikey in list_key: std_err[ikey] = numpy.zeros(nac) ntot_err[ikey] = numpy.zeros(nac) for filev in list_files: fid = netCDF4.Dataset(filev, 'r') ipath = int(filev[-6:-3]) #if ipath > 400: # continue # if ipath%2==0: # continue try: ref['uac'] = numpy.array(fid.variables['u_ac_true'][:]) ref['uac'] = scipy.ndimage.uniform_filter(ref['uac'], size=fsize) except: print(filev) continue ref['uac'][numpy.abs(ref['uac']) > 10] = numpy.nan ref['ual'] = numpy.array(fid.variables['u_al_true'][:]) ref['ual'] = scipy.ndimage.uniform_filter(ref['ual'], size=fsize) ref['ual'][numpy.abs(ref['ual']) > 10] = numpy.nan skim['uacm'] = numpy.array(fid.variables['u_ac_noerr'][:]) skim['uacm'][numpy.abs(skim['uacm']) > 10] = numpy.nan skim['ualm'] = numpy.array(fid.variables['u_al_noerr'][:]) skim['ualm'][numpy.abs(skim['ualm']) > 10] = numpy.nan skim['uac'] = numpy.array(fid.variables['u_ac_obs'][:]) skim['uac'][numpy.abs(skim['uac']) > 10] = numpy.nan skim['ual'] = numpy.array(fid.variables['u_al_obs'][:]) skim['ual'][numpy.abs(skim['ual']) > 10] = numpy.nan for ikey in list_fkey: skim[ikey] = numpy.array(fid.variables[ikey][:]) wnd = numpy.sqrt(skim['uwnd']**2 + skim['vwnd']**2) for ikey in list_key: skim[ikey] = numpy.array(fid.variables[ikey][:]) skim[ikey][numpy.abs(skim[ikey]) > 100] = numpy.nan skim['uac'][numpy.isnan(skim[ikey])] = numpy.nan skim['ual'][numpy.isnan(skim[ikey])] = numpy.nan skim[ikey][numpy.where((wnd < 3) & (wnd < 3))] = numpy.nan # skim[ikey][numpy.where((skim['rain'] > 0.1))] = numpy.nan skim['uac'][numpy.where((wnd < 3) & (wnd < 3))] = numpy.nan skim['ual'][numpy.where((wnd <3) & (wnd < 3))] = numpy.nan #skim['uac'][numpy.where((skim['rain'] > 0.1))] = numpy.nan #skim['ual'][numpy.where((skim['rain'] > 0.1))] = numpy.nan #_indwdal = numpy.where(abs(skim['u_al_wdrem']) > 1) #_indwdac = numpy.where(abs(skim['u_ac_wdrem']) > 1) #skim['uac'][_indwdac] = numpy.nan #skim['ual'][_indwdal] = numpy.nan #skim['u_ac_wdrem'][_indwdac] = numpy.nan #skim['u_al_wdrem'][_indwdal] = numpy.nan # skim['ual'][numpy.isnan(skim[ikey])] = numpy.nan nuac = ref['uac'].shape[1] fid.close() if nuac < 61: continue ind0 = 50 ind1 = -50 # skim['ual'] = skim['ualm'] + skim['u_al_instr'] + skim['u_al_dsigma'] + skim['u_al_wdrem'] # skim['uac'] = skim['uacm'] + skim['u_ac_instr'] + skim['u_ac_dsigma'] + skim['u_ac_wdrem'] for i in range(nuac): it_ac = len(numpy.where(numpy.isnan(skim['uac'][ind0:ind1, i]) == False)[0]) if it_ac >= 100: _std = numpy.nanstd(skim['uac'][ind0:ind1, i] - ref['uac'][ind0:ind1, i])*it_ac if numpy.isfinite(_std): std_uac[i] += _std ntot_ac[i] += it_ac else: std_uac[i] += 0 ntot_ac[i] += 0 it_al = len(numpy.where(numpy.isnan(skim['ual'][ind0:ind1, i]) == False)[0]) if it_al >= 100: _std = numpy.nanstd(skim['ual'][ind0:ind1, i] - ref['ual'][ind0:ind1, i])*it_al if numpy.isfinite(_std): std_ual[i] += _std ntot_al[i] += it_al else: std_ual[i] += 0 ntot_al[i] += 0 it_acm = len(numpy.where(numpy.isnan(skim['uacm'][ind0:ind1, i]) == False)[0]) if it_acm >= 100: _std = numpy.nanstd(skim['uacm'][ind0:ind1, i] - ref['uac'][ind0:ind1, i])*it_acm if numpy.isfinite(_std): std_uacm[i] += _std ntot_acm[i] += it_acm else: std_uacm[i] += 0 ntot_acm[i] += 0 it_alm = len(numpy.where(numpy.isnan(skim['ualm'][ind0:ind1, i]) == False)[0]) if it_alm >= 100: _std = numpy.nanstd(skim['ualm'][ind0:ind1, i] - ref['ual'][ind0:ind1, i])*it_alm if numpy.isfinite(_std): std_ualm[i] += _std ntot_alm[i] += it_alm else: std_ualm[i] += 0 ntot_alm[i] += 0 for ikey in list_key: it_alm = len(numpy.where(numpy.isnan(skim[ikey][ind0:ind1, i]) == False)[0]) if it_alm >= 100: _std = numpy.nanstd(skim[ikey][ind0:ind1, i]) * it_alm if numpy.isfinite(_std): std_err[ikey][i] += _std ntot_err[ikey][i] += it_alm else: std_err[ikey][i] += 0 ntot_err[ikey][i] += 0 std_uac = std_uac/ntot_ac std_ual = std_ual/ntot_al std_uacm = std_uacm/ntot_acm std_ualm = std_ualm/ntot_alm f, (ax1, ax2, ax3) = pyplot.subplots(1, 3, sharey= True, figsize=(20,5 )) xac = numpy.arange(-(nac - 1) * posting/2, (nac + 1)* posting/2, posting) indx = numpy.where(abs(xac)<= 40) #std_uac[indx] = numpy.nan #std_ual[indx] = numpy.nan print(xac[numpy.where(std_uac > threshold)]) print(xac[numpy.where(std_ual > threshold)]) _ind = numpy.where((numpy.abs(xac)>50) & (numpy.abs(xac)<400)) print(config, 'uac', numpy.nanmean(std_uac[_ind])) print(config, 'ual', numpy.nanmean(std_ual[_ind])) print(config, 'uacm', numpy.nanmean(std_uacm[_ind])) print(config, 'ualm', numpy.nanmean(std_ualm[_ind])) for ikey in list_key: std_err[ikey] = std_err[ikey] / ntot_err[ikey] print(config, ikey, numpy.nanmean(std_err[ikey][_ind])) ax1.plot(xac, std_uac, 'r', label='across track') ax1.plot(xac, std_ual, 'b', label='along track') ax1.axhline(y=0.15, color="0.5") ax1.set_title('Observation {}'.format(config)) ax1.set_ylim([0.00, 0.4]) ax1.legend() ax2.plot(xac, std_uacm, 'r', label='across track') ax2.plot(xac, std_ualm, 'b', label='along track') ax2.axhline(y=0.15, color="0.5") ax2.set_title('Error-free {}'.format(config)) ax2.legend() ax3.plot(xac, std_uacm, 'r', label='across track regridding') ax3.plot(xac, std_ualm, 'b', label='along track regridding') for ikey in list_key: std_err[ikey][indx] = numpy.nan ax3.plot(xac, std_err[ikey], label=ikey) ax3.axhline(y=0.15, color="0.5") ax3.set_title('Error decomposition {}'.format(config)) ax3.legend() pyplot.savefig('{}.pdf'.format(datadir_output)) def bin_variables(listfile, listvar, bin_in, modelbox): lonp = numpy.arange(modelbox[0], modelbox[1] + modelbox[2], modelbox[2]) latp = numpy.arange(modelbox[3], modelbox[4] + modelbox[5], modelbox[5]) resol = numpy.sqrt((modelbox[2] * numpy.cos(numpy.deg2rad(latp))) **2 + modelbox[5]**2) dic_v = {} for ifile in listfile: print(ifile) data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:] lat = data['lat'][:] for j in range(len(lonp)): for i in range(len(latp)): ind_key = 10000 * int(i) + int(j) lon = numpy.mod(lon + 180, 360) - 180 dist = numpy.sqrt(((lonp[j] - lon) *numpy.cos(numpy.deg2rad(lat)))**2 + (latp[i] - lat)**2) iiobs = numpy.where(dist < resol[i]) if iiobs[0].shape == (0, ): continue if ind_key not in dic_v.keys(): dic_v[ind_key] = {} for ivar in listvar: ivaro = '{}_obs'.format(ivar) ivart = '{}_noerr'.format(ivar) ivarr = '{}_true'.format(ivar) idvaro = '{}_diff_obs'.format(ivar) idvart = '{}_diff_noerr'.format(ivar) if ivar == 'norm': varo = numpy.sqrt(data['u_ac_obs'][iiobs]**2 + data['u_al_obs'][iiobs]**2) vart = numpy.sqrt(data['u_ac_noerr'][iiobs]**2 + data['u_al_noerr'][iiobs]**2) varr = numpy.sqrt(data['u_ac_true'][iiobs]**2 + data['u_al_true'][iiobs]**2) else: try: varo = numpy.ma.array(data[ivaro][iiobs]) except: print(ifile, ivar, iiobs) vart = data[ivart][iiobs] varr = data[ivarr][iiobs] mask = varo.mask dvaro = numpy.array(abs(varo.data - varr.data)) dvart = numpy.array(abs(vart.data - varr.data)) if varo.any(): if ivaro not in dic_v[ind_key].keys(): dic_v[ind_key][ivaro] = [] if ivart not in dic_v[ind_key].keys(): dic_v[ind_key][ivart] = [] if ivarr not in dic_v[ind_key].keys(): dic_v[ind_key][ivarr] = [] if idvaro not in dic_v[ind_key].keys(): dic_v[ind_key][idvaro] = [] if idvart not in dic_v[ind_key].keys(): dic_v[ind_key][idvart] = [] dic_v[ind_key][ivaro].append(varo[~mask]) dic_v[ind_key][ivart].append(vart[~mask]) dic_v[ind_key][ivarr].append(varr[~mask]) dic_v[ind_key][idvart].append(dvart[~mask]) dic_v[ind_key][idvaro].append(dvaro[~mask]) dic_v[ind_key][ivart].append(varo[~mask]) dic_v[ind_key][ivart].append(varo[~mask]) dic_v[ind_key][ivart].append(varo[~mask]) data.close() with open(bin_in, 'wb') as f: pickle.dump(dic_v, f) def compute_rms(bin_in, bin_out, listvar, modelbox): with open(bin_in, 'rb') as f: dic_v = pickle.load(f) lonp = numpy.arange(modelbox[0], modelbox[1] + modelbox[2], modelbox[2]) latp = numpy.arange(modelbox[3], modelbox[4] + modelbox[5], modelbox[5]) rms = {} std = {} snr = {} cov = {} rms['lon'] = lonp rms['lat'] = latp std['lon'] = lonp std['lat'] = latp snr['lon'] = lonp snr['lat'] = latp for ivar in listvar: _list = ['_obs', '_noerr', '_true', '_diff_obs', 'diff_noerr'] for ivar2 in _list: ivar3 = '{}{}'.format(ivar, ivar2) rms[ivar3] = numpy.full((len(latp), len(lonp)), numpy.nan) std[ivar3] = numpy.full((len(latp), len(lonp)), numpy.nan) snr[ivar3] = numpy.full((len(latp), len(lonp)), numpy.nan) for j in range(len(rms['lon'])): for i in range(len(rms['lat'])): ind_key = 10000 * int(i) + int(j) if ind_key not in dic_v.keys(): continue # import pdb ; pdb.set_trace() for ivar in listvar: ivaro = '{}_obs'.format(ivar) ivart = '{}_noerr'.format(ivar) ivarr = '{}_true'.format(ivar) idvaro = '{}_diff_obs'.format(ivar) idvart = '{}_diff_noerr'.format(ivar) if ivaro not in dic_v[ind_key].keys(): continue varo = numpy.concatenate(dic_v[ind_key][ivar]).ravel() vart = numpy.concatenate(dic_v[ind_key][ivar]).ravel() varr = numpy.concatenate(dic_v[ind_key][ivar]).ravel() rms_true = numpy.sqrt(numpy.nanmean(varr**2)) rms[ivaro][i, j] = numpy.sqrt(numpy.nanmean(varo**2)) std[ivaro][i, j] = numpy.nanstd(varo) snr[ivaro][i, j] = rms_true / std[idvaro][i, j] rms[ivart][i, j] = numpy.sqrt(numpy.nanmean(vart**2)) std[ivart][i, j] = numpy.nanstd(vart) snr[ivart][i, j] = rms_true / std[idvart][i, j] rms[ivarr][i, j] = numpy.sqrt(numpy.nanmean(varr**2)) std[ivarr][i, j] = numpy.nanstd(varr) with open('rms_{}'.format(bin_out), 'wb') as f: pickle.dump(rms, f) with open('std_{}'.format(bin_out), 'wb') as f: pickle.dump(std, f) with open('snr_{}'.format(bin_out), 'wb') as f: pickle.dump(snr, f) def plot_rms(pfile, list_var, outfile, isrms=True, isstd=True, issnr=True, isvar=True): import mod_plot if isrms is True: with open('rms_{}'.format(pfile), 'rb') as f: rms = pickle.load(f) if isstd is True: with open('std_{}'.format(pfile), 'rb') as f: std = pickle.load(f) if issnr is True: with open('snr_{}'.format(pfile), 'rb') as f: snr = pickle.load(f) if isvar is True: with open('cov_{}'.format(pfile), 'rb') as f: cov = pickle.load(f) list_var = rms.keys() for ivar in list_var: if ivar in rms.keys() and isrms is True: lon = rms['lon'] lat = rms['lat'] var = rms[ivar] _outfile = 'rms_{}_{}.png'.format(ivar, outfile) vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') if ivar in std.keys() and isstd is True: lon = std['lon'] lat = std['lat'] var = std[ivar] vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) _outfile = 'std_{}_{}.png'.format(ivar, outfile) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') if ivar in snr.keys(): lon = snr['lon'] lat = snr['lat'] var = snr[ivar] vmin = numpy.nanpercentile(var, 5) vmax = numpy.nanpercentile(var, 95) _outfile = 'snr_{}_{}.png'.format(ivar, outfile) mod_plot.plot_diag(lon, lat, var, _outfile, vmin=vmin, vmax=vmax, cmap='jet') return None if '__main__' == __name__: if len(sys.argv) < 1: print('Provide json file for diagnostics') sys.exit(1) file_param = sys.argv[1] with open(file_param, 'r') as f: params = json.load(f) # rms_l2c(p.outdatadir, p.config) pl2c = params['l2c'] length_al = pl2c['alongtrack_length'] list_config = pl2c['list_config'] list_dir = pl2c['indatadir'] posting = pl2c['posting'] outdir = pl2c['outdatadir'] list_filter_size = pl2c['filter_size'] for i, iconfig in enumerate(list_config): filter_size = list_filter_size[i] indatadir = list_dir[i] # print(indatadir, iconfig) outfile = os.path.join(outdir, 'std_{}'.format(iconfig)) rms_l2c(indatadir, iconfig, outfile, threshold=0.15, fsize=filter_size) outfile = os.path.join(outdir, 'coherency_{}_obs_model'.format(iconfig)) #coherency_l2c((indatadir, indatadir), (iconfig, iconfig), # ('obs','noerr'), length_al, # posting, outfile=outfile, fsize=filter_size) list_var = ('obs', 'obs', 'obs', 'obs') #, 'obs') nal_min = length_al # print(list_dir, list_config) outfile = os.path.join(outdir, 'coherency_obs_{}'.format(list_config[0][:-8])) #coherency_l2c(list_dir, list_config, list_var, nal_min, # posting, outfile=outfile, fsize=filter_size) list_var = ('noerr', 'noerr', 'noerr', 'noerr') #, 'obs') print(list_dir, list_config) outfile = os.path.join(outdir, 'coherency_noerr_{}'.format(list_config[0][:-8])) #coherency_l2c(list_dir, list_config, list_var, nal_min, # posting, outfile=outfile, fsize=filter_size) listvar = ['u_ac', 'u_al', 'ux', 'uy', 'norm'] modelbox2 = params['l2b']['modelbox_bin'] for i, iconfig in enumerate(list_config): filter_size = list_filter_size[i] indatadir = list_dir[i] print(indatadir, iconfig) files = os.path.join(indatadir, '{}_l2c_c01'.format(iconfig)) listfiles = glob.glob('{}*.nc'.format(files)) bin_file = os.path.join(outdir, '{}_l2c.pyo'.format(iconfig)) # bin_variables(listfiles[:10], listvar, bin_file, modelbox2) # bin_file2 = '{}_l2c.pyo'.format(iconfig) # compute_rms(bin_file, bin_file2, listvar, modelbox2) # plot_rms(bin_file2, listvar, iconfig)
23,170
41.282847
101
py
skimulator
skimulator-master/example/params_example_8beams.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") # ------ Name of the configuration (to build output files names) # 8 beams, 45 azimuths, 1024 pulses and cycle length of 37 ms config = "WW3_23W_metop_2018_8a" # 8 beams, 72 azimuths, 0512 pulses and cycle length of 18 ms # config = "WW3_23W_metop_2018_8b" # 8 beams, 108 azimuths, 1024 pulses and cycle length of 37 ms # config = "WW3_23W_metop_2018_8c" # ------ Directory that contains orbit file: dir_setup = os.path.join(home, 'skimulator', 'data') # ------ Directory that contains your own inputs: indatadir = os.path.join(home, 'skimulator', 'example', 'input_fields') # ------ Directory that contains your outputs: outdatadir = os.path.join(home, 'skimulator', 'example', 'skim_output') # ------ Orbit file: #filesat = os.path.join(dir_setup,'orbs1a.txt') filesat = os.path.join(dir_setup,'orbmetop_skim.txt') # ------ Number of days in orbit (optional if specified in orbit file) satcycle = 29 # ------ Satellite elevation (optional if specified in orbit file) sat_elev = 817 * 10**3 # ------ Order of columns (lon, lat, time) in orbit file # (default is (0, 1, 2) with order_orbit_col = None) order_orbit_col = None # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Number of processor for parallelisation proc_number = 1 # ------ Deactivate printing of progress bar to avoid huge log progress_bar = True # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) # ------ Force the computation of the satellite grid: makesgrid = True # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = None #[329.,347., -8.,8.] #------- Rotation speed of the antenna (in tr/min) if '2018_8a' in config: rotation_speed = 4.52 if '2018_8b' in config: rotation_speed = 5.66 if '2018_8c' in config: rotation_speed = 1.89 # Cycle duration cycle = 0.0368 if '2018_8b' in config: cycle = 0.0368 / 2. # cycle = 0.0368 / 2. # for 512 pulses configuration #------- List of position of beams: list_pos = (0, 72*math.pi/180., 144*math.pi/180., 216*math.pi / 180., 288*math.pi/180., 0, math.pi) #------- List of angle of beams in degrees: list_angle = (12, 12, 12, 12, 12, 6, 6) #------- List of timeshift as regard to nadir for 12 degree beams: list_shift = (1, 2, 4, 5, 7, 3, 6) # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, 'list_of_file.txt') # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other option is WW3) model = 'WW3' # ------ First time of the model first_time = '2011-11-15T00:00:00Z' # ------ Grid file name file_grid_model = (os.path.join(indatadir, 'ww3.20111115_cur.nc'),) # ------ Specify if there is a ice mask for high latitudes # (if true, mask is recomputed at each cycle) ice_mask = False # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify list of variable: list_input_var = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0], 'uuss': ['uuss', 'uss', 0], 'vuss': ['vuss', 'uss', 0], 'ice': ['ice', 'ice', 0], 'mssd': ['mssd', 'msd', 0], 'mssx': ['mssu', 'mss', 0], 'mssy':['mssc', 'mss', 0], 'ssh': ['wlv', 'wlv', 0], 'hs':['hs', 'hs', 0], 'uwnd': ['uwnd', 'wnd', 0], 'vwnd': ['vwnd', 'wnd', 0]} # ------ Specify longitude variable: lon = ('longitude',) # ------ Specify latitude variable: lat = ('latitude',) # ------ Specify number of time in file: dim_time = 24 # ------ Time step between two model outputs (in days): timestep = 1/24. # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 35*24 # ------ Not a number value: model_nan = -32767. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'linear' # -----------------------# # SKIM error parameters # -----------------------# # List of errors to compute noise = ['Altimeter', 'Instrument', 'Attitude', 'DSigma', 'Rain', 'WaveDoppler'] # Number of seed for automatic computation nseed = 0 # Length of repeat for altimetric noise len_repeat = 20000 # Todo compute automatically with cycle, number of beams and satellite velocity delta_al = 2 yaw_file = os.path.join(dir_setup, 'sample_req1.nc') # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Choice of instrument configuration instr_configuration = 'A' # ------- Coefficient SNR to retrieve instrumental noise from sigma, # Recommanded value for 1024 pulses: 3e-2, for 512 pulses: 3sqrt(2)e-3 snr_coeff = 6e-3 if '2018_8b' in config: snr_coeff = 1.4142*6e-3 # ------- File which provide the AOCS error: yaw_file = os.path.join(dir_setup, 'sample_req1.nc') ## -- Geophysical error ## ---------------------- # ------ Consider ice in sigma0 computation ice = True # ------ Rain file containing scenarii (python file): rain_file = os.path.join(dir_setup, 'rain_eq_atl.pyo') # ------ Threshold to flag data: rain_threshold = 0.15 wet_tropo = False # -----------------------# # L2C computation # -----------------------# # config name for L2d: config_l2c = '' # Length resolution to select neighbors (in km): resol = 40 # Grid resolution for l2c (alongtrack, acrosstrack) grid (in km): posting = 5 # Remove noisy data around nadir (in km): ac_threshold = 20 # List of variables to be interpolated on the swath: list_input_var_l2c = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]} # -----------------------# # L2D computation # -----------------------# # config name for L2d: config_l2d = '' # Length resolution to select neighbors (multiplication factor): resol_spatial_l2d = 1 # Temporal resolution to select neighbors (multiplication factor): resol_temporal_l2d = 1 # Grid resolution for l2d (lat, lon) grid (in degrees): posting_l2d = (0.1, 0.1) # Time domain: (start_time, end_time, dtime) in days: time_domain = (7, 23, 1) # Spatial domain (lon_min, lon_max, lat_min, lat_max): spatial_domain = [0, 360, -90, 90] # List of variables to be interpolated on the grid: list_input_var_l2d = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]}
7,726
36.692683
80
py
skimulator
skimulator-master/example/params_example_6beams.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") # ------ Name of the configuration (to build output files names) #config="WW3_EQ_metop_2018_6a" # 6 beams, 60 azimuths, 512 pulses and cycle length of 37/2 ms #config="WW3_EQ_metop_2019_6a" config="WW3_EQ_metop_2019_6b" config="WW3_EQ_metop_2019_6c" # 6 beams, ?? azimuths, 1024 pulses and cycle length of 37 ms #config = "WW3_EQ_metop_2019_6b" # ------ Directory that contains orbit file: dir_setup = os.path.join(home, 'skimulator', 'data') # ------ Directory that contains your own inputs: indatadir = os.path.join(home, 'skimulator', 'example', 'input_fields') # ------ Directory that contains your outputs: outdatadir = os.path.join(home, 'skimulator', 'example', 'skim_output') # ------ Orbit file: #filesat = os.path.join(dir_setup,'orbs1a.txt') filesat = os.path.join(dir_setup,'orbmetop_skim.txt') # ------ Number of days in orbit (optional if specified in orbit file) satcycle = 29 # ------ Satellite elevation (optional if specified in orbit file) sat_elev = 817 * 10**3 # ------ Order of columns (lon, lat, time) in orbit file # (default is (0, 1, 2) with order_orbit_col = None) order_orbit_col = None # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Number of processor for parallelisation proc_number = 1 # ------ Deactivate printing of progress bar to avoid huge log progress_bar = True # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) # ------ Force the computation of the satellite grid: makesgrid = True # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = None # [329.,347., -8.,8.] #------- Rotation speed of the antenna (in tr/min) if '2019_6a' in config: rotation_speed = 6.0 elif '2019_6b' in config: rotation_speed = 6 elif '2019_6c' in config: rotation_speed = 6 else: rotation_speed = 9.26 # ------ Cycle duration cycle = 0.0368 #------- List of position of beams: if '2019_6a' in config: list_pos = (0, 90 * math.pi/180., 180*math.pi/180, 270*math.pi/180, 0) elif '2019_6b' in config: list_pos = (0, 120 * math.pi/180, 240 * math.pi/180, 0, 180 * math.pi / 180) elif '2019_6c' in config: list_pos = (0, 120 * math.pi/180, 240 * math.pi/180, 0, 180 * math.pi / 180) else: list_pos = (0, 120*math.pi/180., 240*math.pi/180., 0, math.pi) #------- List of angle of beams in degrees: if '2019_6a' in config: list_angle = (12, 12, 12, 12, 6) else: list_angle = (12, 12, 12, 6, 6) #------- List of timeshift as regard to nadir for 12 degree beams: if '2019_6a' in config: list_shift = (1, 2, 5, 3, 4) elif '2019_6b' in config: list_shift = (1, 5, 2, 4, 3) elif '2019_6c' in config: list_shift = (1, 4, 2, 5, 3) else: list_shift = (5, 2, 3, 1, 4) # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, 'list_of_file.txt') # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other option is WW3) model = 'WW3' # ------ First time of the model first_time = '2011-11-15T00:00:00Z' # ------ Grid file name file_grid_model = (os.path.join(indatadir, 'ww3.20111115_cur.nc'),) # ------ Specify if there is a ice mask for high latitudes # (if true, mask is recomputed at each cycle) ice_mask = False # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify list of variable: list_input_var = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0], 'uuss': ['uuss', 'uss', 0], 'vuss': ['vuss', 'uss', 0], 'ice': ['ice', 'ice', 0], 'mssd': ['mssd', 'msd', 0], 'mssx': ['mssu', 'mss', 0], 'mssy':['mssc', 'mss', 0], 'ssh': ['wlv', 'wlv', 0], 'hs':['hs', 'hs', 0], 'uwnd': ['uwnd', 'wnd', 0], 'vwnd': ['vwnd', 'wnd', 0]} # ------ Specify longitude variable: lon = ('longitude',) # ------ Specify latitude variable: lat = ('latitude',) # ------ Specify number of time in file: dim_time = 24 # ------ Time step between two model outputs (in days): timestep = 1/24. # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 35*24 # ------ Not a number value: model_nan = -32767. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'linear' # -----------------------# # SKIM error parameters # -----------------------# # List of errors to compute noise = ['Altimeter', 'Instrument', 'Attitude', 'DSigma', 'Rain', 'WaveDoppler'] # Number of seed for automatic computation nseed = 0 # Length of repeat for altimetric noise len_repeat = 20000 # Todo compute automatically with cycle, number of beams and satellite velocity delta_al = 2 yaw_file = os.path.join(dir_setup, 'sample_req1.nc') # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Choice of instrument configuration instr_configuration = 'A' # ------- Coefficient SNR to retrieve instrumental noise from sigma, # Recommanded value for 1024 pulses: 3e-2, for 512 pulses: 3sqrt(2)e-3 snr_coeff = 1.4142*6e-3 # ------- File which provide the AOCS error: yaw_file = os.path.join(dir_setup, 'sample_req1.nc') ## -- Geophysical error ## ---------------------- # ------ Consider ice in sigma0 computation ice = True # ------ Rain file containing scenarii (python file): rain_file = os.path.join(dir_setup, 'rain_eq_atl.pyo') # ------ Threshold to flag data: rain_threshold = 0.15 # -----------------------# # L2C computation # -----------------------# # config name for L2d: config_l2c = '' # Length resolution to select neighbors (in km): resol = 40 # Grid resolution for l2c (alongtrack, acrosstrack) grid (in km): posting = 5 # Remove noisy data around nadir (in km): ac_threshold = 20 # List of variables to be interpolated on the swath: list_input_var_l2c = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]} # -----------------------# # L2D computation # -----------------------# # config name for L2d: config_l2d = '' # Length resolution to select neighbors (multiplication factor): resol_spatial_l2d = 1 # Temporal resolution to select neighbors (multiplication factor): resol_temporal_l2d = 1 # Grid resolution for l2d (lat, lon) grid (in degrees): posting_l2d = (0.1, 0.1) # Time domain: (start_time, end_time, dtime) in days: time_domain = (7, 23, 1) # Spatial domain (lon_min, lon_max, lat_min, lat_max): spatial_domain = [0, 360, -90, 90] # List of variables to be interpolated on the grid: list_input_var_l2d = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]}
8,115
36.229358
80
py
skimulator
skimulator-master/example/params_example_8beams_ogcm.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") # ------ Name of the configuration (to build output files names) # 8 beams, 45 azimuths, 1024 pulses and cycle length of 37 ms config = "OGCM_metop_2018_8a" # 8 beams, 72 azimuths, 0512 pulses and cycle length of 18 ms # config = "WW3_23W_metop_2018_8b" # 8 beams, 108 azimuths, 1024 pulses and cycle length of 37 ms # config = "WW3_23W_metop_2018_8c" # ------ Directory that contains orbit file: dir_setup = os.path.join(home, 'skimulator', 'data') # ------ Directory that contains your own inputs: indatadir = os.path.join(home, 'skimulator', 'example', 'input_fields') # ------ Directory that contains your outputs: outdatadir = os.path.join(home, 'skimulator', 'example', 'skim_output') # ------ Orbit file: #filesat = os.path.join(dir_setup,'orbs1a.txt') filesat = os.path.join(dir_setup,'orbmetop_skim.txt') # ------ Number of days in orbit (optional if specified in orbit file) satcycle = 29 # ------ Satellite elevation (optional if specified in orbit file) sat_elev = 817 * 10**3 # ------ Order of columns (lon, lat, time) in orbit file # (default is (0, 1, 2) with order_orbit_col = None) order_orbit_col = None # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Number of processor for parallelisation proc_number = 1 # ------ Deactivate printing of progress bar to avoid huge log progress_bar = True # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) # ------ Force the computation of the satellite grid: makesgrid = True # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = [300, 310, 30, 40] #------- Rotation speed of the antenna (in tr/min) if '2018_8a' in config: rotation_speed = 4.52 if '2018_8b' in config: rotation_speed = 5.66 if '2018_8c' in config: rotation_speed = 1.89 # Cycle duration cycle = 0.0368 if '2018_8b' in config: cycle = 0.0368 / 2. # cycle = 0.0368 / 2. # for 512 pulses configuration #------- List of position of beams: list_pos = (0, 72*math.pi/180., 144*math.pi/180., 216*math.pi / 180., 288*math.pi/180., 0, math.pi) #------- List of angle of beams in degrees: list_angle = (12, 12, 12, 12, 12, 6, 6) #------- List of timeshift as regard to nadir for 12 degree beams: list_shift = (1, 2, 4, 5, 7, 3, 6) # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, 'list_of_file.txt') # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other option is WW3) model = 'NETCDF_MODEL' # ------ First time of the model first_time = '2011-11-15T00:00:00Z' # ------ Grid file name file_grid_model = (os.path.join(indatadir, 'model.20111115_grid.nc'),) # ------ Specify if there is a ice mask for high latitudes # (if true, mask is recomputed at each cycle) ice_mask = False # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify list of variable: list_input_var = {'ucur': ['uo', '', 0], 'vcur': ['vo', '', 0], 'ssh': ['sla', '', 0] } # ------ Specify longitude variable: lon = ('longitude', ) # ------ Specify latitude variable: lat = ('latitude', ) # ------ Specify number of time in file: dim_time = 1 # ------ Time step between two model outputs (in days): timestep = 1 # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 20 # ------ Not a number value: model_nan = -32767. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'linear' # -----------------------# # SKIM error parameters # -----------------------# # List of errors to compute noise = ['Altimeter', 'Instrument', 'Attitude', 'DSigma', 'Rain', 'WaveDoppler'] # Number of seed for automatic computation nseed = 0 # Length of repeat for altimetric noise len_repeat = 20000 # Todo compute automatically with cycle, number of beams and satellite velocity delta_al = 2 yaw_file = os.path.join(dir_setup, 'sample_req1.nc') # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Choice of instrument configuration instr_configuration = 'A' # ------- Coefficient SNR to retrieve instrumental noise from sigma, # Recommanded value for 1024 pulses: 3e-2, for 512 pulses: 3sqrt(2)e-3 snr_coeff = 6e-3 if '2018_8b' in config: snr_coeff = 1.4142*6e-3 # ------- File which provide the AOCS error: yaw_file = os.path.join(dir_setup, 'sample_req1.nc') ## -- Geophysical error ## ---------------------- # ------ Consider ice in sigma0 computation ice = True # ------ Rain file containing scenarii (python file): rain_file = os.path.join(dir_setup, 'rain_eq_atl.pyo') # ------ Threshold to flag data: rain_threshold = 0.15 wet_tropo = False # -----------------------# # L2C computation # -----------------------# # config name for L2d: config_l2c = '' # Length resolution to select neighbors (in km): resol = 40 # Grid resolution for l2c (alongtrack, acrosstrack) grid (in km): posting = 5 # Remove noisy data around nadir (in km): ac_threshold = 20 # List of variables to be interpolated on the swath: list_input_var_l2c = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]} # -----------------------# # L2D computation # -----------------------# # config name for L2d: config_l2d = '' # Length resolution to select neighbors (multiplication factor): resol_spatial_l2d = 1 # Temporal resolution to select neighbors (multiplication factor): resol_temporal_l2d = 1 # Grid resolution for l2d (lat, lon) grid (in degrees): posting_l2d = (0.1, 0.1) # Time domain: (start_time, end_time, dtime) in days: time_domain = (7, 23, 1) # Spatial domain (lon_min, lon_max, lat_min, lat_max): spatial_domain = [0, 360, -90, 90] # List of variables to be interpolated on the grid: list_input_var_l2d = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]}
7,411
35.875622
80
py
skimulator
skimulator-master/example/params_stream.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") + '/src/' # ------ Name of the configuration (to build output files names) # 8 beams, 45 azimuths, 1024 pulses and cycle length of 37 ms rp = 35 config = f'WW3_STREAM_{rp}' # 8 beams, 72 azimuths, 0512 pulses and cycle length of 18 ms # config = "WW3_23W_metop_2018_8b" # 8 beams, 108 azimuths, 1024 pulses and cycle length of 37 ms # config = "WW3_23W_metop_2018_8c" # ------ Directory that contains orbit file: dir_setup = os.path.join(home, 'skimsimulator', 'data') dir_setup = '/mnt/data/data_skim/' # ------ Directory that contains your own inputs: indatadir = os.path.join(home, 'skimsimulator', 'example', 'input_fields') # ------ Directory that contains your outputs: outdatadir = os.path.join(home, 'skimsimulator', 'example', 'skim_output') outdatadir = '/mnt/data/skim_output' # ------ Orbit file: #filesat = os.path.join(dir_setup,'orbs1a.txt') filesat = os.path.join(dir_setup,'orbmetop_skim.txt') # ------ Number of days in orbit (optional if specified in orbit file) satcycle = 29 # ------ Satellite elevation (optional if specified in orbit file) sat_elev = 817 * 10**3 # ------ Order of columns (lon, lat, time) in orbit file # (default is (0, 1, 2) with order_orbit_col = None) order_orbit_col = None # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Number of processor for parallelisation proc_number = 1 # ------ Deactivate printing of progress bar to avoid huge log progress_bar = True # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) # ------ Force the computation of the satellite grid: makesgrid = True # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = [300, 359, 20, 50] #None #[329.,347., -8.,8.] #------- Rotation speed of the antenna (in tr/min) rotation_speed = rp # Cycle duration # PRF 12KHz, 64 pulses / cycle cycle = 64/12000 # cycle = 0.0368 / 2. # for 512 pulses configuration #------- List of position of beams: list_pos = (0,) #------- List of angle of beams in degrees: list_angle = (35,) #------- List of timeshift as regard to nadir for 12 degree beams: list_shift = (1,) # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, 'list_of_file.txt') # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other option is WW3) model = 'WW3' # ------ First time of the model first_time = '2011-11-15T00:00:00Z' # ------ Grid file name file_grid_model = (os.path.join(indatadir, 'ww3.20111115_hs.nc'), os.path.join(indatadir, 'ww3.20111115_cur12.nc'), os.path.join(indatadir, 'ww3.20111115_rain.nc')) # ------ Specify if there is a ice mask for high latitudes # (if true, mask is recomputed at each cycle) ice_mask = False # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify list of variable: list_input_var = {'ucur': ['ucur', 'cur12', 1], 'vcur': ['vcur', 'cur12', 1], 'uuss': ['uuss', 'uss', 0], 'vuss': ['vuss', 'uss', 0], 'ice': ['ice', 'ice', 0], 'mssd': ['mssd', 'msd', 0], 'mssx': ['mssu', 'mss', 0], 'mssy':['mssc', 'mss', 0], 'ssh': ['wlv', 'wlv', 0], 'hs':['hs', 'hs', 0], 'uwnd': ['uwnd', 'wnd', 0], 'vwnd': ['vwnd', 'wnd', 0]} # ------ Specify longitude variable: lon = ('longitude', 'longitude', 'lon') # ------ Specify latitude variable: lat = ('latitude', 'latitude', 'lat') # ------ Specify number of time in file: dim_time = 24 # ------ Time step between two model outputs (in days): timestep = 1/24. # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 35*24 # ------ Not a number value: model_nan = 0. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'linear' # -----------------------# # SKIM error parameters # -----------------------# # List of errors to compute noise = ['Altimeter', 'Instrument', 'Attitude', 'DSigma', 'Rain', 'WaveDoppler'] # Number of seed for automatic computation nseed = 0 # Length of repeat for altimetric noise len_repeat = 20000 # Todo compute automatically with cycle, number of beams and satellite velocity delta_al = 2 yaw_file = os.path.join(dir_setup, 'sample_req1.nc') # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Choice of instrument configuration instr_configuration = 'A' # ------- Coefficient SNR to retrieve instrumental noise from sigma, # Recommanded value for 1024 pulses: 3e-2, for 512 pulses: 3sqrt(2)e-3 snr_coeff = 6e-3 if '2018_8b' in config: snr_coeff = 1.4142*6e-3 # ------- File which provide the AOCS error: yaw_file = os.path.join(dir_setup, 'sample_req1.nc') ## -- Geophysical error ## ---------------------- # ------ Consider ice in sigma0 computation ice = True # ------ Rain file containing scenarii (python file): rain_file = None #os.path.join(dir_setup, 'rain_eq_atl.pyo') # ------ Threshold to flag data: rain_threshold = 0.15 wet_tropo = False # -----------------------# # L2C computation # -----------------------# # config name for L2d: config_l2c = 'l40' # Length resolution to select neighbors (in km): resol = 40 # Grid resolution for l2c (alongtrack, acrosstrack) grid (in km): posting = 5 # Remove noisy data around nadir (in km): ac_threshold = 20 # List of variables to be interpolated on the swath: list_input_var_l2c = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]} # -----------------------# # L2D computation # -----------------------# # config name for L2d: config_l2d = '' # Length resolution to select neighbors (multiplication factor): resol_spatial_l2d = 1 # Temporal resolution to select neighbors (multiplication factor): resol_temporal_l2d = 1 # Grid resolution for l2d (lat, lon) grid (in degrees): posting_l2d = (0.1, 0.1) # Time domain: (start_time, end_time, dtime) in days: time_domain = (7, 23, 1) # Spatial domain (lon_min, lon_max, lat_min, lat_max): spatial_domain = [0, 360, -90, 90] # List of variables to be interpolated on the grid: list_input_var_l2d = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0]}
7,742
37.142857
80
py
skimulator
skimulator-master/skimulator/const.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ '''Constants that are defined for SKIM simulator. \n It contains Earth parameters as well as SKIM instrument and satellite caracteristics. ''' # ################################ # # EARTH CONSTANTS ## # ################################ # - Earth radius (m) Rearth = 6378. * 10**3 # - Convert degree to km deg2km = 111.11 # - Seconds in a day secinday = 86400. # - Light speed (m/s) C = 2.998*10**8 # ################################### # # SKIM INSTRUMENT CARACTERISTICS ## # ################################### # - Satellite elevation (m) sat_elev = 817*10**3 # - Baseline (m) B = 10 # (in Hz) Fka = 35.75 * 10**9 # Satellite cycle (S1) in days satcycle = 29. # Satellite velocity (m/s) vsat = 6657. # ################################### # # OTHER PARAMETERS ## # ################################### # - Radius to interpolate locally model data on the swath (in km) # data are selected every xal_step points and on a radius of radius_interp radius_interp = 100. # - Sampling to interpolate locally model data on the swath (in km) # Data are selected every xal_step points and on a radius of radius_interp xal_step = 20.
1,840
29.683333
75
py
skimulator
skimulator-master/skimulator/mod_parallel.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. @author <sylvain.herledan@oceandatalab.com> @date 2018-12-26 """ import os import sys import math import time import logging import traceback import collections import multiprocessing logger = logging.getLogger(__name__) #logger = multiprocessing.log_to_stderr() #logger.setLevel(logging.DEBUG) class DyingOnError(Exception): """""" pass class MultiprocessingError(Exception): """""" def __init__(self, exc): """""" self.exc = exc class JobsManager(): """""" def __init__(self, pool_size, status_updater, exc_fmt, err_fmt, init_method=None, init_args=None): """""" self._pool_size = pool_size self.manager = multiprocessing.Manager() self.msg_queue = self.manager.Queue() self.res_queue = self.manager.Queue() self.errors_queue = self.manager.Queue() ctx = multiprocessing.get_context('spawn') if init_method is None: self.pool = ctx.Pool(pool_size) else: self.pool = ctx.Pool(pool_size, initializer=init_method, initargs=init_args) #self.format_exception = (exc_fmt,) self.format_exception = None if exc_fmt is not None: self.format_exception = staticmethod(exc_fmt) self.update_status = None if status_updater is not None: self.update_status = staticmethod(status_updater) self.format_error = None if err_fmt is not None: self.format_error = staticmethod(err_fmt) def show_errors(self): """""" if self.format_error is None: return while not self.errors_queue.empty(): (pid, job_id, extra, exc) = self.errors_queue.get() error_str = self.format_error.__func__(pid, job_id, extra, exc) logger.error(error_str) logger.error(' {}'.format(' \n'.join(exc))) def _error_callback(self, exc): """""" logger.error(exc) raise MultiprocessingError(exc) def handle_message(self, status, msg, die_on_error, progress_bar): """die_on_error""" _ok = (msg[3] is None) if (progress_bar is True) and (self.update_status is not None): _ok = self.update_status.__func__(status, msg) #_ok = mod_tools.update_progress_multiproc(status, msg) if (_ok is False) and (die_on_error is True): # Kill all workers, show error and exit with status 1 self.pool.terminate() self.show_errors() raise DyingOnError return _ok def submit_jobs(self, operation, jobs, die_on_error, progress_bar, delay=0.5, results=None): """""" results_callback = None results_args = None if results is not None: if isinstance(results, collections.Iterable) and 1 < len(results): results_callback, results_args = results[0], results[1:] else: results_callback = results for j in jobs: j.append(self.errors_queue) j.append(self.msg_queue) j.append(self.format_exception.__func__) j.append(operation) # Distribute jobs between workers chunk_size = int(math.ceil(len(jobs) / self._pool_size)) status = {} for n, w in enumerate(self.pool._pool): status[w.pid] = {'done': 0, 'total': 0, 'jobs': None, 'extra': ''} proc_jobs = jobs[n::self._pool_size] status[w.pid]['jobs'] = [j[0] for j in proc_jobs] status[w.pid]['total'] = len(proc_jobs) # Make some room for the progress bars if progress_bar is True: sys.stdout.write('\n' * self._pool_size) sys.stdout.flush() # Start jobs processing if results is True: res = [] tasks = self.pool.map_async(_operation_wrapper, jobs, chunksize=chunk_size, error_callback=self._error_callback, callback=res.append) else: tasks = self.pool.map_async(_operation_wrapper, jobs, chunksize=chunk_size, error_callback=self._error_callback) # Wait until all jobs have been executed ok = True while not tasks.ready(): # Handle progress and error messages if not self.msg_queue.empty(): msg = self.msg_queue.get() _ok = self.handle_message(status, msg, die_on_error, progress_bar) ok = ok and _ok # Handle results from processed jobs on the fly if (results_callback is not None) and (not self.res_queue.empty()): job_res = self.res_queue.get() if results_args is not None: results_callback(job_res, *results_args) else: results_callback(job_res) time.sleep(delay) # Make sure all messages have been processed while not self.msg_queue.empty(): msg = self.msg_queue.get() _ok = self.handle_message(status, msg, die_on_error, progress_bar) ok = ok and _ok # Flush stdout buffer to avoid partial output issues sys.stdout.flush() # Make sure all results have been processed while ((results is not None) and (not self.res_queue.empty())): job_res = self.res_queue.get() if results_args is not None: results_callback(job_res, *results_args) else: results_callback(job_res) # Wait for workers to release resources and synchronize with the main # process self.pool.close() self.pool.join() #""" return ok def _operation_wrapper(*args, **kwargs): """""" _args = args[0] operation = _args.pop() format_exc = _args.pop() msg_queue = _args.pop() errors_queue = _args.pop() try: job_id = _args[0] res = operation(msg_queue, *_args, **kwargs) except: # Error sink exc = sys.exc_info() if format_exc is None: error_msg = traceback.format_exception(exc[0], exc[1], exc[2]) else: error_msg = format_exc(exc) # Pass the error message to both the messages queue and the # errors queue msg_queue.put((os.getpid(), job_id, -1, error_msg)) errors_queue.put((os.getpid(), job_id, -1, error_msg)) return False if res is None: return True else: return res
7,515
31.257511
79
py
skimulator
skimulator-master/skimulator/fitspline2d.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import numpy from . import spline import scipy.sparse.linalg as linalg class fitspline2d: def __init__(self, nx, ny, nxspline, nyspline, xcircle=True, ycircle=True): self.xcircle = xcircle self.ycircle = ycircle self.nx = nx self.ny = ny self.nxspline = nxspline self.nyspline = nyspline x = numpy.arange(nx) + 0.5 y = numpy.arange(ny) + 0.5 if self.xcircle == False: self.spx = spline.spline(3, nxspline - 2) else: self.spx = spline.spline_circular(3, nxspline) if self.ycircle == False: self.spy = spline.spline(3, nyspline - 2) else: self.spy = spline.spline_circular(3,nyspline) self.spx.init_uniform(0, nx) self.spy.init_uniform(0, ny) self.vx, self.ix, self.jx = self.spx.bspline_value(x) self.vy, self.iy, self.jy = self.spy.bspline_value(y) self.nspline=self.nxspline*self.nyspline def init_fit(self, doinvert=True): mat=numpy.zeros([self.nspline, self.nspline]) for ii in range(self.nx): # print(ii) for jj in range(self.ny): for jjx in range(4): for jjy in range(4): for kkx in range(4): for kky in range(4): ixj = numpy.mod(self.ix[ii] + jjx, self.nxspline) ixk = numpy.mod(self.ix[ii] + kkx, self.nxspline) iyj = numpy.mod(self.iy[jj] + jjy, self.nyspline) iyk = numpy.mod(self.iy[jj] + kky, self.nyspline) ind0 = ixj + self.nxspline * iyj ind1 = ixk + self.nxspline * iyk mat[ind0, ind1] += (self.vx[kkx,ii] * self.vy[kky,jj] * self.vx[jjx,ii] * self.vy[jjy,jj]) if self.xcircle == False: for ii in range(self.nyspline): mat[ii * self.nxspline, ii * self.nxspline] += 1 for ii in range(self.nyspline): mat[self.nxspline-1+ii*self.nxspline, ii*self.nxspline] += 1 if self.ycircle == False: for ii in range(self.nxspline): mat[ii, ii] += 1 for ii in range(self.nxspline): ind0 = ii+(self.nyspline-1)*self.nxspline ind1 = (self.nyspline-1)*self.nxspline mat[ind0, ind1] += 1 if doinvert is True: self.imat=numpy.linalg.pinv(mat) self.inv = True else: self.imat = mat self.inv = False def fit(self,data): nspline=self.nspline vec=numpy.zeros([nspline]) for ii in range(self.nx): for jj in range(self.ny): for jjx in range(4): for jjy in range(4): ixj = numpy.mod(self.ix[ii] + jjx, self.nxspline) iyj = numpy.mod(self.iy[jj] + jjy, self.nyspline) vec[ixj + self.nxspline * iyj] += self.vx[jjx, ii]*self.vy[jjy, jj] * data[ii, jj] if self.xcircle == False: for ii in range(self.nyspline): vec[ii*self.nxspline]+=data[0,ii] for ii in range(self.nyspline): vec[self.nxspline-1+ii*self.nxspline]+=data[self.nx-1,ii] if self.ycircle == False: for ii in range(self.nxspline): vec[ii]+=data[ii,0] for ii in range(self.nxspline): vec[ii+(self.nyspline-1)*self.nxspline]+=data[ii,self.ny-1] if self.inv == False: self.wres,infoinv=linalg.cg(self.imat,vec) else: self.wres=numpy.dot(self.imat, vec) def getparam(self): return self.wres def setparam(self,wres): self.wres=wres def transform(self, x, y): n = x.shape[0] res = numpy.zeros([n]) vx, ix, jx = self.spx.bspline_value(x) vy, iy, jy = self.spy.bspline_value(y) self.wres=self.wres.reshape(self.nyspline, self.nxspline) for jj in range(4): for ii in range(4): iyj = numpy.mod(iy + jj, self.nyspline) ixi = numpy.mod(ix + ii, self.nxspline) res[:] += vy[jj, :] * vx[ii, :] * self.wres[iyj, ixi] return res class ted_tas(): def __init__(self, wres, nxspline, nyspline, thedeg): self.mysp2={} self.n_time = 1200 self.n_az = 120 for i in range(3): self.mysp2[i]=fitspline2d(self.n_time, self.n_az, nxspline, nyspline, xcircle=True, ycircle=True) self.mysp2[i].setparam(wres[i,:]) def transform(self, t_orbit, az, t_year): ''' #==================================================================== # t_orbit : orbital time inside [0,1]; 0 begining of the orbit, 1 end of the orbit (if t_orbit <0 or t_orbit>1 then BUG! # az : azimuth in degree # t_year : date in year inside [0,1]; 0 begining of the year, 1 end of the year ''' # Number of points for the time dimension n_time = self.n_time az = numpy.mod(az, 360) # Discretisation step for azimuth delta_az = 360 / self.n_az fit_1 = self.mysp2[0].transform(t_orbit * n_time, az / delta_az) cos_year = numpy.cos(t_year * 2 * numpy.pi) t_orbit = numpy.fmod((t_orbit * n_time), n_time) azd = numpy.fmod(az, 360) / delta_az fit_t2 = self.mysp2[1].transform(t_orbit, azd) fit_t3 = self.mysp2[2].transform(t_orbit, azd) fit_2 = fit_t2 * numpy.cos(t_year * 2 * numpy.pi) fit_3 = fit_t3 * numpy.sin(t_year * 2 * numpy.pi) return (fit_1 + fit_2 + fit_3)
6,754
37.6
106
py
skimulator
skimulator-master/skimulator/build_swath.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import os import sys import numpy import math from scipy import interpolate import skimulator.mod_tools as mod_tools import skimulator.const as const import skimulator.rw_data as rw_data import skimulator.mod_parallel import time import logging import traceback # Define logger level for debug purposes logger = logging.getLogger(__name__) def makeorbit(modelbox, p, orbitfile='orbit_292.txt', filealtimeter=None): '''Computes the swath of SKIM satellites on a subdomain. The path of the satellite is given by the orbit file and the subdomain corresponds to the one in the model. Note that a subdomain can be manually added in the parameters file. \n Inputs are satellite orbit (p.filesat), subdomain (modelbox), List of postion of six degree beams, list of position of twelve degree beams, rotation speed. \n Outputs are netcdf files containing SKIM position (lon, lat number of days in a cycle, distance crossed in a cycle, time) ''' # - Load SKIM orbit ground track logger.info('Load data from orbit file') if p.order_orbit_col is None: volon, volat, votime = numpy.loadtxt(orbitfile, usecols=(0, 1, 2), comments='#', unpack=True) else: ncols = p.order_orbit_col volon, volat, votime = numpy.loadtxt(orbitfile, usecols=ncols, comments='#', unpack=True) dic_sat = {} with open(orbitfile, 'r') as fh: for i, line in enumerate(fh): if line.strip().startswith('#'): key, value = line.strip().split('=') dic_sat[key[1:].strip()] = float(value.strip()) else: break if 'cycle' in dic_sat.keys() and 'elevation' in dic_sat.keys(): p.satcycle = dic_sat['cycle'] p.sat_elev = dic_sat['elevation'] # - If orbit is at low resolution, interpolate at cycle (s) resolution cycle = p.cycle votime = votime * const.secinday # macrocycle is fixed or depend on number of beam logger.info('Interpolate orbit at {} seconds'.format(cycle)) # votime2 = votime + 0 # - If orbit is at low resolution, interpolate at 0.5 s resolution hr_step = 0.5 if numpy.mean(votime[1:] - votime[:-1]) > hr_step: x, y, z = mod_tools.spher2cart(volon, volat) time_hr = numpy.arange(0., votime[-1], hr_step) # Parallelisation of this loop? # jobs = [] # jobs.append([votime, time_hr, x]) # jobs.append([votime, time_hr, y]) # jobs.append([votime, time_hr, z]) # worker f = interpolate.interp1d(votime, x) x_hr = f(time_hr) f = interpolate.interp1d(votime, y) y_hr = f(time_hr) f = interpolate.interp1d(votime, z) z_hr = f(time_hr) lon_hr = numpy.zeros(len(x_hr)) + numpy.nan lat_hr = numpy.zeros(len(x_hr)) + numpy.nan lon_hr, lat_hr = mod_tools.cart2sphervect(x_hr, y_hr, z_hr) # Cut orbit if more than an orbit cycle if p.satcycle is None: p.satcycle = const.satcycle ind = numpy.where((time_hr < p.satcycle * const.secinday)) volon = lon_hr[ind] volat = lat_hr[ind] votime = time_hr[ind] # - Get number of points in orbit nop = numpy.shape(votime)[0] # - Get cycle period. tcycle = votime[nop-1] + votime[1] - votime[0] # shift time if the user needs to shift the time of the orbit if p.shift_time is not None: shift_time = p.shift_time * const.secinday shift_index = numpy.where(votime >= shift_time)[0] volon = numpy.hstack([volon[shift_index[0]:], volon[:shift_index[0]]]) volat = numpy.hstack([volat[shift_index[0]:], volat[:shift_index[0]]]) # shift lon if the user needs to shift the localisation of the orbit if p.shift_lon is not None: volon = volon + p.shift_lon volon = (volon + 360) % 360 # - Rearrange orbit starting from pass 1 # Detect the beginning of pass 1 in orbit txt file. By definition, it is # the first passage at southernmost latitude. dlat = numpy.roll(volat, 1) - volat ind = numpy.where((dlat < 0) & (numpy.roll(dlat, 1) >= 0)) # Shift coordinates, so that the first point of the orbit is the beginning # of pass 1 decal = ind[0][-1] # timeshift = votime[-1] - votime[decal] volon = numpy.hstack([volon[decal:], volon[:decal]]) volat = numpy.hstack([volat[decal:], volat[:decal]]) votime = numpy.hstack([votime[decal:], votime[:decal]]) votime = (votime - votime[0]) % tcycle if votime[numpy.where(votime < 0)]: logger.warn('WARNING: there are negative times in your orbit') del ind # Compute the initial time of each pass dlat = numpy.roll(volat, 1) - volat ind = numpy.where(((dlat < 0) & (numpy.roll(dlat, 1) >= 0)) | ((dlat > 0) & (numpy.roll(dlat, 1) <= 0))) # index=numpy.hstack([0,ind[0]-1]) index = ind[0] passtime = votime[index] # Times of pass # - Extract region matnpbox = numpy.zeros((nop)) if modelbox[0] > modelbox[1]: matnpbox[numpy.where((((modelbox[0] - 1) <= volon) | (volon <= (modelbox[1] + 1))) & ((modelbox[2] - 1) <= volat) & ((modelbox[3] + 1) >= volat))] = 1 else: matnpbox[numpy.where(((modelbox[0] - 1) <= volon) & (volon <= (modelbox[1] + 1)) & ((modelbox[2] - 1) <= volat) & ((modelbox[3] + 1) >= volat))] = 1 norp = int(numpy.sum(matnpbox)) # Initialize total distance travelled by the satellite since the first # point of the cycle in the subdomain at low (orbital file) resolution x_al_lr = numpy.zeros((norp)) lon_lr = numpy.zeros((norp)) lat_lr = numpy.zeros((norp)) stime_lr = numpy.zeros((norp)) # Initialize vector with accumulated distance travelled by the satellite indp = 0 distance = numpy.zeros((nop)) # Compute and store distances and coordinates that are in the defined # subdomain logger.info('Compute SKIM nadir coordinate in the new domain') for i in range(0, nop - 1): if p.progress_bar is True: mod_tools.update_progress(float(i) / float(nop-1), None, None) if abs(volon[i + 1] - volon[i]) > 1: if volon[i + 1] > 180.: volon[i + 1] = volon[i + 1] - 360 if volon[i] > 180.: volon[i] = volon[i] - 360 distance[i+1] = (distance[i] + numpy.sqrt(((volon[i+1]-volon[i]) * const.deg2km*numpy.cos(volat[i+1] * 2*math.pi/360.))**2 + ((volat[i+1] - volat[i]) * const.deg2km)**2)) # numpy.sum(dl[:i]) volon[i + 1] = (volon[i + 1] + 360) % 360 if matnpbox[i]: x_al_lr[indp] = distance[i] lon_lr[indp] = (volon[i] + 360) % 360 lat_lr[indp] = volat[i] stime_lr[indp] = votime[i] indp += 1 # - Interpolate orbit at cycle resolution (default is cycle=0.0096) # Detect gap in time in stime (to detect step in x_al, lon and lat) dstime = stime_lr[:] - numpy.roll(stime_lr[:], 1) ind = numpy.where(dstime > 3*(votime[1] - votime[0])) index = numpy.hstack([0, ind[0]]) nindex = numpy.shape(index)[0] # Initialize along track distance, time and coordinates at cycle # resolution if nindex > 1: dgap = numpy.zeros((nindex)) for i in range(1, nindex): dgap[i] = stime_lr[index[i]] - stime_lr[max(index[i] - 1, 0)] Ninterp = int(((stime_lr[-1] - stime_lr[0] - sum(dgap)) / float(p.cycle)) + 1) x_al = numpy.zeros((Ninterp)) stime = numpy.zeros((Ninterp)) lon = numpy.zeros((Ninterp)) lat = numpy.zeros((Ninterp)) imin = 0 imax = 0 substract_point = 0 for i in range(0, nindex - 1): imax = imin + int((stime_lr[index[i+1]-1] - stime_lr[index[i]]) / float(p.cycle)) + 1 if imax <= (imin + 1): x_al[imin] = x_al_lr[index[i]] stime[imin] = stime_lr[index[i]] lon[imin] = lon_lr[index[i]] lat[imin] = lat_lr[index[i]] else: slicei = slice(index[i], index[i + 1]) nofp = (stime_lr[index[i+1] - 1]-stime_lr[index[i]]) / p.cycle if (nofp).is_integer(): imax = imax - 1 substract_point += 1 stime[imin: imax] = numpy.arange(stime_lr[index[i]], stime_lr[index[i+1] - 1], p.cycle) x_al[imin: imax] = numpy.interp(stime[imin: imax], stime_lr[slicei], x_al_lr[slicei]) loncirc = numpy.rad2deg(numpy.unwrap(numpy.deg2rad( lon_lr[slicei]))) lon[imin: imax] = numpy.interp(stime[imin: imax], stime_lr[slicei], loncirc) lat[imin: imax] = numpy.interp(stime[imin: imax], stime_lr[slicei], lat_lr[slicei]) imin = imax if substract_point > 0: substract_point = 1 _tmp = numpy.arange(stime_lr[index[-1]], stime_lr[index[-1]] + (Ninterp - imin)*p.cycle - p.cycle * substract_point, p.cycle) if len(_tmp) > len(stime[imin:]): _tmp2 = _tmp[:-1] elif len(_tmp) < len(stime[imin:]): _tmp2 = numpy.arange(stime_lr[index[-1]], stime_lr[index[-1]] + (Ninterp - imin)*p.cycle, p.cycle) #logger.error('damnit, there is a bug, contact me') else: _tmp2 = _tmp stime[imin:] = _tmp2 #numpy.arange(stime_lr[index[-1]], stime_lr[index[-1]] # + (Ninterp - imin)*p.cycle # - p.cycle * substract_point, p.cycle) x_al[imin:] = numpy.interp(stime[imin:], stime_lr[index[-1]:], x_al_lr[index[-1]:]) loncirc = numpy.rad2deg(numpy.unwrap(numpy.deg2rad( lon_lr[index[-1]:]))) lon[imin:] = numpy.interp(stime[imin:], stime_lr[index[-1]:], loncirc) lat[imin:] = numpy.interp(stime[imin:], stime_lr[index[-1]:], lat_lr[index[-1]:]) else: Ninterp = int(((stime_lr[-2] - stime_lr[0]) / float(p.cycle)) + 1) stime = numpy.arange(stime_lr[0], stime_lr[-2], p.cycle) x_al = numpy.interp(stime, stime_lr[:-1], x_al_lr[:-1]) loncirc = numpy.rad2deg(numpy.unwrap(numpy.deg2rad(lon_lr[:-1]))) lon = numpy.interp(stime, stime_lr[:-1], loncirc) lat = numpy.interp(stime, stime_lr[:-1], lat_lr[:-1]) #x_al = numpy.arange(x_al_lr[0], x_al_lr[-2], p.delta_al) #stime = numpy.interp(x_al, x_al_lr[:-1], stime_lr[:-1]) #lon = numpy.interp(x_al, x_al_lr[:-1], loncirc) #lat = numpy.interp(x_al, x_al_lr[:-1], lat_lr[:-1]) lon = lon % 360 # Save orbit data in Sat_SKIM object orb = rw_data.Sat_SKIM(ifile='{}.nc'.format(os.path.basename(orbitfile))) orb.x_al = x_al orb.time = stime orb.lon = lon orb.lat = lat orb.cycle = tcycle orb.al_cycle = distance[-1] orb.passtime = numpy.sort(passtime) orb.timeshift = p.timeshift return orb def orbit2swath(modelbox, p, orb, die_on_error): '''Computes the swath of SKIM satellites on a subdomain from an orbit. The path of the satellite is given by the orbit file and the subdomain corresponds to the one in the model. Note that a subdomain can be manually added in the parameters file. \n Inputs are satellite orbit (p.filesat), subdomain (modelbox), Swath parameters (half gap distance p.halfgap, half swath distance p.halfswath, along track resolution p.delta_al, across track resolution p.delta_ac). \n Outputs are netcdf files containing SKIM grid (along track distance x_al, radial angle, longitude lon and latitude lat, number of days in a cycle cycle, distance crossed in a cycle cycle_al, time''' ''' Compute orbit from Swath ''' # - Load altimeter orbit x_al = orb.x_al stime = orb.time lon = orb.lon lat = orb.lat tcycle = orb.cycle al_cycle = orb.al_cycle passtime = orb.passtime # - Computation of SKIM grid and storage by passes logger.info('\n Compute SKIM grid') # Detect first pass that is in the subdomain ipass0 = 0 # Check that list of position, shift and angle have the same # dimension if len(p.list_pos) != len(p.list_shift) or \ len(p.list_pos) != len(p.list_angle) or \ len(p.list_angle) != len(p.list_shift): logger.error('Wrong length in list_pos, list_shift' 'or list_angle') sys.exit(1) # Loop on all passes after the first pass detected (note that ipass is # actually stored as ipass + 1 to have the first pass at 1 and ascending jobs = [] p2 = mod_tools.todict(p) for ipass in range(ipass0, numpy.shape(passtime)[0]): jobs.append([ipass, p2, passtime, stime, x_al, tcycle, al_cycle, lon, lat, orb.timeshift]) ok = make_skim_grid(p.proc_count, jobs, die_on_error, p.progress_bar) if p.progress_bar is True: mod_tools.update_progress(1, 'All swaths have been processed', ' ') else: logger.info('All swaths have been processed') return ok def make_skim_grid(_proc_count, jobs, die_on_error, progress_bar): """ Compute SWOT grids for every pass in the domain""" # - Set up parallelisation parameters proc_count = min(len(jobs), _proc_count) status_updater = mod_tools.update_progress_multiproc jobs_manager = skimulator.mod_parallel.JobsManager(proc_count, status_updater, exc_formatter, err_formatter) ok = jobs_manager.submit_jobs(worker_method_grid, jobs, die_on_error, progress_bar) if not ok: # Display errors once the processing is done jobs_manager.show_errors() return ok def exc_formatter(exc): """Format exception returned by sys.exc_info() as a string so that it can be serialized by pickle and stored in the JobsManager.""" error_msg = traceback.format_exception(exc[0], exc[1], exc[2]) return error_msg def err_formatter(pid, ipass, cycle, exc): """Transform errors stored by the JobsManager into readable messages.""" msg = None if cycle < 0: msg = '/!\ Error occurred while processing pass {}'.format(ipass) else: _msg = '/!\ Error occurred while processing cycle {} on pass {}' msg = _msg.format(cycle, ipass) return msg def worker_method_grid(*args, **kwargs): msg_queue, ipass, p2 = args[:3] passtime, stime, x_al, tcycle, al_cycle, lon, lat, timeshift = args[3:] p = mod_tools.fromdict(p2) # Compute rotating beams omega = p.rotation_speed * 2 * math.pi / 60. # Number of beam to lighten: nbeam = len(p.list_shift) + 1 sat_elev= p.sat_elev if sat_elev is None: sat_elev = const.sat_elev # Detect indices corresponding to the pass if ipass == numpy.shape(passtime)[0]-1: ind = numpy.where((stime >= passtime[ipass]))[0] else: ind = numpy.where((stime >= passtime[ipass]) & (stime < passtime[ipass+1]))[0] nind = numpy.shape(ind)[0] # Compute swath grid if pass is in the subdomain if nind > 5: # Initialize SKIM grid, grid variables filesgrid = '{}_p{:03d}.nc'.format(p.filesgrid, ipass + 1) sgrid = rw_data.Sat_SKIM(ifile=filesgrid) sgrid.x_al = x_al[ind] x_al_nad = x_al[ind] x_al_nad = x_al_nad[0::nbeam] sgrid.cycle = tcycle sgrid.al_cycle = al_cycle sgrid.ipass = ipass + 1 # Compute nadir coordinate and initialize angles lonnad = (lon[ind] + 360) % 360 latnad = + lat[ind] timenad = + stime[ind] lon_beam = [lonnad[0::nbeam]] lat_beam = [latnad[0::nbeam]] time_beam = [timenad[0::nbeam]] # n_nad = numpy.shape(lon_beam[0])[0] angle_beam = [numpy.zeros(numpy.shape(lon_beam[0]))] radial_angle_tot = [numpy.zeros(numpy.shape(lon_beam[0]))] xal_beam = [x_al_nad] xac_beam = [x_al_nad * 0] xal_beam_tot = [x_al_nad] # x_al_tmp = x_al_nad - x_al_nad[0] inclination_angle = numpy.zeros(numpy.shape(lonnad)) inclination_angle[1:] = numpy.arctan((latnad[1:] - latnad[:-1]) / numpy.cos(latnad[1:] * math.pi / 180.) / (lonnad[1:] - lonnad[:-1])) inclination_angle[0] = inclination_angle[1] # Loop on beam to construct cycloid for angle, shift, beam in zip(p.list_pos, p.list_shift, p.list_angle): # Angle projected on the earth rc = (const.Rearth * (beam * math.pi/180 - numpy.arcsin(const.Rearth * numpy.sin(math.pi - beam * math.pi/180) / (const.Rearth + sat_elev))) * 10**(-3)) timebeamshift = timenad[shift::nbeam] beam_angle = omega * timebeamshift + angle xal = -(rc * numpy.sin(beam_angle)) / const.deg2km xac = (rc * numpy.cos(beam_angle)) / const.deg2km sign_ac = 1 # Even pass: descending if ((ipass + 1) % 2 == 0): # inclination = -inclination_angle[shift::nbeam] + math.pi inclination = inclination_angle[shift::nbeam] inclination_save = inclination_angle[0::nbeam] radial_angle = -beam_angle + inclination - math.pi/2. sign_ac = -1 # Odd pass: ascending else: inclination = math.pi + inclination_angle[shift::nbeam] # inclination = + inclination_angle[shift::nbeam] inclination_save = inclination_angle[0::nbeam] #+ numpy.pi radial_angle = -beam_angle + inclination - math.pi / 2. lon_tmp = (lonnad[shift::nbeam] + (xal * numpy.cos(inclination) + xac * numpy.sin(inclination)) / numpy.cos(latnad[shift::nbeam] * math.pi / 180.)) lat_tmp = (latnad[shift::nbeam] + (xal * numpy.sin(inclination) - xac * numpy.cos(inclination))) lon_tmp = (lon_tmp + 360) % 360 # Concatenate list for each beam angle lon_beam.append(lon_tmp) lat_beam.append(lat_tmp) xal_beam.append(xal * const.deg2km * sign_ac) xac_beam.append(xac * const.deg2km * sign_ac) xal_beam_tot.append(sgrid.x_al[shift::nbeam]) time_beam.append(timebeamshift) angle_beam.append(beam_angle) radial_angle_tot.append(radial_angle) # Save Sgrid object sgrid.timeshift = timeshift sgrid.lon = lon_beam sgrid.lat = lat_beam sgrid.time = time_beam sgrid.x_al = xal_beam sgrid.x_al_tot = xal_beam_tot sgrid.x_ac = xac_beam sgrid.list_angle = p.list_angle sgrid.list_pos = p.list_pos sgrid.beam_angle = angle_beam sgrid.radial_angle = radial_angle_tot sgrid.angle = inclination_save # Remove grid file if it exists and save it if os.path.exists(filesgrid): os.remove(filesgrid) sgrid.write_swath(p) msg_queue.put((os.getpid(), ipass, None, None)) return None
21,162
42.455852
83
py
skimulator
skimulator-master/skimulator/run_simulator.py
'''Main program: Usage: run_simulator(file_param) \n If no param file is specified, the default one is exemple/params_exemple.txt \n In the first part of the program, model coordinates are read and the SKIM swath is computing accordingly. \n The SKIM grid parameters are saved in netcdf files, if you don't want to recompute them, set maksgrid (in params file) to False.\n In the second part of the program, errors are computed on SKIM grid for each pass, for each cycle. The error free velocity is the velocity interpolated from the model at each timestep. Note that there is no temporal interpolation between model files and thus if several files are used in the velocity interpolation, some discontinuities may be visible. \n OUTPUTS are netcdf files containing the requested errors, the error free radial velocity and the radial velocity with errors. There is one file every pass and every cycle. \n \n Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. ''' import os import numpy import glob import sys import time import datetime import logging import traceback import pkg_resources from . import build_swath from . import rw_data #import skimulator.build_error as build_error #import skimulator.build_error2 as build_error from .error import generator from . import mod_tools from . import mod_run as run from . import mod_parallel # Define logger level for debug purposes logger = logging.getLogger(__name__) #logger = multiprocessing.log_to_stderr() #logger.setLevel(logging.DEBUG) # - Define global variables for progress bars istep = 0 ntot = 1 ifile = 0 def run_simulator(p, die_on_error=False): '''Main routine to run simulator, input is the imported parameter file, no outputs are returned but netcdf grid and data files are written as well as a skimulator.output file to store all used parameter. ''' # - Initialize some parameters values timestart = datetime.datetime.now() mod_tools.initialize_parameters(p) mod_tools.check_path(p) model = p.model ''' p.timeshift = getattr(p, 'timeshift', 0) if p.shift_time is None: p.timeshift = 0 ''' # - Read list of user model files """ if p.file_input is not None: model_data, list_file = run.load_coordinate_model(p) # if no modelbox is specified (modelbox=None), the domain of the input # data is taken as a modelbox # coordinates from the region defined by modelbox are selected if p.modelbox is not None: modelbox = numpy.array(p.modelbox, dtype='float') # Use convert to 360 data modelbox[0] = (modelbox[0] + 360) % 360 if modelbox[1] != 360: modelbox[1] = (modelbox[1] + 360) % 360 else: if p.file_input is not None: modelbox = model_data.calc_box(p) else: logger.error('modelbox should be provided if no model file is' 'provided') sys.exit(1) # - Extract data on modelbox # TODO: do only this step if modelbox is defined? Do it later? if p.file_input is not None: model_data.read_coordinates(p) # Select model data in the region modelbox model_data.model_index_lon = {} model_data.model_index_lat = {} model_data.model_index = {} model_data.vloncirc = {} for key in model_data.vlon.keys(): _lon = + model_data.vlon[key] _lat = + model_data.vlat[key] if p.grid == 'regular': if modelbox[0] < modelbox[1]: _tmp = numpy.where(((modelbox[0]-1) <= _lon) & (_lon <= (modelbox[1]+1)))[0] model_data.model_index_lon[key] = + _tmp else: _tmp = numpy.where(((modelbox[0]-1) <= _lon) | (_lon <= (modelbox[1]+1)))[0] model_data.model_index_lon[key] = + _tmp _tmp = numpy.where(((modelbox[2]-1) <= _lat) & (_lat <= (modelbox[3]+1)))[0] model_data.model_index_lat[key] = + _tmp model_data.vlon[key] = + _lon[model_data.model_index_lon[key]] model_data.vlat[key] = + _lat[model_data.model_index_lat[key]] else: if modelbox[0] < modelbox[1]: _tmp = numpy.where(((modelbox[0]-1) <= _lon) & (_lon <= (modelbox[1]+1)) & ((modelbox[2]-1) <= _lat) & (_lat <= (modelbox[3]+1))) model_data.model_index[key] = + _tmp else: _tmp = numpy.where(((modelbox[0]-1) <= _lon) | (_lon <= (modelbox[1]+1)) & ((modelbox[2]-1) <= _lat) & (_lat <= (modelbox[3]+1))) model_data.model_index[key] = + _tmp # prevent IDL issues _wlon = model_data.vlon[key] model_data.vloncirc[key] = numpy.rad2deg(numpy.unwrap(_wlon)) model_data.model = model else: model_data = [] list_file = [] # avoid issue with 0=360 for global modelbox if modelbox[1] == 0: modelbox[1] = 359.99 # - Make SKIM grid if necessary """ if p.makesgrid is True: logger.info('\n Force creation of SKIM grid') # make nadir orbit orb = build_swath.makeorbit(modelbox, p, orbitfile=p.filesat) # build swath around this orbit ok = False try: ok = build_swath.orbit2swath(modelbox, p, orb, die_on_error) except mod_parallel.MultiprocessingError: logger.error('An error occurred with the multiprocessing ' 'framework') traceback.print_exception(*sys.exc_info()) sys.exit(1) except mod_parallel.DyingOnError: logger.error('An error occurred and all errors are fatal') sys.exit(1) if not ok: logger.error('Errors occurred while generating grid files') sys.exit(1) logger.info("\n SKIM Grids and nadir tracks have been written in" "{}".format(p.outdatadir)) logger.info("-----------------------------------------------") # - Compute interpolated velocity and errors for each pass, at each # cycle logger.info('Compute interpolated velocity and errors:') # load all SKIM grid files (one for each pass) listsgridfile = sorted(glob.glob(p.filesgrid + '_p*.nc')) if not listsgridfile: logger.error('\n There is no SKIM grid file in {}, run simulator with' 'option makesgrid set to true in your params' 'file'.format(p.outdatadir)) sys.exit(1) # Build model time steps from parameter file modeltime = numpy.arange(0, p.nstep*p.timestep, p.timestep) # Remove the grid from the list of model files if p.file_input and p.file_grid_model is None: logger.info("WARNING: the first file is not used to build data") list_file.remove(list_file[0]) # - Loop on SKIM grid files jobs = [] p2 = mod_tools.todict(p) #time_yaw = None #vac_yaw = None error_generator = generator.Generator(p) #if p.attitude is True and os.path.isfile(p.yaw_file): # time_yaw, vac_yaw = build_error.load_yaw_aocs(p.yaw_file) # # time_yaw = time_yaw / 86400 for sgridfile in listsgridfile: jobs.append([sgridfile, p2, listsgridfile, list_file, modelbox, model_data, modeltime, error_generator]) ok = False try: ok = make_skim_data(p.proc_count, jobs, die_on_error, p.progress_bar) except mod_parallel.MultiprocessingError: logger.error('An error occurred with the multiprocessing framework') traceback.print_exception(*sys.exc_info()) sys.exit(1) except mod_parallel.DyingOnError: logger.error('An error occurred and all errors are fatal') sys.exit(1) # - Write Selected parameters in a txt file timestop = datetime.datetime.now() timestop = timestop.strftime('%Y%m%dT%H%M%SZ') timestart = timestart.strftime('%Y%m%dT%H%M%SZ') rw_data.write_params(p, os.path.join(p.outdatadir, 'skim_simulator.output')) if ok is True: if p.progress_bar is True: __ = mod_tools.update_progress(1, 'All passes have been processed', '') else: __ = logger.info('All passes have been processed') logger.info("\n Simulated skim files have been written in " "{}".format(p.outdatadir)) logger.info(''.join(['-'] * 61)) sys.exit(0) logger.error('\nERROR: At least one of the outputs was not saved.') sys.exit(1) def make_skim_data(_proc_count, jobs, die_on_error, progress_bar): """ Compute SWOT-like data for all grids and all cycle, """ # - Set up parallelisation parameters proc_count = min(len(jobs), _proc_count) status_updater = mod_tools.update_progress_multiproc jobs_manager = mod_parallel.JobsManager(proc_count, status_updater, exc_formatter, err_formatter) ok = jobs_manager.submit_jobs(worker_method_skim, jobs, die_on_error, progress_bar) if not ok: # Display errors once the processing is done jobs_manager.show_errors() return ok def exc_formatter(exc): """Format exception returned by sys.exc_info() as a string so that it can be serialized by pickle and stored in the JobsManager.""" error_msg = traceback.format_exception(exc[0], exc[1], exc[2]) return error_msg def err_formatter(pid, grid, cycle, exc): """Transform errors stored by the JobsManager into readable messages.""" msg = None if cycle < 0: msg = '/!\ Error occurred while processing grid {}'.format(grid) else: _msg = '/!\ Error occurred while processing cycle {} on grid {}' msg = _msg.format(cycle, grid) return msg def worker_method_skim(*args, **kwargs): msg_queue, sgridfile, p2, listsgridfile = args[:4] list_file, modelbox, model_data, modeltime, error_generator = args[4:] p = mod_tools.fromdict(p2) # Load SKIM grid files (Swath and nadir) sgrid = run.load_sgrid(sgridfile, p) # duplicate SKIM grids to assure that data are not modified and are # saved properly sgrid_tmp = run.load_sgrid(sgridfile, p) sgrid.gridfile = sgridfile # Convert cycle in days sgrid.cycle /= 86400. sgrid_tmp.cycle /= 86400. sgrid_tmp.indi = None # Select model data around the swath to reduce interpolation cost in # griddata ### TODO comment to be removed? # - Generate SKIM like data # Compute number of cycles needed to cover all nstep model timesteps rcycle = (p.timestep * p.nstep)/float(sgrid.cycle) ncycle = int(rcycle) error = generator.Generator(p) first_time = datetime.datetime.strptime(p.first_time, '%Y-%m-%dT%H:%M:%SZ') # Loop on all cycles for cycle in range(0, ncycle+1): # if ifile > (p.nstep*p.timestep + 1): # break # Create SKIM-like data msg_queue.put((os.getpid(), sgridfile, cycle + 1, None)) if p.file_input is None: model_data = [] # Initialize all list of variables (each beam is appended to the # list) # Initialize velocity, indices and mask to empty lists output_var = {} #for key in p.list_output: # output_var[key] = [] list_key = list(p.list_input_var.keys()) list_key.append('vindice') for key in list_key: output_var[key] = [] # TODO: proof # sigma0 complusory # uwnd, vwnd compulsory for Instrument and WaveDoppler # hs uwnd vwnd mssu mssc mssclose for WaveDoppler # Loop over the beams to interpolate model data time_all = [] for i in range(len(p.list_pos) + 1): sgrid_tmp.lon = sgrid.lon[i] sgrid_tmp.lat = sgrid.lat[i] sgrid_tmp.time = sgrid.time[i] time = sgrid_tmp.time / 86400. + sgrid.cycle * cycle # Interpolate observation on SKIM swath create = run.interp_on_swath(cycle, list_file, modelbox, sgrid_tmp, model_data, modeltime, p) output_var_i, time = create time_all.append(time) for key in output_var_i.keys(): output_var[key].append(output_var_i[key]) for key, value in output_var.items(): output_var[key] = numpy.transpose(numpy.array(value)) time_arr = numpy.transpose(numpy.array(time_all)) radial_angle = numpy.transpose(sgrid.radial_angle) ac_angle = numpy.array(sgrid.angle) radial_angle = numpy.array(sgrid.radial_angle) output_var.update(run.compute_geov(output_var, ac_angle, radial_angle, p)) dic_grid = {'lon': sgrid.lon, 'lat': sgrid.lat, 'incl': sgrid.incl, 'angle': sgrid.angle, 'ipass': sgrid.ipass} #ac_angle m ## TODO remove sgrid.x_al_nadir variable all_error = error.generate(sgrid.x_al, output_var, sgrid.angle, time_arr, sgrid.cycle, first_time, radial_angle, sgrid.x_al[:, 0], sgrid.x_ac) # for key, value in output_var.items(): # output_var[key] = numpy.transpose(value) for k, v in all_error.items(): output_var[k] = v # Compute correction p.delta_azim = 15 all_corr = error.correct(dic_grid, output_var, radial_angle, p.delta_azim) for k, v in all_corr.items(): output_var[k] = v # Add error to true interpolated fields obs = error.generate_obs(output_var, all_error, all_corr) for k, v in obs.items(): output_var[k] = v # Save outputs in a netcdf file if ((~numpy.isnan(numpy.array(output_var['vindice']))).any() or not p.file_input): sgrid.ncycle = cycle run.save_SKIM(cycle, sgrid, time_all, output_var, p) del time if p.file_input is not None: for key in model_data.vlon.keys(): model_data.vlon[key] = (model_data.vlon[key] + 360) % 360 modelbox[0] = (modelbox[0] + 360) % 360 modelbox[1] = (modelbox[1] + 360) % 360 del sgrid msg_queue.put((os.getpid(), sgridfile, None, None))
15,674
39.29563
79
py
skimulator
skimulator-master/skimulator/cli.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import sys import skimulator.mod_tools as mod_tools import argparse import logging logger = logging.getLogger(__name__) def run_script(): """Run SKIM Simulator""" from . import run_simulator # Setup logging main_logger = logging.getLogger() main_logger.handlers = [] handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) main_logger.addHandler(handler) main_logger.setLevel(logging.INFO) parser = argparse.ArgumentParser() parser.add_argument('params_file', nargs='?', type=str, default=None, help='Path of the parameters file') parser.add_argument('--die-on-error', action='store_true', default=False, help='Force simulation to quit on first error') parser.add_argument('--debug', action='store_true', default=False, help='Display debug log messages') args = parser.parse_args() if args.params_file is None: logger.error('Please specify a parameter file') sys.exit(1) if args.debug is True: main_logger.setLevel(logging.DEBUG) file_param = args.params_file p = mod_tools.load_python_file(file_param) try: run_simulator.run_simulator(p, args.die_on_error) except KeyboardInterrupt: logger.error('\nInterrupted by user (Ctrl+C)') sys.exit(1) sys.exit(0) def run_l2c(): """Run L2C reconstruction""" import skimulator.regridding as regridding # Setup logging main_logger = logging.getLogger() main_logger.handlers = [] handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) main_logger.addHandler(handler) main_logger.setLevel(logging.INFO) parser = argparse.ArgumentParser() parser.add_argument('params_file', nargs='?', type=str, default=None, help='Path of the parameters file') parser.add_argument('--die-on-error', action='store_true', default=False, help='Force simulation to quit on first error') parser.add_argument('--debug', action='store_true', default=False, help='Display debug log messages') args = parser.parse_args() if args.params_file is None: logger.error('Please specify a parameter file') sys.exit(1) if args.debug is True: main_logger.setLevel(logging.DEBUG) file_param = args.params_file p = mod_tools.load_python_file(file_param) try: regridding.run_l2c(p, args.die_on_error) except KeyboardInterrupt: logger.error('\nInterrupted by user (Ctrl+C)') sys.exit(1) sys.exit(0) def run_l2d(): """Run L2D reconstruction""" import skimulator.regridding_l2d as regridding # Setup logging main_logger = logging.getLogger() main_logger.handlers = [] handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) main_logger.addHandler(handler) main_logger.setLevel(logging.INFO) parser = argparse.ArgumentParser() parser.add_argument('params_file', nargs='?', type=str, default=None, help='Path of the parameters file') parser.add_argument('--die-on-error', action='store_true', default=False, help='Force simulation to quit on first error') parser.add_argument('--debug', action='store_true', default=False, help='Display debug log messages') args = parser.parse_args() if args.params_file is None: logger.error('Please specify a parameter file') sys.exit(1) if args.debug is True: main_logger.setLevel(logging.DEBUG) file_param = args.params_file p = mod_tools.load_python_file(file_param) try: regridding.run_l2d(p, args.die_on_error) except KeyboardInterrupt: logger.error('\nInterrupted by user (Ctrl+C)') sys.exit(1) sys.exit(0) def interpolate_l2d(): """Run L2D reconstruction""" import skimulator.regridding_l2d as regridding # Setup logging main_logger = logging.getLogger() main_logger.handlers = [] handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) main_logger.addHandler(handler) main_logger.setLevel(logging.INFO) parser = argparse.ArgumentParser() parser.add_argument('params_file', nargs='?', type=str, default=None, help='Path of the parameters file') parser.add_argument('--die-on-error', action='store_true', default=False, help='Force simulation to quit on first error') parser.add_argument('--debug', action='store_true', default=False, help='Display debug log messages') args = parser.parse_args() if args.params_file is None: logger.error('Please specify a parameter file') sys.exit(1) if args.debug is True: main_logger.setLevel(logging.DEBUG) file_param = args.params_file p = mod_tools.load_python_file(file_param) try: regridding.offline_interpolation(p) #, args.die_on_error) except KeyboardInterrupt: logger.error('\nInterrupted by user (Ctrl+C)') sys.exit(1) sys.exit(0)
5,891
31.196721
77
py
skimulator
skimulator-master/skimulator/regridding_l2d.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ '''L2D computation from L2b SKIM like data ''' import numpy import os import sys import glob import ctypes import datetime import time import math import netCDF4 from . import const from . import rw_data as rw from . import mod_tools from . import mod_run as mod import skimulator.mod_parallel import logging import traceback import multiprocessing logger = logging.getLogger(__name__) # Using global variables is discouraged but this one will only be used as a # simple way to share the observations vector with worker processes (the main # process writes to this variable, the workers should only read from it) obs_vector = {} def build_grid(p) -> dict: # Domain (spatial grid) if p.spatial_domain is not None: modelbox = numpy.array(p.spatial_domain, dtype='float') # Use convert to 360 data #modelbox[0] = (modelbox[0] + 360) % 360 #if modelbox[1] != 360: # modelbox[1] = (modelbox[1] + 360) % 360 else: logger.error('Please provide modelbox_l2d for L2d reconstruction') sys.exit(1) lon0, lon1, lat0, lat1 = modelbox global_domain = False if (lon1 - lon0) > 200: global_domain = True # Use convert to 360 data lon0o = (lon0 + 360) % 360 if lon1 != 360: lon1o = (lon1 + 360) % 360 # modelbox[0] = (modelbox[0] + 360) % 360 # if modelbox[1] != 360: # modelbox[1] = (modelbox[1] + 360) % 360 dlon, dlat = p.posting_l2d # ##################################### # Spatial grid construction # ##################################### grd = make_grid(lon0, lon1, dlon, lat0, lat1, dlat) dlatlr = dlat dlonlr = dlon grd['dlon'] = dlon grd['dlonlr'] = dlonlr grd['dlat'] = dlat grd['dlatlr'] = dlatlr grd['nxlr'] = int((lon1 - lon0) / dlonlr) grd['nylr'] = int((lat1 - lat0) / dlatlr) grd['modelbox'] = (lon0, lon1, lat0, lat1) return grd def run_l2d(p, die_on_error=False): config = p.config mod_tools.initialize_parameters(p) resols = p.resol_spatial_l2d # km resolt = p.resol_temporal_l2d # days grd = build_grid(p) lon0, lon1, lat0, lat1 = grd['modelbox'] # Filtering as a function of latitude (10d at Equator, 14 d at 60º) resolt = (-4 * numpy.cos(numpy.deg2rad(grd['lat'])) + 9) * resolt # ##################################### # READ L2B OBS # ##################################### tcycle = p.satcycle # Time domain time0, time1, dt = p.time_domain # TODO change resoltmax = numpy.max(resolt) c0 = int((max(time0 - resoltmax, 0)) / tcycle) + 1 c1 = int((time1 + resoltmax) / tcycle) + 1 obs = {} # c0 = 1 ; c1 = 2 p2 = mod_tools.todict(p) jobs = [] print(datetime.datetime.now()) for cycle in range(c0, c1 + 1, 1): _pat = '{}_c{:02d}_p*'.format(p.config, cycle) pat = os.path.join(p.outdatadir, _pat) listfile = glob.glob(pat) # Create specific funtion to retrieve time_unit if not listfile: continue filev = os.path.join(p.outdatadir, listfile[0]) _, _, time_unit = read_l2b(filev) for pattern in listfile: jobs.append([p2, pattern, lon0, lon1, lat0, lat1, time0, time1, resoltmax, grd['dlonlr'], grd['dlatlr']]) ok = False task = 0 results = [] try: ok, results = build_obs(p.proc_count, jobs, die_on_error, p.progress_bar) except skimulator.mod_parallel.MultiprocessingError: logger.error('An error occurred with the multiprocessing framework') traceback.print_exception(*sys.exc_info()) sys.exit(1) except skimulator.mod_parallel.DyingOnError: logger.error('An error occurred and all errors are fatal') sys.exit(1) print() print('Aggregating results...', datetime.datetime.now()) for i in range(len(results)): obs_rjobs = results[i] #key, ind_key, obs_r = results[i] for j in range(len(obs_rjobs)): try: obs_r = obs_rjobs[j] except: obs_r = obs_rjobs obs_r_keys = list(obs_r.keys()) for key in obs_r_keys: if key not in obs.keys(): obs[key] = {} for ind_key in obs_r[key].keys(): if ind_key not in obs[key].keys(): obs[key][ind_key] = [] obs[key][ind_key] = numpy.concatenate((obs[key][ind_key], obs_r[key][ind_key])) del(obs_r[key]) del(results) print('Building shared obs vector...', datetime.datetime.now()) # Populate the shared observations vector global obs_vector obs_vector = {} for key in obs: obs_vector[key] = {} obs_ind_keys = list(obs[key].keys()) for ind_key in obs_ind_keys: _buffer = multiprocessing.RawArray(ctypes.c_float, numpy.size(obs[key][ind_key])) narray = numpy.frombuffer(_buffer, dtype='float32') numpy.copyto(narray, obs[key][ind_key]) obs_vector[key][ind_key] = _buffer # Release memory from original array to keep the total amount of # RAM as low as possible if 'time' != key: del(obs[key][ind_key]) # release memory as soon as possible print(datetime.datetime.now()) print('Memory housekeeping...', datetime.datetime.now()) keys = list(obs.keys()) for key in keys: if 'time' != key: del(obs[key]) print(datetime.datetime.now()) #import pdb ; pdb.set_trace() # ##################################### # Independant time loop for each analysis # ##################################### enstime = numpy.arange(time0, time1 + dt, dt) time_model = numpy.arange(time0, time1, p.timestep) # Center at noon window = dt / 2. enstime = enstime + window indice_mask = make_mask(p, 'ucur', grd) list_key = {'ucur':'ux_true', 'vcur':'uy_true'} #, 'uwnd':'uwnd', #'vwnd': 'vwnd'} list_input = {} for key in list_key: list_input[key] = p.list_input_var_l2d[key] for it, timeref in enumerate(enstime): print(it, timeref) iobs = {} for ind_key in obs['time'].keys(): iobs[ind_key] = numpy.where((obs['time'][ind_key] > (timeref - numpy.max(resoltmax))) & (obs['time'][ind_key] < (timeref + numpy.max(resoltmax))))[0] make_oi(p, p2, grd, iobs, resols, timeref, resolt, indice_mask, die_on_error=die_on_error) # Interpolate model data to have a true reference #indice_model = it #print('read model') #model_data, out_var, list_file = read_model(p, it, p.dim_time, # list_input) #print('interpolate model') #if global_domain is False: # interpolate_model(p, model_data, out_var, grd, list_key) pattern = os.path.join(p.outdatadir, '{}_{}_l2d_'.format(config, p.config_l2d)) save_l2d(pattern, timeref, window, time_unit, grd) print(datetime.datetime.now()) def offline_interpolation(p): pattern = os.path.join(p.outdatadir, '{}_{}_l2d_'.format(p.config, p.config_l2d)) listfile = glob.glob(os.path.join(p.outdatadir, '{}*.nc'.format(pattern))) str_format = '%Y-%m-%dT%H:%M:%SZ' list_key = {'ucur':'ux_true', 'vcur':'uy_true' } #, 'uwnd': 'uwnd', 'vwnd':'vwnd',} list_input = {} dimtime = 'time' dimlon = 'lon' dimlat = 'lat' longname = { "ux_noerr": "Error-free zonal velocity", "uy_noerr": "Error-free meridional velocity", "ux_obs": "Observed zonal velocity", "uy_obs": "Observed meridional velocity", "ux_model": "Error-free zonal velocity", "uy_model": "Error-free meridional velocity", "ux_true": "True zonal velocity", "uy_true": "True meridional velocity", "uwnd": "Eastward wind", "vwnd": "Northward wind", "uuss": "Eastward Stockes drift", "vuss": "Northward Stockes drift", } unit = {"ux_noerr": "m/s", "ux_obs": "m/s", "uy_noerr": "m/s", "uy_obs": "m/s", "ux_true": "m/s", "uy_true": "m/s", "uwnd": "m/s", "vwnd": "m/s", "uuss": "m/s", "vuss": "m/s", } for key in list_key: list_input[key] = p.list_input_var_l2d[key] for ifile in listfile: fid = netCDF4.Dataset(ifile, 'a') start_date = datetime.datetime.strptime(fid.time_coverage_start, str_format) end_date = datetime.datetime.strptime(fid.time_coverage_end, str_format) model_start = datetime.datetime.strptime(p.first_time, str_format) it = -(model_start - start_date).days print('read l2d coordinates') grd = {} grd['lon'] = fid.variables['lon'][:] grd['lat'] = fid.variables['lat'][:] grd['lon2'], grd['lat2'] = numpy.meshgrid(grd['lon'], grd['lat']) print('read model') print(start_date, model_start) model_data, out_var, list_file = read_model(p, it, p.dim_time, list_input) print('interpolate model') interpolate_model(p, model_data, out_var, grd, list_key) list_var = ('ux_true', 'uy_true') for key in list_var: nvar = '{}'.format(key) var = fid.createVariable(nvar, 'f4', (dimtime, dimlat, dimlon), fill_value=-1.36e9) value = grd[key] try: var.units = unit[str(key)] except: var.units = '' try: var.long_name = longname[str(key)] except: var.long_name = str(key) if value.any(): mask = numpy.isnan(value) value[numpy.where(mask)] = -1.36e9 mask_ind = numpy.where(value < -1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value > 1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value == numpy.PINF) value[mask_ind] = -1.36e9 var[0, :, :] = value def worker_build_obs(*args, **kwargs): msg_queue, p2, pattern = args[:3] p = mod_tools.fromdict(p2) lon0, lon1, lat0, lat1, time0, time1, resolt, dlonlr, dlatlr = args[3:12] res_queue = args[-1] filev = os.path.join(p.outdatadir, pattern) obs = {} if os.path.isfile(filev): obs_i, mask_data, time_unit = read_l2b(filev) _lon = numpy.mod(obs_i['lon'] + 180, 360) - 180 # if lon1o < lon0o: # mask_lon = ((obs_i['lon'] < lon0o) & (obs_i['lon'] > lon1o)) # else: # mask_lon = ((obs_i['lon'] > lon0o) & (obs_i['lon'] < lon1o)) if lon1 < lon0: mask_lon = ((_lon < lon0) & (_lon > lon1)) else: mask_lon = ((_lon > lon0) & (_lon < lon1)) mask_lat = ((obs_i['lat'] > lat0) & (obs_i['lat'] < lat1)) mask_time = ((obs_i['time'] > time0 - resolt) & (obs_i['time'] < time1 + resolt)) mask = (mask_time & mask_lat & mask_lon & mask_data) #if not numpy.any(mask): # msg_queue.put((os.getpid(), filev, None, None)) # return None for key in obs_i.keys(): obs_i[key] = obs_i[key][mask] _lon = _lon[mask] ind_i = numpy.floor((obs_i['lat'] - lat0) / dlatlr) ind_j = numpy.floor(numpy.mod(_lon - lon0, 360) / dlonlr) unique_i = list(set(ind_i)) unique_j = list(set(ind_j)) for key in obs_i.keys(): if key not in obs.keys(): obs[key] = {} for i in unique_i: for j in unique_j: mask_ind = numpy.where((ind_i==i) & (ind_j == j)) if mask_ind[0].any(): ind_key = 10000 * int(i) + int(j) # '{}_{}'.format(int(i), int(j)) for key in obs_i.keys(): if ind_key not in obs[key].keys(): obs[key][ind_key] = [] obs[key][ind_key] = numpy.concatenate((obs[key][ind_key], obs_i[key][mask_ind])) res_queue.put(obs) # Pass result to parent method msg_queue.put((os.getpid(), filev, None, None)) def build_obs(_proc_count, jobs, die_on_error, progress_bar): """ Compute SWOT-like data for all grids and all cycle, """ # - Set up parallelisation parameters proc_count = min(len(jobs), _proc_count) status_updater = mod_tools.update_progress_multiproc jobs_manager = skimulator.mod_parallel.JobsManager(proc_count, status_updater, exc_formatter, err_formatter) # Include the results queue as last argument for each job for j in jobs: j.append(jobs_manager.res_queue) res = [] ok = jobs_manager.submit_jobs(worker_build_obs, jobs, die_on_error, progress_bar, results=res.append) if not ok: # Display errors once the processing is done jobs_manager.show_errors() return ok, res def save_l2d(filenc, timeref, window, time_unit, grd): dateformat = '%Y-%m-%dT%H:%M:%SZ' datewformat = '%Y%m%dT%H%M%S' time_std = netCDF4.num2date(timeref - window, time_unit) time_start = time_std.strftime(format=dateformat) time_std = netCDF4.num2date(timeref + window, time_unit) time_end = time_std.strftime(format=dateformat) time_std = netCDF4.num2date(timeref, time_unit) time_middle = time_std.strftime(format=datewformat) pattern = '{}{}.nc'.format(filenc, time_middle) grd['time'] = timeref metadata = {} metadata['first_time'] = time_unit metadata['file'] = pattern metadata['time_coverage_start'] = time_start metadata['time_coverage_end'] = time_end metadata['file'] = pattern if os.path.exists(pattern): os.remove(pattern) if 'uy_true' in grd.keys(): rw.write_l2d(metadata, grd, ux_noerr=grd['ux_noerr'], uy_noerr=grd['uy_noerr'], ux_obs=grd['ux_obs'], uy_obs=grd['uy_obs'], ux_true=grd['ux_true'], uy_true=grd['uy_true']) else: rw.write_l2d(metadata, grd, ux_noerr=grd['ux_noerr'], uy_noerr=grd['uy_noerr'], ux_obs=grd['ux_obs'], uy_obs=grd['uy_obs']) #, ux_true=grd['ux_true'], #uy_true=grd['uy_true']) def read_l2b(nfile, model_nan=0): fid = netCDF4.Dataset(nfile, 'r') obs_i = {} obs_i['ux'] = numpy.ma.array(fid.variables['ucur'][:]).flatten() obs_i['uy'] = numpy.ma.array(fid.variables['vcur'][:]).flatten() if 'ur_true' in fid.variables.keys(): obs_i['ur_true'] = numpy.ma.array(fid.variables['ur_true'][:]).flatten() if 'ur_obs' in fid.variables.keys(): obs_i['ur_obs'] = numpy.ma.array(fid.variables['ur_obs'][:]).flatten() if 'ussr' in fid.variables.keys(): obs_i['ussr'] = numpy.ma.array(fid.variables['ussr'][:]).flatten() if 'ussr_est' in fid.variables.keys(): obs_i['ussr_est'] = numpy.ma.array(fid.variables['ussr_est'][:]).flatten() if 'instr' in fid.variables.keys(): obs_i['instr'] = numpy.ma.array(fid.variables['instr'][:]).flatten() if 'uwnd' in fid.variables.keys(): vwnd = numpy.ma.array(fid.variables['vwnd'][:]).flatten() uwnd = numpy.ma.array(fid.variables['uwnd'][:]).flatten() nwnd = numpy.sqrt(uwnd**2 + vwnd**2) if 'wd' in fid.variables.keys(): obs_i['wdre'] = numpy.ma.array(fid.variables['wd'][:]).flatten() if 'wd_est' in fid.variables.keys(): obs_i['wdre'] -= numpy.ma.array(fid.variables['wd_est'][:]).flatten() obs_i['lon'] = numpy.ma.array(fid.variables['lon'][:]).flatten() #obs_i['lon'] = numpy.mod(obs_i['lon'] -180, 360) + 180 obs_i['lon'] = numpy.mod(obs_i['lon'] + 360, 360) obs_i['lat'] = numpy.ma.array(fid.variables['lat'][:]).flatten() obs_i['time'] = numpy.ma.array(fid.variables['time'][:]).flatten() angle = numpy.ma.array(fid.variables['radial_angle'][:]).flatten() obs_i['angle'] = numpy.mod(angle, 2 * numpy.pi) mask_invalid = (numpy.ma.getmaskarray(obs_i['ux']) | numpy.ma.getmaskarray(obs_i['uy']) | numpy.ma.getmaskarray(obs_i['ur_true']) | numpy.ma.getmaskarray(obs_i['ur_obs'])) mask_invalid = (mask_invalid | (obs_i['ux']==model_nan) | (obs_i['uy']==model_nan) | (obs_i['ur_true']==model_nan) | (obs_i['ur_obs']==model_nan)) if 'uwnd' in fid.variables.keys(): mask_invalid = (mask_invalid | (nwnd < 4)) if 'wd' in fid.variables.keys(): mask_invalid = (mask_invalid | (abs(obs_i['wdre'])>0.5)) mask_data = ~(mask_invalid) time_unit = fid.variables['time'].units fid.close() return obs_i, mask_data, time_unit def make_grid(lon0, lon1, dlon, lat0, lat1, dlat): grd = {} grd['lon'] = numpy.arange(lon0, lon1 + dlon, dlon) grd['lon'] = numpy.mod(grd['lon'] + 360, 360) - 360 grd['lat'] = numpy.arange(lat0, lat1 + dlat, dlat) grd['lon'] = numpy.mod(grd['lon'] +360, 360) grd['lon2'], grd['lat2'] = numpy.meshgrid(grd['lon'], grd['lat']) grd['nx'] = len(grd['lon']) grd['ny'] = len(grd['lat']) return grd def make_oi(p, p2, grd, iobs, resols, timeref, resolt, index, die_on_error=False): indices = [(j, i) for j, i in zip(index[0], index[1])] # Split the list of (j, i) couples into chunks that the workers will # process in parallel. The method to decide how many chunks are required is # purely empirical chunk_size = int(len(indices) / (3 * p.proc_count)) # 3 is a magic number chunk_size = max(chunk_size, 1) # avoid chunk_size = 0 due to rounding ind_count = math.ceil(len(indices) / chunk_size) jobs = [] for k in range(0, ind_count): jobs.append([p2, grd, iobs, resols, timeref, resolt, index, indices[k * chunk_size :(k + 1) * chunk_size]]) ok = False try: ok = par_make_oi(grd, p.proc_count, jobs, die_on_error, p.progress_bar) except skimulator.mod_parallel.MultiprocessingError: logger.error('An error occurred with the multiprocessing framework') traceback.print_exception(*sys.exc_info()) sys.exit(1) except skimulator.mod_parallel.DyingOnError: logger.error('An error occurred and all errors are fatal') sys.exit(1) def init_oi_worker(obs_dict): """Initialize the obs_vector global variable with the value passed by the parent process. obs_vector is not modified by the worker, so sharing this object between the main process and the workers should not consume additional memory due to Copy On Write fork/spawn behavior (at least on Linux).""" global obs_vector obs_vector = obs_dict def par_make_oi(grd, _proc_count, jobs, die_on_error, progress_bar): """ Compute SWOT-like data for all grids and all cycle, """ global obs_vector # - Set up parallelisation parameters proc_count = min(len(jobs), _proc_count) status_updater = mod_tools.update_progress_multiproc jobs_manager = skimulator.mod_parallel.JobsManager(proc_count, status_updater, exc_formatter, err_formatter, init_oi_worker, (obs_vector,)) for j in jobs: j.append(jobs_manager.res_queue) jobs_res = [] ok = jobs_manager.submit_jobs(worker_make_oi, jobs, die_on_error, progress_bar, results=jobs_res.append) print() print('par_make_oi complete', datetime.datetime.now()) if not ok: # Display errors once the processing is done jobs_manager.show_errors() # Reconstruct results grids grd['ux_noerr'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) grd['uy_noerr'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) grd['ux_obs'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) grd['uy_obs'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) grd['ux_uss'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) grd['uy_uss'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) #grd['ux_'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) #grd['uy_uwd'] = numpy.full((grd['ny'], grd['nx']), numpy.nan) for k in range(0, len(jobs_res)): for res in jobs_res[k]: j, i, ux_noerr, uy_noerr, ux_obs, uy_obs = res grd['ux_noerr'][j, i] = ux_noerr grd['uy_noerr'][j, i] = uy_noerr grd['ux_obs'][j, i] = ux_obs grd['uy_obs'][j, i] = uy_obs del(jobs_res) return ok def worker_make_oi(*args, **kwargs): msg_queue = args[0] p2, grd, iobs, resols, timeref, resolt, index, indices = args[1:9] res_queue = args[-1] p = mod_tools.fromdict(p2) global obs_vector jobs_res = [] for j, i in indices: obs = {} ni = 1 nj = 1 dlat = grd['dlat'] dlatlr = grd['dlatlr'] dlon = grd['dlon'] dlonlr = grd['dlonlr'] resoltij = resolt[j] # Filtering as a function of latitude (90km at Eqautor, 40 km at 60º) resolsij = (100 * numpy.cos(numpy.deg2rad(grd['lat'][j])) - 10) * resols jlr = int(numpy.floor(j*dlat/dlatlr)) ilr = int(numpy.floor(i*dlon/dlonlr)) for jx in range(max(0, jlr-nj), min(jlr + nj + 1, grd['nylr'])): for jy in range(max(0, ilr-ni), min(ilr + ni + 1, grd['nxlr'])): # ind_key = '{}_{}'.format(jx, jy) ind_key = 10000 * int(jx) + int(jy) # '{}_{}'.format(int(i), int(j)) if ind_key not in iobs.keys(): continue if ~iobs[ind_key].any(): continue for key in obs_vector.keys(): if key not in obs.keys(): obs[key] = [] # Rebuild numpy array from shared memory (no copy) _buffer = obs_vector[key][ind_key] narray = numpy.frombuffer(_buffer, dtype='float32') _obs = narray[numpy.array(iobs[ind_key], dtype='int')] obs[key] = numpy.concatenate(([obs[key], _obs])) # TODO: to be optimized, especially for global, ... flat_ind = j + i*grd['nylr'] if 'lon' not in obs.keys(): continue # Handle IDL and Greenwich line _lonobs = numpy.mod(obs['lon'] + 360, 360) _longrd = numpy.mod(grd['lon2'][j, i] + 360, 360) dlon = 180 - abs(abs(_lonobs - _longrd) - 180) # dist = 110. * (numpy.cos(numpy.deg2rad(grd['lat'][j]))**2 * (dlon)**2 # + (obs['lat'] - grd['lat2'][j, i])**2)**0.5 fac = 1 if abs(grd['lat'][j]) < 10: fac = fac + (10 - abs(grd['lat'][j])) / 8 dist = 110. * (1 / fac*numpy.cos(numpy.deg2rad(grd['lat'][j]))**2 * (dlon)**2 + fac * (obs['lat'] - grd['lat2'][j, i])**2)**0.5 iiobs=numpy.where((dist < resolsij))[0] if len(iiobs)>=2: H = numpy.zeros((len(iiobs), 2)) H[:, 0] = numpy.cos(obs['angle'][iiobs]) H[:, 1] = numpy.sin(obs['angle'][iiobs]) win_s = numpy.exp(-dist[iiobs]**2/(0.5*resolsij)**2) time_cen = obs['time'][iiobs] - timeref win_t = numpy.exp(-time_cen**2/(0.5 * resoltij)**2) # exp window Ri = win_s * win_t RiH = numpy.tile(Ri, (2, 1)).T*H M = numpy.dot(H.T, RiH) if numpy.linalg.cond(M) < 1e3: Mi = numpy.linalg.inv(M) eta_true = numpy.dot(numpy.dot(Mi, RiH.T), obs['ur_true'][iiobs]) eta_obs = numpy.dot(numpy.dot(Mi, RiH.T), obs['ur_obs'][iiobs]) jobs_res.append((j, i, eta_true[0], eta_true[1], eta_obs[0], eta_obs[1])) # Pass results for all the processed (j, i) to the parent process res_queue.put(jobs_res) # Notify parent process about job completion msg_queue.put((os.getpid(), j, i, None)) def read_model(p, ifile, dim_time, list_input): model_data, list_file = mod.load_coordinate_model(p) model_data.read_coordinates(p) nfile = int(ifile) print(nfile) filename = os.path.join(p.indatadir, list_file[nfile]) print(filename) model_step_ctor = getattr(rw, p.model) out_var = {} for i in range(dim_time): model_step = model_step_ctor(p, ifile=(filename, ), list_input_var=list_input, time=i) model_step.read_var(p) for key in list_input.keys(): if not key in out_var.keys(): out_var[key] = model_step.input_var[key] / dim_time else: out_var[key] = (out_var[key] + model_step.input_var[key] / dim_time) return model_data, out_var, list_file def make_mask(p, key, grid): """ Return indices of points on the ocean (non masked value in the model) """ list_input = {key: p.list_input_var_l2d[key]} model_data, out_var, list_file = read_model(p, 0, 1, list_input) mask_ucur = numpy.ma.getmaskarray(out_var['ucur']) mask_ucur = (mask_ucur | numpy.isnan(out_var['ucur'])) out_var['ucur'] = numpy.ma.array(out_var['ucur'], mask=mask_ucur) list_key = {'ucur':'mask'} interpolate_model(p, model_data, out_var, grid, list_key) mask_index = numpy.where(~numpy.ma.getmaskarray(grid['mask'])) # & (grid['mask'] != 0) # & (~numpy.isnan(grid['mask']))) # TODO TO proof if mask_index is empty return mask_index def interpolate_model(p, model_data, model_var, grd, list_key): import pyresample as pr wrap_lon = pr.utils.wrap_longitudes geom = pr.geometry.SwathDefinition interp = mod.interpolate_irregular_pyresample lon = wrap_lon(grd['lon2']) for ikey, okey in list_key.items(): if len(p.list_input_var[ikey]) > 2: grid_number = p.list_input_var[ikey][2] else: grid_number = 0 _lon = wrap_lon(model_data.vlon[grid_number]) _lat = model_data.vlat[grid_number] if len(numpy.shape(_lon)) <= 1: _lon, _lat = numpy.meshgrid(_lon, _lat) swath_def = geom(lons=_lon, lats=_lat) grid_def = geom(lons=lon, lats=grd['lat2']) var = model_var[ikey] grd[okey] = interp(swath_def, var, grid_def, p.resol, interp_type=p.interpolation) grd[okey][grd[okey] == p.model_nan] = numpy.nan return grd def exc_formatter(exc): """Format exception returned by sys.exc_info() as a string so that it can be serialized by pickle and stored in the JobsManager.""" error_msg = traceback.format_exception(exc[0], exc[1], exc[2]) return error_msg def err_formatter(pid, it, cycle, exc): """Transform errors stored by the JobsManager into readable messages.""" msg = '/!\ Error occurred while processing it {}'.format(it) #if cycle < 0: # msg = '/!\ Error occurred while processing grid' #else: # _msg = '/!\ Error occurred while processing cycle {}' # msg = _msg.format(cycle) return msg
29,349
38.715832
87
py
skimulator
skimulator-master/skimulator/regridding.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ '''L2C computation from L2b SKIM like data ''' import numpy import os import sys import glob import scipy.interpolate import datetime import time import skimulator.const as const import skimulator.rw_data as rw import skimulator.mod_tools as mod_tools import skimulator.mod_run as mod import skimulator.mod_parallel import logging import traceback logger = logging.getLogger(__name__) def make_obs(p, data, grid, obs): #obs['vrt'] = obs['vrt'][ind] ind = obs['ind'] obs['vxt'] = numpy.squeeze(numpy.array(data['ucur']).flatten())[ind] obs['vyt'] = numpy.squeeze(numpy.array(data['vcur']).flatten())[ind] obs['vmodr'] = numpy.squeeze(numpy.array(data['ur_true']).flatten())[ind] if 'Instrument' in p.noise: obs['instr'] = numpy.squeeze(numpy.array(data['instr']).flatten())[ind] if 'WaveDoppler' in p.noise: if 'wd_est' in data.keys(): obs['wdre'] = numpy.squeeze((numpy.array(data['wd']) - numpy.array(data['wd_est'])).flatten())[ind] else: obs['wdre'] = numpy.squeeze(numpy.array(data['wd']).flatten())[ind] obs['lon'] = numpy.squeeze(numpy.array(data['lon']).flatten())[ind] obs['lat'] = numpy.squeeze(numpy.array(data['lat']).flatten())[ind] obs['lon_nadir'] = numpy.squeeze(numpy.array(data['lon_nadir'][:])) obs['lat_nadir'] = numpy.squeeze(numpy.array(data['lat_nadir'][:])) obs['time'] = numpy.squeeze(numpy.array(data['time']).flatten())[ind] # + (cycle-1)*self.tcycle obs['time_nadir'] = numpy.array(data['time_nadir'][:]) obs['vindice'] = numpy.squeeze(numpy.array(data['vindice']).flatten())[ind] obs['dir'] = numpy.squeeze(numpy.mod(grid.angle + numpy.pi/2, 2*numpy.pi).flatten())[ind] obs['angle'] = numpy.squeeze(numpy.mod(grid.radial_angle, 2*numpy.pi).flatten())[ind] obs['al_nadir'] = grid.x_al_nadir if p.type_satellite == 'stream': obs['al'] = (numpy.array(grid.x_al) + obs['al_nadir']) # + numpy.tile(obs['al_nadir'], (obs['nbeam'], 1)).transpose()) else: obs['al'] = (numpy.array(grid.x_al) + numpy.tile(obs['al_nadir'], (obs['nbeam'], 1)).transpose()) obs['al'] = numpy.squeeze(obs['al'].flatten())[ind] obs['ac'] = numpy.squeeze(numpy.array(grid.x_ac).flatten())[ind] return obs def across_track(lon, lat, posting, max_ac, desc=False): npoints = 1 nind = len(lon) SatDir = numpy.zeros((int(nind/npoints), 3)) SatLoc = numpy.zeros((int((nind)/npoints), 3)) s2cart = mod_tools.spher2cart(lon[::npoints], lat[::npoints]) SatLoc[:, 0], SatLoc[:, 1], SatLoc[:, 2] = s2cart # Compute satellite direction (SatLoc is periodic) SatDir[1: -1, 0] = ((SatLoc[2:, 0] - SatLoc[: -2, 0]) / numpy.sqrt(SatLoc[1: -1, 0]**2 + SatLoc[1: -1, 1]**2 + SatLoc[1: -1, 2]**2)) SatDir[1: -1, 1] = ((SatLoc[2:, 1] - SatLoc[: -2, 1]) / numpy.sqrt(SatLoc[1: -1, 0]**2 + SatLoc[1: -1, 1]**2 + SatLoc[1: -1, 2]**2)) SatDir[1: -1, 2] = ((SatLoc[2:, 2] - SatLoc[: -2, 2]) / numpy.sqrt(SatLoc[1: -1, 0]**2 + SatLoc[1: -1, 1]**2 + SatLoc[1: -1, 2]**2)) SatDir[-1, :] = SatDir[-2, :] SatDir[0, :] = SatDir[1, :] # Rotate from earth center around satellite direction to compute # swath points of angles between the borders of the swath in left # and right swath nhalfswath = int(max_ac/posting) new_lon = numpy.zeros((nind, 2*nhalfswath+1)) new_lat = numpy.zeros((nind, 2*nhalfswath+1)) ac = numpy.zeros((2*nhalfswath+1)) for i in range(0, nind, npoints): for j in range(0, int(nhalfswath)+1): #j = int(nhalfswath) - j ac[nhalfswath + j] = j*posting ac[nhalfswath - j] = -j*posting R = mod_tools.rotationmat3D(float((j*posting) / (const.Rearth*10**-3)), SatDir[int(i/npoints), :]) ObsLoc = numpy.dot(R, SatLoc[int(i/npoints)]) cs = mod_tools.cart2spher(ObsLoc[0], ObsLoc[1], ObsLoc[2]) new_lon[i, nhalfswath+j], new_lat[i, nhalfswath+j] = cs ObsLoc = numpy.dot(numpy.transpose(R), SatLoc[int(i/npoints)]) cs = mod_tools.cart2spher(ObsLoc[0], ObsLoc[1], ObsLoc[2]) new_lon[i, nhalfswath-j], new_lat[i, nhalfswath-j] = cs new_lon = new_lon[:, ::-1] new_lat = new_lat[:, ::-1] return new_lon, new_lat, ac def make_grid(grid, obs, posting, desc=False): grd = {} if obs['ac'].max() < 0: return None max_ac = abs(obs['ac']).max() # OI grid grd['dal'] = posting # if desc is True: # ind = numpy.where(obs['al_nadir'] > obs['al'].min()) # obs['al_nadir'] = obs['al_nadir'][ind] grd['al'] = numpy.arange(obs['al_nadir'].min(), obs['al_nadir'].max(), grd['dal']) #grd['lon'] = scipy.interpolate.griddata(_alac, obs['lon'], _inalac, # method='linear') #grd['lat'] = scipy.interpolate.griddata(_alac, obs['lat'], _inalac, # method='linear') lon360 = numpy.mod(obs['lon_nadir'] + 180, 360) - 180 lon = scipy.interpolate.griddata(obs['al_nadir'], obs['lon_nadir'], (grd['al']), method='linear') lat = scipy.interpolate.griddata(obs['al_nadir'], obs['lat_nadir'], (grd['al']), method='linear') grd['lon'], grd['lat'], grd['ac'] = across_track(lon, lat, posting, max_ac + grd['dal'], desc=desc) grd['ac2'], grd['al2'] = numpy.meshgrid(grd['ac'], grd['al']) grd['nal'], grd['nac'] = numpy.shape(grd['ac2']) grd['time'] = scipy.interpolate.griddata(obs['al_nadir'], obs['time_nadir'], (grd['al']), method='linear') grd['time_nadir'] = obs['time_nadir'] grd['angle'] = numpy.full((grd['nal'], grd['nac']), numpy.nan) grd['angle'][:, :-1] = numpy.angle((grd['lon'][:, 1:] - grd['lon'][:, :-1]) * (numpy.cos(numpy.deg2rad(grd['lat'][:, 1:]))) + 1j * (grd['lat'][:, 1:] - grd['lat'][:, :-1])) grd['angle'][:, -1]= grd['angle'][:, -2] #grd['vobsal'] = numpy.full((grd['nal'], grd['nac']), numpy.nan) #grd['vobsac'] = numpy.full((grd['nal'], grd['nac']), numpy.nan) #grd['vmodal'] = numpy.full((grd['nal'], grd['nac']), numpy.nan) #grd['vmodac'] = numpy.full((grd['nal'], grd['nac']), numpy.nan) return grd def perform_oi_1(grd, obs, resol, desc=False): obsal = numpy.full((grd['nal'], grd['nac']), numpy.nan) obsac = numpy.full((grd['nal'], grd['nac']), numpy.nan) for j in range(grd['nac']): for i in range(grd['nal']): dist = numpy.sqrt((obs['ac'] - grd['ac2'][i, j])**2 + (obs['al'] - grd['al2'][i, j])**2) ios = numpy.where((dist < resol))[0] if len(ios) >= 4: H = numpy.zeros((len(ios), 2)) H[:, 0] = numpy.cos(obs['dir'][ios]) H[:, 1] = numpy.sin(obs['dir'][ios]) std_err = numpy.ones((len(ios))) # rectangular filtering window for now... # Ri = std_err**-2 Ri=numpy.exp(-dist[ios]**2/(0.5*resol)**2) # exp window RiH = numpy.tile(Ri, (2, 1)).T * H M = numpy.dot(H.T, RiH) Mi = numpy.linalg.inv(M) eta_obs = numpy.dot(numpy.dot(Mi, RiH.T), obs['vobsr'][ios]) #eta_mod = numpy.dot(numpy.dot(Mi, RiH.T), obs['vmodr'][ios]) j2 = j #if desc is True: # j2 = grd['nac'] - 1 - j obsal[i, j2]=eta_obs[0] obsac[i, j2]=eta_obs[1] #grd['vmodal'][i, j2]=eta_mod[0] #grd['vmodac'][i, j2]=eta_mod[1] return obsal, obsac """ def perform_oi_2(grd, obs, resol): # - In parameter file ## TODO - # Number of pixel (resolution for healpix) nside = 256 # Number of diamonds for healpix ndiam = 12 ntotpixel = nside * nside * ndiam # Conditionning threshold thresh_cond = 10 ph = 2 * numpy.pi - numpy.deg2rad(lon) th = numpy.pi / 2 - numpy.deg2rad(lat) pidx = heal.ang2pix(nside, th, ph) for i in range(nbeam): for j in range(ndata): if ur[j, i] > -1E9: ip = pidx[j,i] # compute imulated model im[ip, 1] += u[j, i] im[ip, 2] += v[j,i] nim[ip] += 1 # compute covariance(s) model co = numpy.cos(rangle[j,i]) si = numpy.sin(rangle[j,i]) w = ww[j,i] cov[ip, 0, 0] += co * co cov[ip, 1, 0] += si * co cov[ip, 0, 1] += si * co cov[ip, 1, 1] += si * si cov2[ip, 0, 0] += w * co * co cov2[ip, 1, 0] += w * si * co cov2[ip, 0, 1] += w * si * co cov2[ip, 1, 1] += w * si * si # compute data vector model vec[ip, 0] += co * ur[j,i] vec[ip, 1] += si * ur[j,i] # compute data noise vector model vec2[ip, 0] += w* co * uro[j,i] vec2[ip, 1] += w * si * uro[j,i] # compute doppler projection for k in range(3): vecdop[k, ip, 0] += w * co * tdop[j,i,k] vecdop[k, ip, 1] += w * si * tdop[j,i,k] """ def read_model(p, indice): model_data, list_file = mod.load_coordinate_model(p) model_data.read_coordinates(p) list_model_step = [] for ifile in indice: nfile = int(ifile /p.dim_time) filetime = ifile - nfile * p.dim_time _tmpfilename = list_file[nfile] filename = os.path.join(p.indatadir, _tmpfilename) model_step_ctor = getattr(rw, p.model) model_step = model_step_ctor(p, ifile=(filename, ), list_input_var=p.list_input_var, time=filetime) model_step.read_var(p) list_model_step.append(model_step) return model_data, list_model_step, list_file def interpolate_model(p, model_data, list_model_step, grd, list_obs, list_key, desc=False): import pyresample as pr wrap_lon = pr.utils.wrap_longitudes geom = pr.geometry.SwathDefinition interp = mod.interpolate_irregular_pyresample lon = wrap_lon(grd['lon']) # list_key = {'ucur':'u_model', 'vcur':'v_model'} for ikey, okey in list_key.items(): grd[okey] = numpy.full(numpy.shape(grd['lat']), numpy.nan) if ikey not in p.list_input_var.keys(): continue if len(p.list_input_var[ikey]) > 2: grid_number = p.list_input_var[ikey][2] else: grid_number = 0 _lon = wrap_lon(model_data.vlon[grid_number]) _lat = model_data.vlat[grid_number] if len(numpy.shape(_lon)) <= 1: _lon, _lat = numpy.meshgrid(_lon, _lat) swath_def = geom(lons=_lon, lats=_lat) for i in range(len(list_model_step)): model_step = list_model_step[i] if desc is False: ind_lat = numpy.where(grd['lat']> list_obs[i]) else: ind_lat = numpy.where(grd['lat']< list_obs[i]) model_step.read_var(p) grid_def = geom(lons=lon[ind_lat], lats=grd['lat'][ind_lat]) #grid_def = pr.geometry.SwathDefinition(lons=lon, #[ind_lat], # lats=grd['lat']) #[ind_lat]) var = model_step.input_var[ikey] _tmp = interp(swath_def, var, grid_def, 4*p.resol, interp_type=p.interpolation) grd[okey][ind_lat] = _tmp return grd def write_l2(outfile, grd, obs, cycle, passn, firsttime): if os.path.exists(outfile): os.remove(outfile) metadata = {} metadata['file'] = outfile dateformat = '%Y-%m-%dT%H:%M:%SZ' time_model = datetime.datetime.strptime(firsttime, '%Y-%m-%dT%H:%M:%SZ') grdtime0 = numpy.nanmin(grd['time_nadir']) grdtime1 = numpy.nanmax(grd['time_nadir']) if numpy.isnan(grdtime0): grdtime0 = 0 if numpy.isnan(grdtime1): grdtime1 = 0 day = numpy.floor(grdtime0) seconds = (grdtime0 - day) * 86400 time0 = time_model + datetime.timedelta(day, seconds) day = numpy.floor(grdtime1) seconds = (grdtime1 - day) * 86400 time1 = time_model + datetime.timedelta(day, seconds) metadata['time_coverage_start'] = time0.strftime(format=dateformat) metadata['time_coverage_end'] = time1.strftime(format=dateformat) metadata['cycle'] = cycle metadata['pass'] = passn metadata['first_time'] = firsttime #geolocation = {} #geolocation['lon'] rw.write_l2c(metadata, grd, u_ac_obs=obs['ur_obs']['ac'], u_al_obs=obs['ur_obs']['al'], u_ac_noerr=obs['ur_true']['ac'], u_al_noerr=obs['ur_true']['al'], angle=grd['angle'], ux_obs=obs['ur_obs']['x'], uy_obs=obs['ur_obs']['y'], ux_noerr=obs['ur_true']['x'], uy_noerr=obs['ur_true']['y'], ux_true=grd['u_model'], uy_true=grd['v_model'], u_ac_true=grd['vtrueac'], u_al_true=grd['vtrueal'], u_ac_instr=obs['instr']['ac'], u_al_instr=obs['instr']['al'], #u_ac_wdrem=grd['uwdreac'], u_al_wdrem=grd['uwdreal'], #u_ac_uss_obs=grd['ussobsac'], u_al_uss_obs=grd['ussobsal'], #u_ac_uss_oi=grd['ussoiac'], u_al_uss_oi=grd['ussoial'], #u_ac_uss_true=grd['usstrueac'], u_al_uss_true=grd['usstrueal'], uwnd=grd['uwnd'], vwnd=grd['vwnd'], rain=grd['rain'], # mssu=grd['mssu'], mssc=grd['mssc'], #u_ac_wd=obs['wd']['ac'], u_al_wd=obs['wd']['al'], #u_ac_dsigma=grd['dsigmaac'], u_al_dsigma=grd['dsigmaal'], ) def run_l2c(p, die_on_error=False): timestart = datetime.datetime.now() pattern = os.path.join(p.outdatadir, '{}_c*'.format(p.config)) list_file = glob.glob(pattern) gpath = os.path.join(p.outdatadir, '{}_grid'.format(p.config)) mod_tools.initialize_parameters(p) iterate = 0 # - Loop on SKIM grid files jobs = [] p2 = mod_tools.todict(p) for ifile in list_file: jobs.append([ifile, p2, gpath]) ok = False try: ok = make_skim_l2c(p.proc_count, jobs, die_on_error, p.progress_bar) except skimulator.mod_parallel.MultiprocessingError: logger.error('An error occurred with the multiprocessing framework') traceback.print_exception(*sys.exc_info()) sys.exit(1) except skimulator.mod_parallel.DyingOnError: logger.error('An error occurred and all errors are fatal') sys.exit(1) # - Write Selected parameters in a txt file timestop = datetime.datetime.now() timestop = timestop.strftime('%Y%m%dT%H%M%SZ') timestart = timestart.strftime('%Y%m%dT%H%M%SZ') rw.write_params(p, os.path.join(p.outdatadir, 'skim_l2c.output')) if ok is True: if p.progress_bar is True: __ = mod_tools.update_progress(1, 'All passes have been processed', '') else: __ = logger.info('All passes have been processed') logger.info("\n Simulated skim files have been written in " "{}".format(p.outdatadir)) logger.info(''.join(['-'] * 61)) sys.exit(0) logger.error('\nERROR: At least one of the outputs was not saved.') sys.exit(1) def make_skim_l2c(_proc_count, jobs, die_on_error, progress_bar): """ Compute SWOT-like data for all grids and all cycle, """ # - Set up parallelisation parameters proc_count = min(len(jobs), _proc_count) status_updater = mod_tools.update_progress_multiproc jobs_manager = skimulator.mod_parallel.JobsManager(proc_count, status_updater, exc_formatter, err_formatter) ok = jobs_manager.submit_jobs(worker_method_l2c, jobs, die_on_error, progress_bar) if not ok: # Display errors once the processing is done jobs_manager.show_errors() return ok def exc_formatter(exc): """Format exception returned by sys.exc_info() as a string so that it can be serialized by pickle and stored in the JobsManager.""" error_msg = traceback.format_exception(exc[0], exc[1], exc[2]) return error_msg def err_formatter(pid, grid, cycle, exc): """Transform errors stored by the JobsManager into readable messages.""" msg = None if cycle < 0: msg = '/!\ Error occurred while processing grid {}'.format(grid) else: _msg = '/!\ Error occurred while processing cycle {} on grid {}' msg = _msg.format(cycle, grid) return msg def worker_method_l2c(*args, **kwargs): msg_queue, ifile, p2, gpath = args[:] p = mod_tools.fromdict(p2) passn = int(ifile[-6:-3]) if passn % 2 == 0: desc = True else: desc = False cycle = int(ifile[-10:-8]) fileg = '{}_p{:03d}.nc'.format(gpath, passn) data = rw.Sat_SKIM(ifile=ifile) grid = rw.Sat_SKIM(ifile=fileg) list_var_l2c = ['ur_obs', 'ur_true', 'instr', 'wdre', 'wd', 'dsigma', 'ussr', 'ussr_est'] list_var_l2c = ['ur_obs', 'ur_true', 'instr', 'wd'] l2b_dic = data.load_data(p, list_var_l2c, ucur=[], vcur=[], time=[], lon_nadir=[], lat_nadir=[], lon=[], lat=[], time_nadir=[], vindice=[], uwnd=[], vwnd=[]) #instr=[], uwd=[], uwd_est=[], dsigma=[], rain=[], ussr=[], ussr_est=[], uwnd=[], vwnd=[]) grid.load_swath(p, radial_angle=[], angle=[], x_al=[], x_al_nadir=[], x_ac=[]) dic = {} l2c_dic = {} for key in list_var_l2c: dic[key] = {} l2c_dic[key] = {} test = numpy.array(l2b_dic['ur_true']) if len(numpy.shape(test))==2: onebeam = False nil, nbeams = numpy.shape(test) else: onebeam = True nil = numpy.shape(test)[0] nbeams = 1 sbeam_incid = numpy.zeros((nil, nbeams)) ### TODO Change this nwnd = numpy.sqrt(numpy.array(data.uwnd)**2 + numpy.array(data.vwnd)**2) for key in list_var_l2c: dic[key]['vobsr'] = numpy.squeeze(numpy.array(l2b_dic[key]).flatten()) dic[key] = flag_data(dic[key], p.rain_threshold) if key == 'ur_true': dic[key]['vmodr'] = numpy.squeeze(numpy.array(l2b_dic[key]).flatten()) dic[key]['nsamp'], dic[key]['nbeam'] = nil, nbeams dic = select_valid_data(dic) if len(dic['ur_obs']['vobsr'][:]) > 2 and len(data.lon_nadir) >2: try: for key in list_var_l2c: dic[key] = make_obs(p, l2b_dic, grid, dic[key]) grd = make_grid(grid, dic[list_var_l2c[0]], p.posting, desc=desc) grdnoerr = make_grid(grid, dic['ur_true'], p.posting, desc=desc) ## TODO proof error if grd is None: logger.info('no grid found') return if grdnoerr is None: logger.info('no grid found') return # OI for key in list_var_l2c: if key == 'ur_true': _grd = grdnoerr.copy() else: _grd = grd.copy() l2c_dic[key]['al'], l2c_dic[key]['ac'] = perform_oi_1(_grd, dic[key], p.resol, desc=desc) except: logger.info(f'pass {passn}') return vindice = l2b_dic['vindice'] diff_indice = vindice[1:] - vindice[:-1] ind = numpy.where(diff_indice != 0)[0] first_lat = numpy.min(grd['lat']) sign_uv = 1 for key in list_var_l2c: grd[key] = l2c_dic[key] #grd['vmodac'] = + grdnoerr['vobsac'][:] #grd['vmodal'] = + grdnoerr['vobsal'][:] if desc is True: first_lat = numpy.max(grd['lat']) sign_uv = -1 if ind.any(): vindice = [dic['ur_obs']['vindice'][0], dic['ur_obs']['vindice'][ind[0] + 1]] ind_lat = [first_lat, dic['ur_obs']['lat'][ind[0] + 1]] else: vindice = [dic['ur_obs']['vindice'][0],] ind_lat = [first_lat,] vindice = numpy.array(vindice) vmask = (vindice<0) if numpy.any(vmask): vindice[numpy.where(vmask)] = 0 model_data, model_step, list_file2 = read_model(p, vindice) list_model_key = {'ucur':'u_model', 'vcur':'v_model', 'uwnd': 'uwnd', 'vwnd': 'vwnd','mssx':'mssu', 'mssy': 'mssc', 'uuss': 'uuss', 'vuss': 'vuss' } if (p.rain is True) and (p.rain_file is None): list_model_key['rain'] = 'rain' else: grd['rain'] = numpy.array([]) grd = interpolate_model(p, model_data, model_step, grd, ind_lat, list_model_key, desc=desc) ac_thresh = p.ac_threshold grd, l2c_dic = alac2xy(grd, l2c_dic, ac_thresh, sign_uv) grd, l2c_dic = mask_from_model(grd, l2c_dic, list_model_key) pattern_out = '{}{}_l2c_c{:02d}_p{:03d}.nc'.format(p.config, p.config_l2c, cycle, passn) outfile = os.path.join(p.outdatadir, pattern_out) write_l2(outfile, grd, l2c_dic, cycle, passn, p.first_time) msg_queue.put((os.getpid(), ifile, None, None)) def alac2xy(grd: dict, l2c_dic: dict, ac_thresh: float, sign_uv: float )-> (dict, dict): for key in l2c_dic.keys(): l2c_dic[key]['ac'][numpy.abs(grd['ac2']) < ac_thresh] = numpy.nan l2c_dic[key]['x'] = sign_uv * (l2c_dic[key]['ac'] * numpy.cos(grd['angle']) + l2c_dic[key]['al'] * numpy.cos(grd['angle']+numpy.pi/2)) l2c_dic[key]['y'] = sign_uv * (l2c_dic[key]['ac'] * numpy.sin(grd['angle']) + l2c_dic[key]['al'] * numpy.sin(grd['angle']+numpy.pi/2)) grd['vtrueac'] = sign_uv *(grd['u_model']*numpy.cos(grd['angle']) + grd['v_model'] * numpy.sin(grd['angle'])) grd['vtrueal'] = sign_uv * (-grd['u_model']*numpy.sin(grd['angle']) + grd['v_model']*numpy.cos(grd['angle'])) return grd, l2c_dic def mask_from_model(grd: dict, l2c_dic: dict, list_model_key: list ) -> (dict, dict): _masku = numpy.ma.getmaskarray(grd['u_model']) _maskv = numpy.ma.getmaskarray(grd['v_model']) mask = ((grd['u_model'] == 0) | (grd['v_model'] == 0) | _masku | _maskv | (abs(grd['u_model']) > 10) | (abs(grd['v_model']) > 10)) for skey in list_model_key.keys(): key = list_model_key[skey] grd[key][mask] = numpy.nan grd['vtrueac'][mask] = numpy.nan grd['vtrueal'][mask] = numpy.nan for key in l2c_dic.keys(): for key2 in l2c_dic[key].keys(): l2c_dic[key][key2][mask] = numpy.nan return grd, l2c_dic def select_valid_data(dic: dict) -> dict: if 'wdre' in dic.keys() and 'ur_true' in dic.keys(): ind_mod = numpy.where((dic['ur_true']['vmodr'] > -100) & (abs(dic['wdre']['vobsr']) < 100) & (numpy.isfinite(dic['wdre']['vobsr'])))[0] dic['ur_true']['vobsr'] = dic['noerr']['vobsr'][ind_mod] dic['ur_true']['ind'] = ind_mod elif 'ur_true' in dic.keys(): ind_mod = numpy.where((dic['ur_true']['vmodr'] > -100))[0] dic['ur_true']['vobsr'] = dic['ur_true']['vobsr'][ind_mod] dic['ur_true']['ind'] = ind_mod else: ind_mod = None if 'ur_obs' in dic.keys(): ind_obs = numpy.where((abs(dic['ur_obs']['vobsr']) < 100) & (numpy.isfinite(dic['ur_obs']['vobsr'])))[0] dic['ur_obs']['vobsr'] = dic['ur_obs']['vobsr'][ind_mod] dic['ur_obs']['ind'] = ind_mod else: ind_mod = None if 'wdre' in dic.keys(): indwd = numpy.where((abs(dic['wdre']['vobsr']) < 100) & (numpy.isfinite(dic['wdre']['vobsr'])))[0] dic['wdre']['vobsr'] = dic['wdre']['vobsr'][indwd] dic['wd']['vobsr'] = dic['wd']['vobsr'][indwd] dic['wdre']['ind'] = indwd elif 'wd'in dic.keys(): dic['wd']['vobsr'] = dic['wd']['vobsr'][ind_mod] dic['wd']['ind'] = ind_mod for key in dic.keys(): if key != 'ur_true' and key != 'ur_obs' and key != 'wdre' and key != 'wd': dic[key]['vobsr'] = dic[key]['vobsr'][ind_mod] dic[key]['ind'] = ind_mod return dic def flag_data(dic: dict, rain_threshold: float) -> dict: if 'rain' in dic.keys(): dic['obs']['vobsr'][dic['obs']['rain'] > rain_threshold] = numpy.nan #obs['vobsr'][obs['rain']>p.rain_threshold] = numpy.nan if 'dsigma' in dic.keys(): dic['dsigma'][dic['dsigma']>0.5] = numpy.nan if 'wdre' in dic.keys(): dic['wdre']['vobsr'][abs(dic['wdre']['vobsr'])>1] = numpy.nan dic['wdre']['vobsr'][dic['nwnd'].flatten() < 4] = numpy.nan return dic
26,773
41.633758
109
py
skimulator
skimulator-master/skimulator/mod_run.py
'''Module to create one beam data: \n \n Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. ''' import os from scipy import interpolate import numpy import math import glob import sys import time import pickle import datetime import logging from . import build_swath from . import rw_data from . import mod_tools from . import const from .error import utils from typing import Tuple, Dict import multiprocessing # Define logger level for debug purposes logger = logging.getLogger(__name__) #logger = multiprocessing.log_to_stderr() #logger.setLevel(logging.DEBUG) def load_sgrid(sgridfile, p): '''Load SKIM swath and Nadir data for file sgridfile ''' # Load SKIM swath file sgrid = rw_data.Sat_SKIM(ifile=sgridfile) cycle = 0 x_al = [] x_ac = [] al_cycle = 0 timeshift = 0 sgrid.load_swath(p, cycle=cycle, x_al=x_al, x_al_nadir=[], x_ac=x_ac, al_cycle=al_cycle, timeshift=timeshift) # sgrid.loncirc = [] # for i in range(len(sgrid.lon)): # sgrid.loncirc.append(numpy.rad2deg(numpy.unwrap(sgrid.lon[i]))) # Extract the pass number from the file name ipass = int(sgridfile[-6: -3]) sgrid.ipass = ipass return sgrid def load_coordinate_model(p): model = p.model if p.file_input is not None: list_file = [line.strip() for line in open(p.file_input)] else: list_file = None # - Read model input coordinates ''' # If a list of model files are specified, read model file coordinates if p.file_input is not None: model_data_ctor = getattr(rw_data, model) if p.file_grid_model is not None: _filename = list(p.file_grid_model) else: logger.info("WARNING: First file of list of files is used for" "coordinates only") _filename = list_file[0].split(',') filename = [] for ifile in _filename: filename.append(os.path.join(p.indatadir, ifile)) #filename = os.path.join(p.indatadir, list_file[0]) model_data = model_data_ctor(p, ifile=filename, lon=list(p.lon), lat=list(p.lat)) return model_data, list_file def interpolate_regular_1D(lon_in, lat_in, var, lon_out, lat_out, Teval=None): ''' Interpolation of data when grid is regular and coordinate in 1D. ''' # To correct for IDL issues ind_sort = numpy.argsort(lon_in) lon_in = lon_in[ind_sort] var = var[:, ind_sort] if numpy.max(lon_in) > 359 and numpy.min(lon_in) < 1: ind_in1 = numpy.where(lon_in <= 180) ind_in2 = numpy.where(lon_in > 180) # lon_in[lon_in > 180] = lon_in[lon_in > 180] - 360 # lon_in = np.mod(lon_in - (lref - 180), 360) + (lref - 180) # lon_in = numpy.rad2deg(numpy.unwrap(numpy.deg2rad(lon_in))) ind_out1 = numpy.where(lon_out <= 180) ind_out2 = numpy.where(lon_out > 180) # lon_out[lon_out > 180] = lon_out[lon_out > 180] - 360 # lon_out = numpy.rad2deg(numpy.unwrap(numpy.deg2rad(lon_out))) interp = interpolate.RectBivariateSpline mask_teval = (numpy.isnan(var) | numpy.ma.getmaskarray(var)) if Teval is None: Teval = numpy.zeros(numpy.shape(lon_out)) if ind_out1[0].any(): _tmp = interp(lat_in, lon_in[ind_in1], mask_teval[:, ind_in1[0]], kx=1, ky=1, s=0) Teval[ind_out1] = _tmp.ev(lat_out[ind_out1], lon_out[ind_out1]) if ind_out2[0].any(): _tmp = interp(lat_in, lon_in[ind_in2], mask_teval[:, ind_in2[0]], kx=1, ky=1, s=0) Teval[ind_out2] = _tmp.ev(lat_out[ind_out2], lon_out[ind_out2]) # Trick to avoid nan in interpolation var_mask = + var var_mask[numpy.isnan(var_mask) | numpy.ma.getmaskarray(var_mask)] = 0. # Interpolate variable var_out = numpy.full(numpy.shape(lon_out), numpy.nan) if ind_out1[0].any(): _tmp = interp(lat_in, lon_in[ind_in1], var_mask[:, ind_in1[0]], kx=1, ky=1, s=0) var_out[ind_out1] = _tmp.ev(lat_out[ind_out1], lon_out[ind_out1]) if ind_out2[0].any(): _tmp = interp(lat_in, lon_in[ind_in2], var_mask[:, ind_in2[0]], kx=1, ky=1, s=0) var_out[ind_out2] = _tmp.ev(lat_out[ind_out2], lon_out[ind_out2]) else: mask_teval = (numpy.isnan(var) | numpy.ma.getmaskarray(var)) # Interpolate mask if it has not been done (Teval is None) interp = interpolate.RectBivariateSpline if Teval is None: _Teval = interp(lat_in, lon_in, mask_teval, kx=1, ky=1, s=0) Teval = _Teval.ev(lat_out, lon_out) # Trick to avoid nan in interpolation var_mask = + var var_mask[numpy.isnan(var_mask) | numpy.ma.getmaskarray(var_mask)] = 0. # Interpolate variable _var_out = interp(lat_in, lon_in, var_mask, kx=1, ky=1, s=0) var_out = _var_out.ev(lat_out, lon_out) # Mask variable with Teval var_out[Teval > 0] = numpy.nan #var_out[Teval > 0] = numpy.nan #var_out[abs(var_out) > 1000] = numpy.nan return var_out, Teval def interpolate_irregular_pyresample(swath_in, var, grid_out, radius, interp_type='nearest'): ''' Interpolation of data when grid is irregular and pyresample is installed.''' import pyresample as pr interp = pr.kd_tree.resample_nearest if interp_type == 'nearest': interp = pr.kd_tree.resample_nearest radius_n = radius * 10**3 var_out = interp(swath_in, var, grid_out, radius_of_influence=radius_n, epsilon=100) else: interp = pr.kd_tree.resample_gauss radius_g = radius * 3 * 10**3 sigma_g = radius * 10**3 var_out = interp(swath_in, var, grid_out, radius_of_influence=radius_g, sigmas=sigma_g, fill_value=None) return var_out def interp_on_swath(cycle, list_file, modelbox, sgrid, model_data, modeltime, p): '''Create SKIM and nadir errors err and errnad, interpolate model velocity\ model_data on swath and nadir track, compute SKIM-like and nadir-like data for cycle, SKIM grid sgrid and ngrid. ''' # Initialiaze errors and velocity output_var_i = {} shape_i = numpy.shape(sgrid.lon)[0] for key in p.list_input_var.keys(): output_var_i[key] = numpy.full(shape_i, numpy.nan) output_var_i['vindice'] = numpy.full(shape_i, numpy.nan) date1 = cycle * sgrid.cycle # Definition of the time in the model time = sgrid.time / 86400. + date1 # convert seconds in days lon = sgrid.lon lat = sgrid.lat timeshift = sgrid.timeshift / 86400. # in days # Look for satellite data that are beween step-p.timestep/2 and # step+p.timestep/2 if p.file_input is not None: lon2D = {} lat2D = {} # meshgrid in 2D for interpolation purposes for key in model_data.vlon.keys(): if p.grid == 'irregular': lon2D[key], lat2D[key] = numpy.meshgrid(model_data.vlon[key], model_data.vlat[key]) index_filemodel = numpy.where(((time[-1] - timeshift) >= (modeltime-p.timestep/2.)) & ((time[0] - timeshift) < (modeltime+p.timestep/2.))) # local variable to find time record in file for WW3 nfile = 0 time_offset = 0 # At each step, look for the corresponding time in the satellite data for ifile in index_filemodel[0]: # If there are satellite data, Get true velcoity from model if numpy.shape(index_filemodel)[1] > 0: # Select part of the track that corresponds to the time of the # model (+-timestep/2) ind_time = numpy.where(((time-timeshift) >= (modeltime[ifile]-p.timestep/2.)) & ((time-timeshift) < (modeltime[ifile]+p.timestep/2.))) else: logger.error('No model file is found around time') sys.exit(1) # Load data from this model file # if output from ww3, time dimension is >1 (hourly outputs, # one file per month): conversion of ifile into file number # and record number ## TODO : Clean filetime = (ifile - time_offset)%p.dim_time # read next file when the time dimension is reached if filetime >= (time_offset + p.dim_time): time_offset += p.dim_time nfile += 1 filetime = (ifile - time_offset)%p.dim_time nfile = int(ifile /p.dim_time) filetime = ifile - nfile * p.dim_time _tmpfilename = list_file[nfile] filename = os.path.join(p.indatadir, _tmpfilename) model_step_ctor = getattr(rw_data, model_data.model) model_step = model_step_ctor(p, ifile=(filename, ), list_input_var=p.list_input_var, time=filetime) input_var_i = {} if p.grid == 'regular': model_step.read_var(p, ind_lon=model_data.ind_lon) for key in model_step.input_var.keys(): grid_key = model_step.numgrid[key] _indlat = model_data.model_index_lat[grid_key] _tmp = model_step.input_var[key][_indlat, :] _indlon = model_data.model_index_lon[grid_key] input_var_i[key] = +_tmp[:, _indlon] else: model_step.read_var(p, index=None) for key in model_step.input_var.keys(): _ind = model_data.model_index[model_step.numgrid[key]] input_var_i[key] = + model_step.input_var[key][_ind] # - Interpolate Model data on a SKIM grid and/or along the # nadir track # if grid is regular, use interpolate.RectBivariateSpline to # interpolate if p.grid == 'regular' and \ len(numpy.shape(model_data.vlon[0])) == 1: # Flatten satellite grid and select part of the track # corresponding to the model time for key in model_step.input_var.keys(): mlon = model_data.vlon[model_step.numgrid[key]] mlat = model_data.vlat[model_step.numgrid[key]] _tmp, Teval_u = interpolate_regular_1D(mlon, mlat, input_var_i[key], lon[ind_time[0]], lat[ind_time[0]]) output_var_i[key][ind_time[0]] = + _tmp else: # Grid is irregular, interpolation can be done using # pyresample module if it is installed or griddata # function from scipy. # Note that griddata is slower than pyresample functions. try: import pyresample as pr lon_wrap = pr.utils.wrap_longitudes geom = pr.geometry.SwathDefinition lon = lon_wrap(lon) grid_def = geom(lons=lon, lats=lat) for key in model_step.input_var.keys(): grid_key = model_step.numgrid[key] _ind = model_data.model_index[grid_key] _mlon = lon_wrap(model_data.vlon[grid_key]) _mlat = model_data.vlat[grid_key] if len(_mlon[0]) <= 1: _mlon = lon_wrap(lon2D[grid_key]) _mlat = lat2D[grid_key] swath_def = geom(lons=_mlon[_ind], lats=_mlat[_ind]) _tmp = interpolate_irregular_pyresample( swath_def, input_var_i[key], grid_def, p.posting, interp_type=p.interpolation) output_var_i[key][ind_time[0]] = + _tmp except ImportError: for key in model_step.input_var.keys(): grid_key = model_step.numgrid[key] _ind = model_data.model_index[grid_key] _mlon = model_data.vlon[grid_key] _mlat = model_data.vlat[grid_key] if len(_mlon) <= 1: _mlon = lon2D[grid_key] _mlat = lat2D[grid_key] lonravel = + _mlon[_ind].ravel() latravel = + _mlat[_ind].ravel() _tmp = + input_var_i[key].ravel() interp = interpolate.griddata((lonravel, latravel), _tmp, (lon[ind_time[0]], lat[ind_time[0]]), method=p.interpolation) output_var_i[key][ind_time[0]] = + interp # Force value outside modelbox at nan if modelbox[0] > modelbox[1]: for key in model_step.input_var.keys(): _ind = numpy.where(((lon > modelbox[0]) & (lon < modelbox[1])) | (lat < modelbox[2]) | (lat > modelbox[3])) output_var_i[key][_ind] = numpy.nan else: for key in model_step.input_var.keys(): _ind = numpy.where((lon < modelbox[0]) | (lon > modelbox[1]) | (lat < modelbox[2]) | (lat > modelbox[3])) output_var_i[key][_ind] = numpy.nan output_var_i['vindice'][ind_time[0]] = ifile else: pass # TODO to proof: creation of empty mss if no mss and p.instr is True return output_var_i, time def compute_geov(input_dic:dict, ac_angle: numpy.array, radial_angle, p, fc: int=1)-> Dict[str, numpy.ndarray]: _dic = {} # list compulsory " ucur, vcur + ice + require compute_sigma_water sigma0 = compute_sigma(input_dic, p.list_angle, ac_angle, p, fc=fc) _dic['sigma0'] = sigma0 _dic['ur_true'] = utils.proj_radial(input_dic['ucur'][:, fc:], input_dic['vcur'][:, fc:], radial_angle) _dic['ssh_true'] = input_dic['ssh'] _dic['radial_angle'] = radial_angle return _dic def compute_sigma_water(input_var, beam_angle, radial_angle, fc: int=1): required = ('mssx', 'mssy', 'mssd', 'uwnd', 'vwnd', 'uwnd', 'ucur', 'vcur') missing = [_ for _ in required if _ not in input_var.keys()] if 0 < len(missing): logger.info('Missing file to compute sigma, instrumental error not' ' computed') logger.info('Missing parameters: {}'.format(', '.join(missing))) return None else: mssd = numpy.deg2rad(input_var['mssd']) mssu = + input_var['mssx'] mssc = + input_var['mssy'] uwnd = + input_var['uwnd'] vwnd = + input_var['vwnd'] ucur = + input_var['ucur'] vcur = + input_var['vcur'] mssxl = mssu * numpy.cos(mssd)**2 + mssc * numpy.sin(mssd)**2 mssyl = mssu * numpy.sin(mssd)**2 + mssc * numpy.cos(mssd)**2 mssxyl = (mssu - mssc) * numpy.sin(2 * mssd) / 2 nwr = numpy.sqrt((uwnd - ucur)**2 + (vwnd - vcur)**2) mask = (nwr == 0) nwr[mask] = numpy.nan uwnd[mask] = numpy.nan vwnd[mask] = numpy.nan ucur[mask] = numpy.nan vcur[mask] = numpy.nan wrd = numpy.pi / 2 - numpy.arctan2(vwnd - vcur, uwnd - ucur) mssshort = numpy.log(nwr + 0.7) * 0.009 # Replace nan values by 0 to avoid a runtime warning (nan values # will be restored afterwards) mssshort_nanmask_ind = numpy.where(numpy.isnan(mssshort)) mssshort[mssshort_nanmask_ind] = 0 mssshort[mssshort < 0] = 0 mssshort[mssshort_nanmask_ind] = numpy.nan # restore nan values # Directionality for short wave mss (if 0.5: isotrophic) facssdw = 0.6 mssds = facssdw * mssshort msscs = mssshort - mssds mssxs = msscs * numpy.sin(wrd)**2 + mssds * numpy.cos(wrd)**2 mssys = mssds * numpy.sin(wrd)**2 + msscs * numpy.cos(wrd)**2 mssxys = abs(mssds - msscs) * numpy.sin(2* wrd) / 2 input_var['mssx'] = mssxs + mssxl input_var['mssy'] = mssys + mssyl input_var['mssxy'] = mssxys + mssxyl input_var['mssc'] = mssc input_var['mssu'] = mssu R2 = 0.55 nsample = numpy.shape(radial_angle)[0] rbeam_angle = [numpy.deg2rad(beam_angle), ]*nsample mssx = (mssxs + mssxl)[:, fc:] mssy = (mssys + mssyl)[:, fc:] mssxy = (mssxys + mssxyl)[:, fc:] mask = ((mssx == 0) | (mssy == 0)) mssx[mask] = numpy.nan mssy[mask] = numpy.nan expo = (-0.5 * numpy.tan(rbeam_angle)**2 * (numpy.cos(radial_angle)**2 *mssy + numpy.sin(radial_angle)**2 * mssx - numpy.sin(2 * radial_angle) * mssxy) / (mssx * mssy)) coeff = R2 / (2 * numpy.cos(rbeam_angle)**4 * numpy.sqrt(mssx * mssy)) sigma_water = coeff * numpy.exp(expo) # TODO sigma_water[:, 0] = numpy.nan #del input_var['mssd'] return sigma_water def compute_sigma(output_var_i, beam_angle, radial_angle, p, fc: int=1): sigma_water = compute_sigma_water(output_var_i, beam_angle, radial_angle, fc=fc) if (p.ice is True) and ('ice' in output_var_i.keys()): vice = output_var_i['ice'][:, fc:] mat_beam = numpy.array([beam_angle,]*numpy.shape(vice)[0]) sigma_ice = numpy.ones(numpy.shape(vice)) sigma_ice[mat_beam == 6] = 2.5 # TODO proof when angle is different from 6 and 12 #if beam_angle == 6: # sigma_ice = 2.5 #elif beam_angle == 12: # sigma_ice = 1 #else: # logger.info('beam angle is {} but should be either 6 or 12, ' # 'sigma_ice is set to 1'.format(beam_angle)) mask = (numpy.isnan(vice)) c_ice = numpy.ma.MaskedArray(vice, mask=mask) c_ice[c_ice.mask] = 0 sigma0 = (1 - c_ice) * sigma_water + c_ice * sigma_ice else: sigma0 = sigma_water return sigma0 def save_SKIM(cycle, sgrid, time, outdata, p): file_output = '{}_c{:02d}_p{:03d}.nc'.format(p.file_output, cycle + 1, sgrid.ipass) OutputSKIM = rw_data.Sat_SKIM(ifile=file_output, lon=sgrid.lon, lat=sgrid.lat, time=time, x_al=sgrid.x_al, cycle=sgrid.cycle) OutputSKIM.gridfile = sgrid.gridfile OutputSKIM.ipass = sgrid.ipass OutputSKIM.ncycle = sgrid.ncycle OutputSKIM.write_data(p, outdata) #ur_model=ur_model, index=vindice, # uss_err=err_uss, ur_uss=ur_uss, std_uss=std_uss, # nadir_err=[err.nadir, ], ur_obs=ur_obs, # instr=err_instr, u_model=u_model, v_model=v_model, # errdcos=errdcos) return None
20,831
43.8
79
py
skimulator
skimulator-master/skimulator/grid_check.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import os import zlib import base64 import struct import hashlib import logging logger = logging.getLogger(__name__) def _text_file_md5(file_path, block_size=1024*1024): """""" if not os.path.exists(file_path): raise FileNotFoundError(file_path) md5 = hashlib.md5() with open(file_path, 'r', encoding='utf-8') as f: while True: data = f.read(block_size) if not data: break md5.update(data.encode('utf-8')) file_hash = md5.hexdigest() return file_hash def get_hash_from_grid_params(p): """""" orbit_path = p.filesat # Path to a text file containing the elevation, the cycle and the track of the satellite orbit_cols = p.order_orbit_col # None|int[3] (order of columns in orbit file) modelbox = p.modelbox # None|float[4] (lonmin, lonmax, latmin, latmax) rotation_speed = p.rotation_speed # float (rotation speed of the antenna in tr/min) cycle = p.cycle # float (beam cycle duration) list_pos = p.list_pos # float[N] (position of each beam) list_angle = p.list_angle # float[N] (inclination of each beam) list_shift = p.list_shift # int[N] (time shift wrt nadir for each beam) shift_lon = p.shift_lon # None|float (longitude shift in degrees) shift_time = p.shift_time # None|float (time shift in days) orbit_file_hash = _text_file_md5(orbit_path) orbit_md5 = orbit_file_hash.encode('utf-8') logger.debug('Orbit file MD5: {}'.format(orbit_file_hash)) # Flags for optional parameters has_cols = orbit_cols is not None has_modelbox = modelbox is not None has_shift_lon = shift_lon is not None has_shift_time = shift_time is not None fmt = '!B32sffI{}{}{}'.format('f' * len(list_pos), 'f' * len(list_angle), 'i' * len(list_shift)) logger.debug('Mandatory params serialization format: {}'.format(fmt)) flags = 0b00000000 opts = [] if has_cols: opts.append(struct.pack('!BBB', orbit_cols)) else: flags = flags + 0b00000001 if has_modelbox: opts.append(struct.pack('!ffff', modelbox)) else: flags = flags + 0b00000010 if has_shift_lon: opts.append(struct.pack('!f', shift_lon)) else: flags = flags + 0b00000100 if has_shift_time: opts.append(struct.pack('!f', shift_time)) else: flags = flags + 0b00001000 h = struct.pack(fmt, flags, orbit_md5, rotation_speed, cycle, len(list_pos), *list_pos, *list_angle, *list_shift) for d in opts: h = h + d return h def get_grid_params_from_hash(h): """""" flags, orbit_md5, rot, cycle, nbeams = struct.unpack('!B32sffI', h[:45]) has_cols = 0 >= (flags & 0b00000001) has_modelbox = 0 >= (flags & 0b00000010) has_shift_lon = 0 >= (flags & 0b00000100) has_shift_time = 0 >= (flags & 0b00001000) istart = 45 iend = 45 + 4 * nbeams # assuming 1 float <=> 4 bytes list_pos = struct.unpack('!{}f'.format(nbeams), h[istart:iend]) istart = iend iend = istart + 4 * nbeams # assuming 1 float <=> 4 bytes list_angle = struct.unpack('!{}f'.format(nbeams), h[istart:iend]) istart = iend iend = istart + 4 * nbeams # assuming 1 int <=> 4 bytes list_shift = struct.unpack('!{}i'.format(nbeams), h[istart:iend]) istart = iend if has_cols: iend = istart + 3 orbit_cols = struct.unpack('!BBB', h[istart:iend]) istart = iend else: orbit_cols = None if has_modelbox: iend = istart + 4 * 4 # assuming 1 float <=> 4 bytes modelbox = struct.unpack('!ffff', h[istart:iend]) istart = iend else: modelbox = None if has_shift_lon: iend = istart + 4 # assuming 1 float <=> 4 bytes shift_lon = struct.unpack('!f', h[istart:iend])[0] istart = iend else: shift_lon = None if has_shift_time: iend = istart + 4 # assuming 1 float <=> 4 bytes shift_time = struct.unpack('!f', h[istart:iend])[0] istart = iend else: shift_time = None result = {'order_orbit_col': orbit_cols, 'rotation_speed': rot, 'cycle': cycle, 'list_pos': list_pos, 'list_angle': list_angle, 'list_shift': list_shift, 'modelbox': modelbox, 'shift_lon': shift_lon, 'shift_time': shift_time, 'orbit_file_md5': orbit_md5} logger.debug('') logger.debug('=== Reconstructed parameters ===') logger.debug('orbit file MD5: {}'.format(orbit_md5.decode('utf-8'))) logger.debug('rotation_speed: {}'.format(rot)) logger.debug('cycle: {}'.format(cycle)) logger.debug('nbeams: {}'.format(nbeams)) logger.debug('list_pos: {}'.format(list_pos)) logger.debug('list_angle: {}'.format(list_angle)) logger.debug('list_shift: {}'.format(list_shift)) logger.debug('order_orbit_col: {}'.format(orbit_cols)) logger.debug('modelbox: {}'.format(modelbox)) logger.debug('shift_lon: {}'.format(shift_lon)) logger.debug('shift_time: {}'.format(shift_time)) return result def get_b64_gzipped_hash(p): """""" h = get_hash_from_grid_params(p) compressed = zlib.compress(h, 9) b64_bytes = base64.b64encode(compressed) b64_str = b64_bytes.decode('utf-8') return b64_str def revert_b64_gzipped_hash(b64_str): """""" b64_bytes = b64_str.encode('utf-8') compressed = base64.b64decode(b64_bytes) h = zlib.decompress(compressed) p = get_grid_params_from_hash(h) return p if '__main__' == __name__: import sys import logging import params_8beams_fram as p main_logger = logging.getLogger() main_logger.handlers = [] handler = logging.StreamHandler() handler.setLevel(logging.DEBUG) main_logger.addHandler(handler) main_logger.setLevel(logging.DEBUG) try: h = compute_grid_hash(p) except FileNotFoundError: _, e, _ = sys.exc_info() logger.error('Orbit file "{}" not found'.format(e.args[0])) # Compress information as much as possible... print(base64.b64encode(zlib.compress(h, 9)).decode('utf-8')) # Maybe use the CRC32 in the filename and include the hash in the NetCDF # attributes print(zlib.crc32(h)) read_grid_params_from_hash(h)
7,140
30.879464
116
py
skimulator
skimulator-master/skimulator/__init__.py
# ======================================================================= # General Documentation """Utilities for SKIM Science Simulator for the ocean Some useful online help commands for the package: * help(skimulator): Help for the package. A list of all modules in this package is found in the "Package Contents" section of the help output. * help(skimulator.M): Details of each module "M", where "M" is the module's name. #----------------------------------------------------------------------- # Additional Documentation # Authors: Lucile Gaultier # # Modification History: # - Jan 2021: V4.0 # - Jul 2019: V3.0 # - Oct 2018: V2.0 # - Dec 2017: V1.0 # - Mar 2017: Original by Lucile Gaultier, ODL # # Notes: # - Written for Python 3.9, tested with Python 3.7, 3.9 # #----------------------------------------------------------------------- Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ # ----------------------------------------------------------------------- # ---------------- Module General Import and Declarations --------------- # - Set module version to package version: __version__ = '4.0' __author__ = 'Lucile Gaultier <lucile.gaultier@oceandatalab.com>' __date__ = '2021-01-01' __email__ = 'lucile.gaultier@oceandatalab.com' __url__ = '' __description__ = ('SKIM Simulator') __author_email__ = ('lucile.gaultier@oceandatalab.com') __keywords__ = () import os # Try to improve accuracy of the version number by reading the build number in # the share/VERSION.txt file if it is available try: import pkg_resources _version_path = pkg_resources.resource_filename('skimulator', 'share/VERSION.txt') if os.path.exists(_version_path): with open(_version_path, 'rt') as f: _commits = f.readline().strip() __version__ = '{}.{}'.format(__version__, _commits) _ = f.readline() # Commit hash, useful for debug __date__ = f.readline().strip() except ImportError: pass except KeyError: # It seems that some versions of setuptools/pkg_resources try to read the # file once its path has been rebuilt, which causes an error when the file # does not exist. pass # - If you're importing this module in testing mode, or you're running # pydoc on this module via the command line, import user-specific # settings to make sure any non-standard libraries are found: import sys if (__name__ == "__main__") or \ ("pydoc" in os.path.basename(sys.argv[0])): import user # - Find python version number __python_version__ = sys.version[:3] # - Import numerical array formats try: import numpy except ImportError: print(''' Numpy is not available on this platform, ''') # - Import scientific librairies try: import scipy except ImportError: print("""Scipy is not available on this platform, """) # - Import netcdf reading librairies try: import netCDF4 except ImportError: print(''' netCDF4 is not available on this machine, ''') # reading and writing netcdf functions in rw_data.py won't work'''
3,814
31.887931
78
py
skimulator
skimulator-master/skimulator/spline.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import numpy #/******************************************************** #* bplsine_init_uniform * #* * #* initialize an allocated bspline struct with * #* uniform knots between a and b * #* * #* - bspline : a pointer to an allocated bspline * #* - a, b : interval of the bspline * #* */ class spline: def __init__(self, k, gp2): self.k = k self.g = gp2 - 2 # number of interior knots self.Nknots = self.g + 2 + 2 * self.k # total number of knots self.knots = numpy.zeros([self.Nknots]) self.Nvals = self.g + self.k + 1 self.Ncoefs = self.Nvals self.coefs = numpy.zeros([self.Ncoefs]) self.dp = numpy.zeros([self.k + 1]) self.dm = numpy.zeros([self.k + 1]) def init_uniform(self, a, b): self.a = a self.b = b k = self.k g = self.g Nknots = self.Nknots for i in range(k): self.knots[i] = a for i in numpy.arange(k,k + g+1,1): self.knots[i] = a + (i - k) / (g + 1.) * (b - a) for i in numpy.arange(k + g + 1,Nknots,1): self.knots[i] = b def find_interval_uniform(self,x): k = self.k g = self.g a = self.a b = self.b j = numpy.floor(((x - a) / (b - a) * (g + 1.)) + k).astype('int') if (x < a).any(): j[numpy.where(x < a)] = -1 if (x > b).any(): j[numpy.where(x > b)] = -1 if (j > g + k).any(): j[numpy.where(j > g + k)] = g+k return(j) def bspline_value (self, x): k = self.k g = self.g a = self.a b = self.b dp = self.dp dm = self.dm n=len(x) # Find interval i=self.find_interval_uniform (x) if (i < 0).any(): return([0, 0, 0, 0], -1, -2) # interval of vals affected istart = i - k iend = i vals = numpy.zeros([k + 1, n]) dp = numpy.zeros([self.k + 1, n]) dm = numpy.zeros([self.k + 1, n]) vals[0, :] = 1. for s in numpy.arange(1, k + 1, 1): dp[s - 1,:] = self.knots[i + s] - x dm[s - 1,:] = x - self.knots[i + 1 - s] prev = numpy.zeros([n]) old = vals[0 ,:] for r in range(1, s + 1, 1): M = old[:] / (dp[r - 1, :] + dm[s + 1 - r - 1, :]) old = 1*vals[r , :] vals[r - 1 , :] = prev[:] + dp[r - 1, :] * M[:] vals[r , :] = dm[s + 1 - r - 1, :] * M[:] prev = 1*vals[r , :] return (vals,istart,iend) def init_val(self,yy): self.coefs=yy def get_val(self,xx): vals,istart,iend=self.bspline_value(xx) sy=0*xx # print(istart.min(),istart.max()) # print(iend.min(),iend.max()) for i in range(4): # print((istart[:] + i).max()) sy+=vals[i, :]*self.coefs[istart[:]+i] return(sy) class spline_circular: def __init__(self,k,gp2): self.k = k self.g = gp2 - 1 # number of interior knots self.Nknots = self.g + 2 + 2 * (self.k+1) # total number of knots self.knots = numpy.zeros([self.Nknots]) self.Nvals = self.g + self.k + 1 self.Ncoefs = self.Nvals self.coefs = numpy.zeros([self.Ncoefs]) self.dp = numpy.zeros([self.k + 1]) self.dm = numpy.zeros([self.k + 1]) def init_uniform(self,a,b): self.a = a self.b = b k = self.k g = self.g Nknots = self.Nknots self.knots = (numpy.arange(Nknots)-k)*(b-a)/(g+1)+a def find_interval_uniform(self,x): k = self.k g = self.g a = self.a b = self.b j = numpy.floor(((x - a) / (b - a) * (g + 1.)) + k).astype('int') if (x < a).any(): j[numpy.where(x < a)] = -1 if (x > b).any(): j[numpy.where(x > b)] = -1 if (j > g + k).any(): j[numpy.where(j > g + k)] = g+k return(j) def bspline_value (self, x): k = self.k g = self.g a = self.a b = self.b dp = self.dp dm = self.dm n=len(x) # Find interval i=self.find_interval_uniform (x) if (i < 0).any(): return([0,0,0,0],-1,-2) # interval of vals affected istart = i - k iend = i vals = numpy.zeros([k+1,n]) dp = numpy.zeros([self.k + 1,n]) dm = numpy.zeros([self.k + 1,n]) vals[0,:] = 1. for s in numpy.arange(1,k+1,1): dp[s - 1,:] = self.knots[i + s] - x dm[s - 1,:] = x - self.knots[i + 1 - s] prev = numpy.zeros([n]) old = vals[0 ,:] for r in range(1,s+1,1): M = old[:] / (dp[r - 1,:] + dm[s + 1 - r - 1,:]) old = 1*vals[r ,:] vals[r - 1 ,:] = prev[:] + dp[r - 1,:] * M[:] vals[r ,:] = dm[s + 1 - r - 1,:] * M[:] prev = 1*vals[r ,:] return (vals,istart%(g+1),iend) def init_val(self,yy): self.coefs=yy def get_val(self,xx): vals,istart,iend=self.bspline_value(xx) sy=0*xx for i in range(4): sy+=vals[i,:]*self.coefs[istart[:]+i] return(sy)
6,038
25.603524
74
py
skimulator
skimulator-master/skimulator/rw_data.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ ''' Module to read and write data Contains functions to read variables and coordinates from a netcdf files. \n Contains model classes: \n -ROMS \n -NEMO \n -NETCDF_MODEL \n -WW3 \n Contains satellite class: Sat_SKIM \n Contains file instrumentation class: file_instr \n ''' import skimulator import skimulator.grid_check from netCDF4 import Dataset import numpy import sys import time as ti import logging import datetime import os version = skimulator.__version__ logger = logging.getLogger(__name__) class IncompatibleGridError(Exception): """Raised""" def __init__(self, path, grid_hash, params_hash, *args, **kwargs): """""" self.path = path self.grid = 0 #skimulator.grid_check.revert_b64_gzipped_hash(grid_hash) self.p = 0 #self.p = skimulator.grid_check.revert_b64_gzipped_hash(params_hash) def write_params(params, pfile): """ Write parameters that have been selected to run swot_simulator. """ with open(pfile, 'w') as f: for key in dir(params): if not key[0:2] == '__': f.write('{} = {}\n'.format(key, params.__dict__[key])) def read_coordinates(nfile, nlon, nlat, twoD=True): ''' General routine to read coordinates in a netcdf file. \n Inputs are file name, longitude name, latitude name. \n Outputs are longitudes and latitudes (2D arrays).''' # - Open Netcdf file try: fid = Dataset(nfile, 'r') except IOError: logger.error('There was an error opening the file {}'.format(nfile)) sys.exit(1) # - Read 1d or 2d coordinates try: vartmp = fid.variables[nlat] except: logger.error('Coordinates {} not found in file {}'.format(nlat, nfile)) sys.exit(1) try: vartmp = fid.variables[nlon] except: logger.error('Coordinates {} not found in file {}'.format(nlon, nfile)) sys.exit(1) if len(vartmp.shape) == 1: lon_tmp = numpy.array(fid.variables[nlon][:]).squeeze() lat_tmp = numpy.array(fid.variables[nlat][:]).squeeze() if twoD: lon, lat = numpy.meshgrid(lon_tmp, lat_tmp) else: lon = lon_tmp lat = lat_tmp elif len(vartmp.shape) == 2: lon = numpy.array(fid.variables[nlon][:, :]).squeeze() lat = numpy.array(fid.variables[nlat][:, :]).squeeze() if not twoD: lon = lon[0, :] lat = lat[:, 0] else: logger.warn('unknown dimension for lon and lat') fid.close() return lon, lat def read_var(nfile, var, index=None, time=0, depth=0, model_nan=None): ''' General routine to read variables in a netcdf file. \n Inputs are file name, variable name, index=index to read part of the variables, time=time to read a specific time, depth=depth to read a specific depth, model_nan=nan value ''' # - Open Netcdf file try: fid = Dataset(nfile, 'r') except IOError: logger.error('There was an error opening the file {}'.format(nfile)) sys.exit(1) # - Check dimension of variable try: vartmp = fid.variables[var] except: logger.error('Variable {} not found in file {}'.format(var, nfile)) sys.exit(1) # - Read variable if index is None: if len(vartmp.shape) == 1: T = numpy.ma.array(fid.variables[var][:]).squeeze() elif len(vartmp.shape) == 2: T = numpy.ma.array(fid.variables[var][:, :]).squeeze() elif len(vartmp.shape) == 3: if time is None: T = numpy.ma.array(fid.variables[var][:, :, :]).squeeze() else: T = numpy.ma.array(fid.variables[var][time, :, :]).squeeze() elif len(vartmp.shape) == 4: if time is None: if depth is None: T = numpy.ma.array(fid.variables[var][:, :, :, :]).squeeze() else: T = numpy.ma.array(fid.variables[var][:, depth, :, :]).squeeze() elif depth is None: T = numpy.ma.array(fid.variables[var][time, :, :, :]).squeeze() else: T = numpy.ma.array(fid.variables[var][time,depth, :, :]).squeeze() else: logger.error('wrong dimension in variables {}'.format(var)) sys.exit(1) else: if len(vartmp.shape) == 1: Ttmp = numpy.ma.array(fid.variables[var][:]).squeeze() T = Ttmp[index] elif len(vartmp.shape) == 2: Ttmp = numpy.ma.array(fid.variables[var][:, :]).squeeze() T = Ttmp[index] elif len(vartmp.shape) == 3: if time is None: U = numpy.ma.array(fid.variables[var][:, :, :]).squeeze() T = U[:, index] else: U = numpy.ma.array(fid.variables[var][time, :, :]).squeeze() T = U[index] elif len(vartmp.shape) == 4: if time is None: if depth is None: U = numpy.ma.array(fid.variables[var][:, :, :, :]).squeeze() T = U[:, :, index] else: U = numpy.ma.array(fid.variables[var][:, depth, :, :]).squeeze() T = U[:, index] elif depth is None: U = numpy.ma.array(fid.variables[var][time, :, :, :]).squeeze() T = U[:, index] else: U = numpy.ma.array(fid.variables[var][time, depth, :, :]).squeeze() T = U[index] else: logger.error('Wrong dimension') sys.exit(1) fid.close() # - Mask value that are NaN #try: # mask = (fid.variables[var]._fill_value == fill_value) #except: # mask = (T == model_nan) # else: # mask = (T == numpy.nan) if model_nan is not None: _mask = numpy.ma.getmaskarray(T) # TODO remove hard coded value? _mask = (_mask | (T == model_nan) | (abs(T) > 1000)) T = numpy.ma.array(T, mask=_mask) #T[numpy.where(T == model_nan)] = numpy.nan T._shared_mask = False return T #numpy.ma.MaskedArray(T, mask=mask) def global_attributes(fid, level): fid.title = 'SKIM {} simulated by SKIM simulator'.format(level) fid.keywords = 'SKIM, Doppler' # Check keywords fid.Conventions = "CF-1.6" fid.summary = 'SKIM data produced' fid.description = "SKIM fixed swath" fid.Metadata_Conventions = "Unidata Dataset Discovery v1.0" fid.history = '{} File created by skimulator version {}'.format(level, version) fid.processing_level = level fid.standard_name_vocabulary = "CF-1.6" fid.creator_name = "Lucile Gaultier" fid.creator_email = "lucile.gaultier@gmail.com" fid.publisher_url = "" fid.project = "SKIM" fid.date_created = ti.strftime("%Y-%m-%dT%H:%M:%SZ") fid.date_modified = ti.strftime("%Y-%m-%dT%H:%M:%SZ") fid.keywords_vocabulary = "" fid.references = "" def write_l2c(metadata, geolocation, **kwargs): ti = datetime.datetime.now() lat = geolocation['lat'] lon = geolocation['lon'] lon = numpy.mod(lon + 180, 360) - 180 tim = geolocation['time'] # - Open Netcdf file in write mode fid = Dataset(metadata['file'], 'w', format='NETCDF4_CLASSIC') # - Create Global attribute global_attributes(fid, 'L2C') fid.description = "SKIM fixed swath" fid.time_coverage_start = metadata['time_coverage_start'] # p.date0+"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.time_coverage_end = metadata['time_coverage_end'] # p.date0 +"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.geospatial_lat_min = "{:.2f}".format(numpy.min(lat)) fid.geospatial_lat_max = "{:.2f}".format(numpy.max(lat)) fid.geospatial_lat_units = "degrees_north" fid.geospatial_lon_max = "{:.2f}".format(numpy.max(lon)) fid.geospatial_lon_min = "{:.2f}".format(numpy.min(lon)) fid.geospatial_lon_units = "degrees_east" fid.project = "SKIM" fid.date_created = ti.strftime("%Y-%m-%dT%H:%M:%SZ") fid.date_modified = ti.strftime("%Y-%m-%dT%H:%M:%SZ") fid.cycle = "{0:d}".format(int(metadata['cycle'])) fid.track = "{} th pass".format(metadata['pass']) # - Create dimensions # if (not os.path.isfile(self.file)): dimlon = 'al' dimlat = 'ac' dimtime = 'time' nlon = numpy.shape(lon)[0] - 1 nlat = numpy.shape(lat)[1] ntime = None fid.createDimension(dimlat, nlat) fid.createDimension(dimtime, ntime) # - Create and write Variables vtime = fid.createVariable('time', 'f8', (dimtime)) vtime.axis = "T" vtime.units = "days since {}".format(metadata['first_time']) vtime.long_name = "Time" vtime.standard_name = "time" vtime.calendar = "gregorian" vtime[:] = tim vlon = fid.createVariable('lon', 'f4', (dimtime, dimlat)) vlon.axis = "X" vlon.long_name = "Longitude" vlon.standard_name = "longitude" vlon.units = "degrees_east" vlon[:, :] = lon val = fid.createVariable('x_al', 'f4', (dimtime)) val.axis = "Y" val.long_name = "Along track distance from beginning of cycle" val.standard_name = "Along track distance" val.units = "km" val[:] = geolocation['al'] vac = fid.createVariable('x_ac', 'f4', (dimlat)) vac.axis = "Y" vac.long_name = "Along track distance from nadir" vac.standard_name = "Across track distance" vac.units = "km" vac[:] = geolocation['ac'] vlat = fid.createVariable('lat', 'f4', (dimtime, dimlat)) vlat.axis = "Y" vlat.long_name = "Latitude" vlat.standard_name = "latitude" vlat.units = "degrees_north" vlat[:, :] = lat longname = { "ux_noerr": "Error-free zonal velocity", "uy_noerr": "Error-free meridional velocity", "ux_obs": "Observed zonal velocity", "uy_obs": "Observed meridional velocity", "u_ac_obs": "Observed across track velocity", "u_al_obs": "Observed along track velocity", "ux_model": "Error-free zonal velocity", "uy_model": "Error-free meridional velocity", "u_ac_noerr": "Error-free across track velocity", "u_al_noerr": "Error-free along track velocity", "angle": "angle of xac with eastward vector", "ux_true": "True zonal velocity", "uy_true": "True meridional velocity", "x_al": "Along track distance from beginning of cycle", "x_ac": "Across track distance from nadir", "u_ac_true": "True across track velocity", "u_al_true": "True along track velocity", "u_ac_instr": "Intrumental across track error", "u_al_instr": "Intrumental along track error", "u_al_wdrem": "Along track error in the wave doppler estimation", "u_ac_wdrem": "Across track error in the wave doppler estimation", "u_al_wd": "Along track wave doppler", "u_ac_wd": "Across track wave doppler", "u_ac_uss_obs": "Estimated across track Stokes drift", "u_al_uss_obs": "Estimated along track Stokes drift", "u_ac_uss_oi": "Error-free across track Stokes drift", "u_al_uss_oi": "Error-free along track Stokes drift", "u_ac_uss_true": "True across track Stokes drift", "u_al_uss_true": "True along track Stokes drift", "u_ac_dsigma":"Across track error due to dsigma", "u_al_dsigma":"Along track error due to dsigma", "uwnd": "Zonal wind at 10m", "vwnd": "Merdional wind at 10m", "rain": "Rain" } unit = {"ux_noerr": "m/s", "ux_obs": "m/s", "u_ac_obs": "m/s", "uy_noerr": "m/s", "uy_obs": "m/s", "u_al_obs": "m/s", "angle": "rad", "ux_model": "m/s", "uy_model": "m/s", "u_ac_noerr": "m/s", "u_al_noerr": "m/s", "u_ac_true": "m/s", "u_al_true": "m/s", "ux_true": "m/s", "uy_true": "m/s", "x_al": "km", "x_ac": "km", "u_ac_instr": "m/s", "u_al_instr": "m/s", "u_ac_wdrem": "m/s", "u_al_wdrem":"m/s", "u_ac_wd": "m/s", "u_al_wd": "m/s", "u_ac_uss_obs":"m/s", "u_al_uss_obs":"m/s", "u_ac_uss_oi": "m/s", "u_al_uss_oi":"m/s", "u_ac_uss_true":"m/s", "u_al_uss_true":"m/s", "u_ac_dsigma":"m/s", "u_al_dsigma":"m/s","uwnd":"m/s", "vwnd":"m/s", "rain":"mm/h" } for key, value in kwargs.items(): if value is not None: nvar = '{}'.format(key) var = fid.createVariable(nvar, 'f4', (dimtime, dimlat), fill_value=-1.36e9) try: var.units = unit[str(key)] except: var.units = '' try: var.long_name = longname[str(key)] except: var.long_name = str(key) if value.any(): mask = numpy.isnan(value) value[numpy.where(mask)] = -1.36e9 mask_ind = numpy.where(value < -1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value > 1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value == numpy.PINF) value[mask_ind] = -1.36e9 var[:, :] = value fid.close() def write_l2d(metadata, geolocation, **kwargs): ti = datetime.datetime.now() lat = geolocation['lat'] lon = geolocation['lon'] tim = geolocation['time'] # - Open Netcdf file in write mode fid = Dataset(metadata['file'], 'w', format='NETCDF4_CLASSIC') # - Create Global attribute global_attributes(fid, 'L2D') fid.description = "SKIM fixed grid" fid.time_coverage_start = metadata['time_coverage_start'] # p.date0+"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.time_coverage_end = metadata['time_coverage_end'] # p.date0 +"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.geospatial_lat_min = "{:.2f}".format(numpy.min(lat)) fid.geospatial_lat_max = "{:.2f}".format(numpy.max(lat)) fid.geospatial_lat_units = "degrees_north" _lon_max = numpy.mod(lon + 360, 360) fid.geospatial_lon_max = "{:.2f}".format(numpy.max(_lon_max)) fid.geospatial_lon_min = "{:.2f}".format(numpy.min(_lon_max)) fid.geospatial_lon_units = "degrees_east" # - Create dimensions # if (not os.path.isfile(self.file)): dimlon = 'lon' dimlat = 'lat' dimtime = 'time' nlon = numpy.shape(lon)[0] nlat = numpy.shape(lat)[0] ntime = None fid.createDimension(dimlat, nlat) fid.createDimension(dimlon, nlon) fid.createDimension(dimtime, ntime) # - Create and write Variables vtime = fid.createVariable('time', 'f8', (dimtime)) vtime.axis = "T" vtime.units = metadata['first_time'] vtime.long_name = "Time" vtime.standard_name = "time" vtime.calendar = "gregorian" vtime[:] = tim vlon = fid.createVariable('lon', 'f4', (dimlon, )) vlon.axis = "X" vlon.long_name = "Longitude" vlon.standard_name = "longitude" vlon.units = "degrees_east" vlon[:] = lon vlat = fid.createVariable('lat', 'f4', (dimlat, )) vlat.axis = "Y" vlat.long_name = "Latitude" vlat.standard_name = "latitude" vlat.units = "degrees_north" vlat[:] = lat longname = { "ux_noerr": "Error-free zonal velocity", "uy_noerr": "Error-free meridional velocity", "ux_obs": "Observed zonal velocity", "uy_obs": "Observed meridional velocity", "ux_model": "Error-free zonal velocity", "uy_model": "Error-free meridional velocity", "ux_true": "True zonal velocity", "uy_true": "True meridional velocity", } unit = {"ux_noerr": "m/s", "ux_obs": "m/s", "uy_noerr": "m/s", "uy_obs": "m/s", "ux_true": "m/s", "uy_true": "m/s" } for key, value in kwargs.items(): if value is not None: nvar = '{}'.format(key) var = fid.createVariable(nvar, 'f4', (dimtime, dimlat, dimlon), fill_value=-1.36e9) try: var.units = unit[str(key)] except: var.units = '' try: var.long_name = longname[str(key)] except: var.long_name = str(key) if value.any(): mask = numpy.isnan(value) value[numpy.where(mask)] = -1.36e9 mask_ind = numpy.where(value < -1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value > 1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value == numpy.PINF) value[mask_ind] = -1.36e9 var[0, :, :] = value fid.close() class Sat_SKIM(): ''' Sat_SKIM class: to read and write data that has been created by SKIM simulator ''' def __init__(self, ifile=None, lon=None, lat=None, lon_nadir=None, lat_nadir=None, time=None, cycle=None, al_cycle=None, x_al=None, timeshift=None): self.file = ifile self.lon = lon self.lat = lat self.lon_nadir = lon_nadir self.lat_nadir = lat_nadir self.time = time self.cycle = cycle self.x_al = x_al self.al_cycle = al_cycle self.timeshift = timeshift def write_swath(self, p, **kwargs): '''Write swath location in Satellite grid file sgridfile.\n Dimensions are time (i.e. along track), x_ac (across track) and cycle (1). \n Variables are longitude, latitude, number of days in a cycle, distance crossed in a cycle, time, along track and across track distances are stored.''' #grid_params_hash = skimulator.grid_check.get_b64_gzipped_hash(p) grid_params_hash = 0 #skimulator.grid_check.get_b64_gzipped_hash(p) # - Open Netcdf file in write mode fid = Dataset(self.file, 'w', format='NETCDF4_CLASSIC') # - Create Global attribute global_attributes(fid, 'swath') fid.description = "SKIM swath" fid.time_coverage_start = self.time[0][0] # p.date0+"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.time_coverage_end = self.time[-1][-1] # p.date0 +"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.geospatial_lat_min = "{:.2f}".format(numpy.min(self.lat[0])) fid.geospatial_lat_max = "{:.2f}".format(numpy.max(self.lat[0])) fid.geospatial_lat_units = "degrees_north" fid.geospatial_lon_max = "{:.2f}".format(numpy.max(self.lon[0])) fid.geospatial_lon_min = "{:.2f}".format(numpy.min(self.lon[0])) fid.geospatial_lon_units = "degrees_east" fid.cycle = "{0:d}".format(int(self.al_cycle)) fid.track = "{} th pass".format(self.ipass) fid.grid_params_hash = grid_params_hash # - Create dimensions # if (not os.path.isfile(self.file)): dimsample = 'sample' maxpos = numpy.argmax(numpy.array(p.list_shift)) nsample = numpy.shape(self.lon[maxpos + 1])[0] fid.createDimension(dimsample, nsample) # fid.createDimension('time_nadir', numpy.shape(self.lon)[0]) dimcycle = 'cycle' fid.createDimension(dimcycle, 1) nbeam = len(self.list_pos) dimnbeam = 'nbeam' fid.createDimension(dimnbeam, nbeam) # - Create and write Variables vtime = fid.createVariable('time', 'f8', (dimsample, dimnbeam)) vtime.axis = "T" vtime.units = "seconds since {}".format(p.first_time) vtime.long_name = "Time" vtime.standard_name = "time" vtime.calendar = "gregorian" vtime_nadir = fid.createVariable('time_nadir', 'f8', (dimsample,)) vtime_nadir.axis = "T" vtime_nadir.units = "seconds since {}".format(p.first_time) vtime_nadir.long_name = "Time at nadir" vtime_nadir.standard_name = "time" vtime_nadir.calendar = "gregorian" vlon = fid.createVariable('lon', 'f4', (dimsample, dimnbeam)) vlon.axis = "X" vlon.long_name = "Longitude" vlon.standard_name = "longitude" vlon.units = "degrees_east" vlon_nadir = fid.createVariable('lon_nadir', 'f4', (dimsample,)) vlon_nadir.axis = "X" vlon_nadir.long_name = "Longitude at nadir" vlon_nadir.standard_name = "longitude" vlon_nadir.units = "degrees_east" vlat = fid.createVariable('lat', 'f4', (dimsample, dimnbeam)) vlat.axis = "Y" vlat.long_name = "Latitude" vlat.standard_name = "latitude" vlat.units = "degrees_north" vlat_nadir = fid.createVariable('lat_nadir', 'f4', (dimsample,)) vlat_nadir.axis = "Y" vlat_nadir.long_name = "Latitude at nadir" vlat_nadir.standard_name = "latitude" vlat_nadir.units = "degrees_north" vx_al = fid.createVariable('x_al', 'f4', (dimsample, dimnbeam)) vx_al.units = "km" vx_al.long_name = "Along track distance from the nadir" # vx_al_tot = fid.createVariable('x_al_total', 'f4', (dimsample, # dimnbeam)) # vx_al_tot.units = "km" # vx_al_tot.long_name = "Along track distance from the beginning of "\ # "the pass projected on nadir" vx_ac = fid.createVariable('x_ac', 'f4', (dimsample, dimnbeam)) vx_ac.units = "km" vx_ac.long_name = "Across track distance from the nadir" vx_al_nadir = fid.createVariable('x_al_nadir', 'f4', (dimsample,)) vx_al_nadir.units = "km" vx_al_nadir.long_name = "Nadir along track distance from the"\ "beginning of the cycle" vangle = fid.createVariable('angle', 'f8', (dimsample, dimnbeam)) vangle.units = "rad" vangle.long_name = "Angle of the beam refered to the across track"\ " direction" vrangle = fid.createVariable('radial_angle', 'f8', (dimsample, dimnbeam)) vrangle.units = "rad" vrangle.long_name = "Radial angle refered to (longitude towards east,"\ " latitude toward north)" for i in range(nbeam + 1): if i == 0: vtime_nadir[:] = self.time[i][:nsample] vlon_nadir[:] = self.lon[i][:nsample] vlat_nadir[:] = self.lat[i][:nsample] vx_al_nadir[:] = self.x_al[i][:nsample] else: vtime[:, i - 1] = self.time[i][:nsample] vlon[:, i - 1] = self.lon[i][:nsample] vlat[:, i - 1] = self.lat[i][:nsample] vx_al[:, i - 1] = self.x_al[i][:nsample] vx_ac[:, i - 1] = self.x_ac[i][:nsample] vangle[:, i - 1] = self.beam_angle[i][:nsample] vrangle[:, i - 1] = self.radial_angle[i][:nsample] vcycle = fid.createVariable('cycle', 'f4', (dimcycle, )) valcycle = fid.createVariable('al_cycle', 'f4', (dimcycle, )) vtimeshift = fid.createVariable('timeshift', 'f4', (dimcycle, )) vcycle[:] = self.cycle vcycle.units = "second" vcycle.long_name = "seconds during a cycle" valcycle[:] = self.al_cycle valcycle.units = "km" valcycle.long_name = " Distance travelled during the pass" vtimeshift[:] = self.timeshift vtimeshift.units = "day" vtimeshift.long_name = "Shift time to match model time" vlistpos = fid.createVariable('beam_position', 'f4', (dimnbeam, )) vlistpos[:] = self.list_pos vlistpos.units = "" vlistpos.long_name = "Beam position" vlistangle = fid.createVariable('beam_angle', 'f8', (dimnbeam, )) vlistangle[:] = self.list_angle vlistangle.units = "" vlistangle.long_name = "Beam angle" vincl = fid.createVariable('inclination', 'f4', (dimsample, )) vincl.units = "rad" vincl.long_name = "Track inclination at nadir" # vincl[:] = numpy.array(self.angle)[0,:nsample] vincl[:] = self.angle[:nsample] fid.close() return None def write_data(self, p, outdata): # **kwargs): '''Write SKIM data in output file file_output Dimensions are x_al (along track distance), x_ac (across track distance). \n Variables are longitude, latitude, index (file number), error-free radial velocity (velocity interpolated from the model and projected with the radial angle), selected errors (instrument, uss bias, radial uss) and velocity with errors. \n ''' # - Open netcdf file in write mode fid = Dataset(self.file, 'w', format='NETCDF4_CLASSIC') global_attributes(fid, 'L2B') fid.description = "SKIM L2B from SKIMulator" try: fid.corresponding_grid = self.gridfile except: pass dateformat = '%Y-%m-%dT%H:%M:%SZ' time_model = datetime.datetime.strptime(p.first_time, '%Y-%m-%dT%H:%M:%SZ') mintime = numpy.nanmin(self.time) maxtime = numpy.nanmax(self.time) day = numpy.floor(mintime) seconds = (mintime - day) * 86400 time0 = time_model + datetime.timedelta(day, seconds) fid.time_coverage_start = time0.strftime(format=dateformat) # p.date0+"YYYY-MM-DDThh:mmZ" #tim0 converted to format day = numpy.floor(maxtime) seconds = (maxtime - day) * 86400 time0 = time_model + datetime.timedelta(day, seconds) fid.time_coverage_end = time0.strftime(format=dateformat) # p.date0 +"YYYY-MM-DDThh:mmZ" #tim0 converted to format fid.geospatial_lat_min = "{:.2f}".format(numpy.min(self.lat)) fid.geospatial_lat_max = "{:.2f}".format(numpy.max(self.lat)) fid.geospatial_lat_units = "degrees_north" fid.geospatial_lon_max = "{:.2f}".format(numpy.max(self.lon)) fid.geospatial_lon_min = "{:.2f}".format(numpy.min(self.lon)) fid.geospatial_lon_units = "degrees_east" fid.cycle = "{} th cycle".format(self.ncycle) fid.track = "{} th pass".format(self.ipass) dimsample = 'sample' fid.createDimension(dimsample, numpy.shape(self.lon[0])[0]) # fid.createDimension('time_nadir', numpy.shape(self.lon)[0]) dimcycle = 'cycle' fid.createDimension(dimcycle, 1) nbeam = len(p.list_pos) dimnbeam = 'nbeam' fid.createDimension(dimnbeam, nbeam) # - Create and write Variables vtime = fid.createVariable('time', 'f8', (dimsample, dimnbeam)) vtime.axis = "T" vtime.units = "days since {}".format(p.first_time) vtime.long_name = "Time" vtime.standard_name = "time" vtime.calendar = "gregorian" vtime_nadir = fid.createVariable('time_nadir', 'f8', (dimsample,)) vtime_nadir.axis = "T" vtime_nadir.units = "days since {}".format(p.first_time) vtime_nadir.long_name = "Time at nadir" vtime_nadir.standard_name = "time" vtime_nadir.calendar = "gregorian" vlon = fid.createVariable('lon', 'f4', (dimsample, dimnbeam)) vlon.axis = "X" vlon.long_name = "Longitude" vlon.standard_name = "longitude" vlon.units = "degrees_east" vlon_nadir = fid.createVariable('lon_nadir', 'f4', (dimsample,)) vlon_nadir.axis = "X" vlon_nadir.long_name = "Longitude at nadir" vlon_nadir.standard_name = "longitude" vlon_nadir.units = "degrees_east" vlat = fid.createVariable('lat', 'f4', (dimsample, dimnbeam)) vlat.axis = "Y" vlat.long_name = "Latitude" vlat.standard_name = "latitude" vlat.units = "degrees_north" vlat_nadir = fid.createVariable('lat_nadir', 'f4', (dimsample,)) vlat_nadir.axis = "Y" vlat_nadir.long_name = "Latitude at nadir" vlat_nadir.standard_name = "latitude" vlat_nadir.units = "degrees_north" for i in range(nbeam + 1): if i == 0: vtime_nadir[:] = self.time[i][:] vlon_nadir[:] = self.lon[i][:] vlat_nadir[:] = self.lat[i][:] else: vtime[:, i - 1] = self.time[i][:] vlon[:, i - 1] = self.lon[i][:] vlat[:, i - 1] = self.lat[i][:] longname = {"sigma0": "sigma0", "ur_true": "Radial velocity interpolated from model", "ucur": "Zonal velocity interpolated from model", "vcur": "Meridional velocity interpolated from model", "ur_obs": "Observed radial velocity (Ur_model+errors)", "index": "Equivalent model output number in list of file", "ur_uss": "Stokes drift radial velocity bias", "uwd": "True Radial Wave doppler componant", "uwd_est": "Estimated radial wave doppler componant", "uwnd": "Eastward wind at 10m ", "vwnd": "Northward wind at 10m ", "altimeter": "Nadir error", "wlv": "SSH interpolated from model", "ssh_obs": "Observed SSH", "ice": "Sea ice concentration", "uuss": "Meridional Stokes drift", "vuss": "Zonal Stokes drift", "instr": "instrumental error velocity", "radial_angle": "Azimuthal angle, trigonometric from East", "sigma0" : "NRCS", "ussr": "Radial Stokes drift", "ussr_est": "Reconstructed radial Stokes drift", "dsigma0": "Velocity due to the azimuthal sigma0 gradient", "dsigm-atm": "Velocity due to atmospheric sigma0 gradient", "yaw_oacs": "Velocity due to uncertainty in gyro", "yaw_ted": "Velocity due to TED", "yaw": "Total velocity due to mispointing", "yaw_rem": "Remaining yaw after offline correction", } unit = {"sigma0": "", "ur_true": "m/s", "ur_obs": "m/s", "index": " ", "ur_uss": "m/s", "uwnd": "m/s", "vwnd": "m/s", "uwd": "m/s", "ucur": "m/s", "vcur": "m/s", "ssh_obs": "m", "wlv": "m", "altimeter": "m", "ssh_obs":"m", "uwd_est": "m/s", "uuss": "m/s", "vuss": "m/s", "radial_angle": "rad", "ussr": "m/s", "ussr_est": "m/s", "dsigma0": "m/s", "disgm-atm":"m/s", "yaw_aocs": "m/s", "yaw_ted": "m/s", "yaw": "m/s", "yaw_rem": "m/s" } list_nadir_only = ("altimeter", ) list_nadir = ("ssh_true", "ssh_obs", "ice", "sigma0", "vindice", "mssu", "mssc", "uwnd", "vwnd") for key, value in outdata.items(): if numpy.shape(value)[1] > nbeam + 1: print(key, 'wrong value dimension', numpy.shape(value)) if value is not None: if key in list_nadir or key in list_nadir_only: if key not in list_nadir_only: nvar_nadir = '{}_nadir'.format(key) else: nvar_nadir = key var_nadir = fid.createVariable(nvar_nadir, 'f4', (dimsample, ), fill_value=-1.36e9) try: var_nadir.units = unit[str(key)] except: var_nadir.units = '' try: var_nadir.long_name = longname[str(key)] except: var_nadir.long_name = str(key) if key not in list_nadir_only: nvar = '{}'.format(key) if key == 'radial_angle': ntype = 'f8' else: ntype = 'f4' var = fid.createVariable(nvar, ntype, (dimsample, dimnbeam), fill_value=-1.36e9) try: var.units = unit[str(key)] except: var.units = '' try: var.long_name = longname[str(key)] except: var.long_name = str(key) mask = numpy.ma.getdata(numpy.isnan(value)) value[mask] = -1.36e9 mask_ind = numpy.where(value < -1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value > 1e7) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value == numpy.PINF) value[mask_ind] = -1.36e9 mask_ind = numpy.where(value == p.model_nan) value[mask_ind] = -1.36e9 if numpy.shape(value)[1] == 1: var_nadir[:] = value[:, 0] elif numpy.shape(value)[1] == nbeam: var[:] = value[:, :] elif numpy.shape(value)[1] == nbeam + 1: var[:] = value[:, 1:] if key in list_nadir: var_nadir[:] = value[:, 0] else: print(key, 'wrong value dimension', numpy.shape(value)) # TODO set range values fid.close() return None def load_swath(self, p, **kwargs): '''Load swath variables stored in Satellite grid file sgridfile. \n (longitude, latitude, number of days in a cycle, crossed distance during a cycle, time, along track and across track position).''' # - Open Netcdf file try: fid = Dataset(self.file, 'r') except IOError: logger.error('There was an error opening the file ' '{}'.format(self.file)) sys.exit(1) if 'grid_params_hash' in fid.ncattrs(): grid_params_hash = 0 #skimulator.grid_check.get_b64_gzipped_hash(p) #if fid.grid_params_hash != grid_params_hash: # raise IncompatibleGridError(self.file, fid.grid_params_hash, # grid_params_hash) # fid = Dataset(self.file, 'r') time = [] lon = [] lat = [] # cycle = [] # x_al = [] listvar = {'time': time, 'lon': lon, 'lat': lat, } self.lon = [] self.lat = [] self.time = [] # - Read variables in listvar and return them for stringvar in listvar: var = fid.variables['{}{}'.format(stringvar, '_nadir')] listvar[stringvar].append(numpy.array(var[:]).squeeze()) var = fid.variables[stringvar] for i in range(len(p.list_pos)): listvar[stringvar].append(numpy.array(var[:, i]).squeeze()) setattr(self, stringvar, listvar[stringvar]) # - Read variables in arguments for key, value in kwargs.items(): var = fid.variables[key] value = numpy.array(fid.variables[key][:]).squeeze() # value[value == var.fill_value] = numpy.nan setattr(self, key, value) self.pos = numpy.array(fid.variables['beam_position'][:]) _dpos = numpy.abs(self.pos - numpy.array(p.list_pos, dtype=numpy.float32)) if (numpy.all(_dpos) > 0.0001): logger.error('List of beam positions has changed,' ' reprocess the grids') sys.exit(1) self.angle = numpy.array(fid.variables['beam_angle'][:]) _dangle = numpy.abs(self.angle - numpy.array(p.list_angle, dtype=numpy.float32)) if (numpy.all(_dangle) > 0.0001): logger.error('List of beam angles has changed,' ' reprocess the grids') sys.exit(1) self.radial_angle = numpy.array(fid.variables['radial_angle'][:]) self.angle = numpy.array(fid.variables['angle'][:]) try: self.corresponding_grid = fid.corresponding_grid except AttributeError: pass try: self.incl = numpy.array(fid.variables['inclination'][:]).squeeze() except: logger.info('inclination variable not found') fid.close() return None def load_data(self, p, list_var, **kwargs): '''Load swath variables stored in Satellite grid file sgridfile. \n (longitude, latitude, number of days in a cycle, crossed distance during a cycle, time, along track and across track position).''' # - Open Netcdf file try: fid = Dataset(self.file, 'r') except IOError: logger.error('There was an error opening the file ' '{}'.format(self.file)) sys.exit(1) # fid = Dataset(self.file, 'r') time = [] lon = [] lat = [] # cycle = [] # x_al = [] #listvar = {'time': time, 'lon': lon, 'lat': lat, } #self.lon = [] #self.lat = [] #self.time = [] # - Read variables in listvar and return them #for stringvar in listvar: # var = fid.variables['{}{}'.format(stringvar, '_nadir')] # listvar[stringvar].append(numpy.array(var[:]).squeeze()) # var = fid.variables[stringvar] # for i in range(len(p.list_pos)): # listvar[stringvar].append(numpy.array(var[:, i]).squeeze()) # setattr(self, stringvar, listvar[stringvar]) # - Read variables in arguments dic = {} for key in list_var: var = fid.variables[key] value = numpy.array(fid.variables[key][:]).squeeze() # value[value == var.fill_value] = numpy.nan setattr(self, key, value) dic[key] = value for key, value in kwargs.items(): var = fid.variables[key] value = numpy.array(fid.variables[key][:]).squeeze() # value[value == var.fill_value] = numpy.nan setattr(self, key, value) dic[key] = value try: self.corresponding_grid = fid.corresponding_grid except AttributeError: pass fid.close() return dic class NETCDF_MODEL(): '''Class to read any netcdf data.\n USAGE is NETCDF_MODEL(file=name of file ,var= variable name, lon=variable longitude, lat=variable latitude, units=).\n Argument file is mandatory, arguments var, lon, lat are specified in params file. \n ''' def __init__(self, p, ifile=None, list_input_var=None, lon=('longitude', ), lat=('latitude', ), depth=0, time=0): if p.list_input_var is None: logger.error('Specify your list_input_var in parameter file') sys.exit(1) self.input_var_list = {'ucur': ['uo', ''], 'vcur': ['vo', ''], } else: self.input_var_list = p.list_input_var self.input_var = {} self.numgrid = {} self.nlon = list(lon) self.nlat = list(lat) self.nfile = ifile self.depth = depth self.time = time self.model_nan = getattr(p, 'model_nan', 0) p.model_nan = self.model_nan logger.debug('Nan Values {}, {}'.format(p.model_nan, self.model_nan)) def read_var(self, p, ind_lon=None, index=None): '''Read variables from netcdf file \n Argument is index=index to load part of the variable.''' ind_lon = None for key, value in self.input_var_list.items(): nfile0 = self.nfile[0] _nfile = '{}{}.nc'.format(nfile0, value[1]) if os.path.exists(_nfile): _tmp = read_var(_nfile, value[0], index=index, time=self.time, depth=self.depth, model_nan=self.model_nan) self.input_var[key] = _tmp if ind_lon is not None: self.input_var[key] = self.input_var[key][:, ind_lon] if len(value) == 3: self.numgrid[key] = value[2] else: self.numgrid[key] = 0 else: logger.info('{} not found'.format(_nfile)) # self.vvar[numpy.where(numpy.isnan(self.vvar))]=0 return None def read_coordinates(self, p, index=None): '''Read coordinates from netcdf file \n Argument is index=index to load part of the variable.''' self.vlon = {} self.ind_lon = {} self.vlat = {} for ikey in range(len(list(self.nfile))): ifile = self.nfile[ikey] if p.grid == 'regular': lon, lat = read_coordinates(ifile, self.nlon[ikey], self.nlat[ikey], twoD=False) lon = numpy.mod(lon + 360, 360) ind_lon = None # numpy.argsort(lon) else: lon, lat = read_coordinates(ifile, self.nlon[ikey], self.nlat[ikey]) self.vlat[ikey] = lat if ind_lon is not None: self.vlon[ikey] = lon[ind_lon] #numpy.mod(lon + 360, 360) self.ind_lon[ikey] = ind_lon else: self.vlon[ikey] = lon self.ind_lon[ikey] = None return None def calc_box(self, p): '''Calculate subdomain coordinates from netcdf file Return minimum, maximum longitude and minimum, maximum latitude''' self.read_coordinates(p) if (numpy.min(self.vlon[0]) < 1.) and (numpy.max(self.vlon[0]) > 359.): _ind = numpy.where(self.vlon[0] > 180.) self.vlon[0][_ind] = self.vlon[0][_ind] - 360 lon1 = (numpy.min(self.vlon[0]) + 360) % 360 lon2 = (numpy.max(self.vlon[0]) + 360) % 360 else: lon1 = numpy.min(self.vlon[0]) lon2 = numpy.max(self.vlon[0]) return [lon1, lon2, numpy.min(self.vlat[0]), numpy.max(self.vlat[0])] class WW3(): '''Class to read ww3 netcdf data.\n USAGE is NETCDF_MODEL(file=name of file ,var= variable name, lon=variable longitude, lat=variable latitude, units=).\n Argument file is mandatory, arguments var, lon, lat are specified in params file. \n ''' def __init__(self, p, ifile=None, list_input_var=None, lon=('longitude',), lat=('latitude',), depth=0, time=0): self.nlon = list(lon) self.nlat = list(lat) self.nfile = ifile self.depth = depth self.time = time if list_input_var is None: self.input_var_list = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 0], 'uuss': ['uuss', 'uss', 0], 'vuss': ['vuss', 'uss', 0], 'ice': ['ice', 'ice', 0], 'mssd': ['mssd', 'msd', 0], 'mssx': ['mssx', 'mss', 0], 'mssy':['mssy', 'mss', 0], 'ssh': ['wlv', 'wlv', 0], 'uwnd': ['uwnd', 'wnd', 0], 'vwnd': ['vwnd', 'wnd', 0]} else: self.input_var_list = p.list_input_var self.input_var = {} self.numgrid = {} self.model_nan = getattr(p, 'model_nan', 0.) p.model_nan = self.model_nan logger.debug('Nan Values {}, {}'.format(p.model_nan, self.model_nan)) def read_var(self, p, ind_lon=None, index=None): '''Read variables from netcdf file \n Argument is index=index to load part of the variable.''' for key, value in self.input_var_list.items(): nfile0 = self.nfile[0] _nfile = '{}{}.nc'.format(nfile0, value[1]) if os.path.exists(_nfile): _tmp = read_var(_nfile, value[0], index=index, time=self.time, depth=self.depth, model_nan=self.model_nan) self.input_var[key] = _tmp if len(value) == 3: self.numgrid[key] = value[2] else: self.numgrid[key] = 0 else: logger.info('{} not found'.format(_nfile)) return None def read_coordinates(self, p, index=None): '''Read coordinates from netcdf file \n Argument is index=index to load part of the variable.''' self.vlon = {} self.ind_lon = {} self.vlat = {} for ikey in range(len(list(self.nfile))): ifile = self.nfile[ikey] if p.grid == 'regular': lon, lat = read_coordinates(ifile, self.nlon[ikey], self.nlat[ikey], twoD=False) else: lon, lat = read_coordinates(ifile, self.nlon[ikey], self.nlat[ikey]) self.vlat[ikey] = lat self.ind_lon[ikey] = None self.vlon[ikey] = (lon + 360) % 360 return None def calc_box(self, p): '''Calculate subdomain coordinates from netcdf file Return minimum, maximum longitude and minimum, maximum latitude''' self.read_coordinates(p) if (numpy.min(self.vlon[0]) < 1.) and (numpy.max(self.vlon[0]) > 359.): _ind = numpy.where(self.vlon[0] > 180.) self.vlon[0][_ind] = self.vlon[0][_ind] - 360 lon1 = (numpy.min(self.vlon[0]) + 360) % 360 lon2 = (numpy.max(self.vlon[0]) + 360) % 360 else: lon1 = numpy.min(self.vlon[0]) lon2 = numpy.max(self.vlon[0]) return [lon1, lon2, numpy.min(self.vlat[0]), numpy.max(self.vlat[0])]
48,046
42.246625
83
py
skimulator
skimulator-master/skimulator/mod_tools.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ ''' Spectral and algebra tools for SKIM simulator. \n Contains the following functions: - load_python_file: load and parse parameter file \n - rotationmat3d: rotate data \n - spher2cart: convert spherical to cartesian coordinates \n - cart2spher: convert cartesian to spherical coordinates \n - cart2spher: convert cartesian to spherical coordinates for vectors \n - update_progress: Progress bar''' import numpy import math import logging import sys import os import types import datetime from . import const # Define logger level for debug purposes logger = logging.getLogger(__name__) def load_python_file(file_path): """Load a file and parse it as a Python module.""" if not os.path.exists(file_path): raise IOError('File not found: {}'.format(file_path)) full_path = os.path.abspath(file_path) python_filename = os.path.basename(full_path) module_name, _ = os.path.splitext(python_filename) module_dir = os.path.dirname(full_path) if module_dir not in sys.path: sys.path.append(module_dir) module = __import__(module_name, globals(), locals(), [], 0) return module def initialize_parameters(p): p.shift_lon = getattr(p, 'shift_lon', None) p.shift_time = getattr(p, 'shift_time', None) if p.shift_time is None: p.timeshift = 0 p.timeshift = getattr(p, 'timeshift', 0) if p.shift_time is None: p.timeshift = 0 model = getattr(p, 'model', 'NETCDF_MODEL') p.model = model p.model_nan = getattr(p, 'model_nan', 0) p.vel_factor = getattr(p, 'vel_factor', 1.) p.first_time = getattr(p, 'first_time', '2018-01-01T00:00:00Z') p.list_input_var = getattr(p, 'list_input_var', None) p.grid = getattr(p, 'grid', 'regular') p.cycle = getattr(p, 'cycle', 0.0368) p.order_orbit_col = getattr(p, 'order_orbit_col', None) p.satcycle = getattr(p, 'satcycle', None) p.sat_elev = getattr(p, 'sat_elev', None) p.ice_mask = getattr(p, 'ice_mask', True) p.vsat = getattr(p, 'vsat', const.vsat) # TODO Remove ? listo = ['wlv', 'ssh_obs', 'ur_true', 'ucur', 'vcur', 'uuss', 'vuss', 'radial_angle', 'vwnd', 'mssx', 'mssy', 'mssxy', 'uwb', 'vindice', 'ur_obs', 'mask', 'uwnd', 'sigma0', 'ice'] p.list_output = getattr(p, 'list_output', listo) p.snr_coeff = getattr(p, 'snr_coeff', 3e-3) p.nadir = getattr(p, 'nadir', False) p.ice = getattr(p, 'ice', False) p.proc_count = getattr(p, 'proc_number', 1) p.progress_bar = getattr(p, 'progress_bar', True) p.ac_threshold = getattr(p, 'ac_threshold', 20) # in km p.delta_al = getattr(p, 'delta_al', 2) #p.resol = getattr(p, 'resol_spatial_l2c', 40) # in km p.resol = getattr(p, 'resol', 40) # in km p.posting = getattr(p, 'posting_l2c', 5) # in km p.resol_spatial_l2d = getattr(p, 'resol_spatial_l2d', 50) # in km p.resol_temporal_l2d = getattr(p, 'resol_temporal_l2d', 8) # in days p.posting_l2d = getattr(p, 'posting_l2d', (0.1, 0.1)) # in degrees # Time domain: (start_time, end_time, dtime) in days p.time_domain = getattr(p, 'time_domain', (5, 25, 1)) # in days p.list_input_var_l2c = getattr(p, 'list_input_var_l2c', p.list_input_var) p.list_input_var_l2d = getattr(p, 'list_input_var_l2d', p.list_input_var) p.config_l2d = getattr(p, 'config_l2d', '') p.config_l2c = getattr(p, 'config_l2c', '') p.config = getattr(p, 'config', 'SKIM') p.rain = getattr(p, 'rain', False) p.rain_file = getattr(p, 'rain_file', None) p.rain_threshold = getattr(p, 'rain_threshold', 0.1) p.yaw_file = getattr(p, 'yaw_file', None) p.instr_configuration = getattr(p, 'instr_configuration', 'A') p.Rearth = const.Rearth p.type_satellite = getattr(p, 'type_satellite', 'skim') make_list_output(p) return None def make_list_output(p): #if p.rain is True: compulsory = ('ur_true', 'ur_obs', 'radial_angle', 'vindice', 'ur_obs') for key in compulsory: if key not in p.list_output: p.list_output.append(key) if p.nadir is True: compulsory = ('ssh_obs',) for key in compulsory: if key not in p.list_output: p.list_output.append(key) if 'Instrument' in p.noise: compulsory = ('instr', 'mssxy', 'sigma0') for key in compulsory: if key not in p.list_output: p.list_output.append(key) if 'WaveDoppler' in p.noise: compulsory = ('mssc', 'mssu') for key in compulsory: if key not in p.list_output: p.list_output.append(key) return None def check_path(p): if os.path.isdir(p.dir_setup) is False: logger.error('Data directory {} not found'.format(p.dir_setup)) sys.exit(1) if p.file_input is not None: if os.path.isdir(p.indatadir) is False: logger.error('Input directory {} not found'.format(p.indatadir)) sys.exit(1) if os.path.isdir(p.outdatadir) is False: logger.warn('Output directory {} did not exist and was ' 'created'.format(p.outdatadir)) os.makedirs(p.outdatadir) filesat_path = os.path.join(p.dir_setup, p.filesat) if os.path.isfile(filesat_path) is False: logger.error('Orbit file {} not found'.format(filesat_path)) sys.exit(1) if p.file_input is not None: if os.path.isfile(p.file_input) is False: logger.error('Model file list {} not found'.format(p.file_input)) sys.exit(1) return None def rotationmat3D(theta, axis): ''' Creates a rotation matrix: Slow method. \n Inputs are rotation angle theta and rotation axis axis. The rotation matrix correspond to a rotation of angle theta with respect to axis axis. \n Return the rotation matrix.''' # mat = numpy.eye(3, 3) axis = axis / math.sqrt(numpy.dot(axis, axis)) a = math.cos(theta/2.) b, c, d = -axis*math.sin(theta/2.) return numpy.array([[a*a + b*b - c*c - d*d, 2*(b*c - a*d), 2*(b*d + a*c)], [2*(b*c + a*d), a*a + c*c - b*b - d*d, 2*(c*d - a*b)], [2*(b*d - a*c), 2*(c*d + a*b), a*a + d*d - b*b - c*c]]) def sign(a): return (a > 0) - (a < 0) def spher2cart(lon, lat): ''' Convert spherical coordinates to cartesian coordinates.\n Inputs are longitude, latitude. \n Return x, y, z''' x = numpy.cos(lon*math.pi/180.) * numpy.cos(lat*math.pi/180.) y = numpy.sin(lon*math.pi/180.) * numpy.cos(lat*math.pi/180.) z = numpy.sin(lat*math.pi/180.) return x, y, z def cart2sphervect(x, y, z): ''' Convert cartesian coordinates to spherical coordinates. \n Inputs are cartiesian coordinates x, y, z. \n Return lon, lat. ''' norm = numpy.sqrt(x*x + y*y + z*z) lat = numpy.arcsin(z/norm) * 180./math.pi lon = numpy.arctan(y/x) % (2*math.pi) if (x < 0).any(): lon[x < 0] = (numpy.arctan(y[x < 0] / x[x < 0]) % (2*math.pi) + x[x < 0] / x[x < 0]*math.pi) lon = lon * 180/math.pi return lon % 360, lat def cart2spher(x, y, z): ''' Convert cartesian coordinates to spherical coordinates. \n Inputs are cartiesian coordinates x, y, z. \n Return lon, lat. ''' norm = numpy.sqrt(x*x + y*y + z*z) lat = numpy.arcsin(z/norm) * 180./math.pi if (x < 0): lon = (numpy.arctan(y/x) % (2*math.pi) + max(-numpy.sign(x), 0)*math.pi) else: lon = numpy.arctan(y/x) % (2*math.pi) lon = lon * 180/math.pi return lon % 360, lat def todict(p): result = {} for attr in dir(p): value = getattr(p, attr) if (isinstance(value, types.ModuleType) or isinstance(value, types.MethodType) or isinstance(value, types.FunctionType) or attr.startswith('__')): continue result[attr] = value return result def fromdict(result): p = type('skimparam', (object,), result) return p def update_progress_multiproc(status, info): """Creation of progress bar: print on screen progress of run, optimized for parrallelised tasks""" if info[3] is not None: # There has been an error return False pid = info[0] grid_name = info[1] if isinstance(grid_name, str): ipass = grid_name[-6:-3] else: ipass = '{:03d}'.format(grid_name) cycle = info[2] count = len(status.keys()) sys.stdout.write(_term_move_up() * count) now = datetime.datetime.now().strftime('%H:%M:%S') for pid in status: if grid_name in status[pid]['jobs']: if cycle is None: # Grid has been completely processed status[pid]['done'] += 1 status[pid]['extra'] = '{}|> pass: {} ....... DONE'.format( now, ipass) else: # Just update extra info status[pid]['extra'] = '{}|> pass: {}, cycle: {:04d}'.format( now, ipass, cycle) bar_size = 20 for pid, proc_state in status.items(): done = math.floor(bar_size * proc_state['done'] / proc_state['total']) todo = bar_size - done proc_elems = ['\n['] proc_elems.extend(['#'] * int(math.ceil(done))) proc_elems.extend([' '] * int(math.ceil(todo))) proc_elems.extend(['] {}'.format(proc_state['extra'])]) sys.stdout.write(''.join(proc_elems)) sys.stdout.flush() return True def update_progress(progress, arg1, arg2): '''Creation of a progress bar: print on screen the progress of the run''' barLength = 20 # Modify this to change the length of the progress bar status = "" if isinstance(progress, int): progress = float(progress) if not isinstance(progress, float): progress = 0 status = "error: progress var must be float\r\n" if progress < 0: progress = 0 status = "Halt...\r\n" if progress >= 1: progress = 1 status = "Done...\r\n" block = int(round(barLength*progress)) if arg1 and arg2: text = "\r[{0}] {1}%, {2}, {3}".format("#"*block + "-"*(barLength-block), "%.2f" % (progress*100), arg1 + ', ' + arg2, status) elif arg1: text = "\r[{0}] {1}%, {2}, {3}".format("#"*block + "-"*(barLength-block), "%.2f" % (progress*100), arg1, status) else: text = "\r[{0}] {1}%, {2} ".format("#"*block + "-"*(barLength-block), "%.2f" % (progress*100), status) sys.stdout.write(text) sys.stdout.flush() return progress def _term_move_up(): # pragma: no cover """Borrowed from https://github.com/tqdm/tqdm/blob/master/tqdm/_tqdm.py MIT 2016 (c) [PR #96] on behalf of Google Inc. MIT 2013 (c) Noam Yorav-Raphael, original author.""" colorama = None return '' if (os.name == 'nt') and (colorama is None) else '\x1b[A'
11,664
36.149682
134
py
skimulator
skimulator-master/skimulator/error/wet_troposphere.py
# Copyright (c) 2020 CNES/JPL # # All rights reserved. Use of this source code is governed by a # BSD-style license that can be found in the LICENSE file. """ Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """ Wet troposphere errors ---------------------- """ from typing import Dict, List, Tuple try: import numba as nb except ImportError: nb = None import numpy as np import scipy.ndimage.filters from . import utils #from .. import F_KA, VOLUMETRIC_MEAN_RADIUS, CELERITY, BASELINE def meshgrid(x: np.ndarray, y: np.ndarray) -> Tuple[np.ndarray, np.ndarray]: nx = x.size ny = y.size mx = np.empty((ny, nx)) my = np.empty((ny, nx)) for idx in range(ny): mx[idx, :] = x for idx in range(nx): my[:, idx] = y return mx, my def calculate_path_delay_lr(beam_positions: List[float], sigma: float, radio_r: np.ndarray, radio_l: np.ndarray, x_al: np.ndarray, x_ac_large: np.ndarray, wt_large: np.ndarray): beam_r = np.empty((x_al.shape[0], )) beam_l = np.empty((x_al.shape[0], )) # Find righ and leftacross track indices in the gaussian # footprint of 2.*p.sigma ind_r = x_ac_large + beam_positions[1] indac_r = np.where((ind_r < 2 * sigma) & (ind_r > -2 * sigma))[0] ind_l = x_ac_large + beam_positions[0] indac_l = np.where((ind_l < 2 * sigma) & (ind_l > -2 * sigma))[0] factor = 1 / (2 * np.pi * sigma**2) for idx, xal in enumerate(x_al): # Find along track indices in the gaussian footprint # of 2.*p.sigma delta_x_al = x_al - xal indal = np.where((delta_x_al <= (2 * sigma)) & (delta_x_al > (-2 * sigma)))[0] slice_al = slice(indal[0], indal[-1] + 1) slice_acr = slice(indac_r[0], indac_r[-1] + 1) slice_acl = slice(indac_l[0], indac_l[-1] + 1) x, y = _meshgrid(x_ac_large[slice_acr], x_al[slice_al] - xal) # Compute path delay on left and right gaussian footprint g = factor * np.exp(-(x**2 + y**2) / (2 * sigma**2)) beam_r[idx] = np.sum( g * wt_large[slice_al, slice_acr]) / np.sum(g) + radio_r[idx] x, y = _meshgrid(x_ac_large[slice_acl], x_al[slice_al] - xal) g = factor * np.exp(-(x**2 + y**2) / (2 * sigma**2)) beam_l[idx] = np.sum( g * wt_large[slice_al, slice_acl]) / np.sum(g) + radio_l[idx] return beam_r, beam_l def calculate_path_delay(sigma: float, radio: np.ndarray, x_al: np.ndarray, x_ac_large: np.ndarray, wt_large: np.ndarray): beam = np.empty((x_al.shape[0], )) # Find across track indices in the gaussian footprint of # 2. * sigma indac = np.where((x_ac_large < 2 * sigma) & (x_ac_large > -2 * sigma))[0] factor = 1 / (2 * np.pi * sigma**2) for idx, xal in enumerate(x_al): delta_x_al = x_al[:] - xal indal = np.where((delta_x_al <= (2 * sigma)) & (delta_x_al > (-2 * sigma)))[0] slice_al = slice(indal[0], indal[-1] + 1) slice_ac = slice(indac[0], indac[-1] + 1) x, y = _meshgrid(x_ac_large[slice_ac], x_al[slice_al] - xal) # Compute path delay on gaussian footprint g = factor * np.exp(-(x**2 + y**2) / (2 * sigma**2)) beam[idx] = np.sum( g * wt_large[slice_al, slice_ac]) / np.sum(g) + radio[idx] return beam class WetTroposphere: """Wet troposphere errors Args: parameters (settings.Parameters): Simulation settings """ ALPHA = 10 LC_MAX = 500 F_MAX = 0.05 def __init__(self, parameters: settings.Parameters) -> None: # Store the generation parameters of the random signal. self.beam_positions = parameters.beam_position self.delta_ac = parameters.delta_ac self.delta_al = parameters.delta_al self.len_repeat = parameters.len_repeat self.nbeam = parameters.nbeam self.nseed = parameters.nseed + 4 self.sigma = parameters.sigma # TODO self.conversion_factor = ( 1 / (F_KA * 2 * np.pi / CELERITY * BASELINE) * (1 + (parameters.height * 1e-3) / VOLUMETRIC_MEAN_RADIUS) * np.pi / 180 * 1e3) # Define power spectrum of error in path delay due to wet tropo freq = np.arange(1 / 3000, 1 / (2 * self.delta_al), 1 / 3000) # Global mean wet tropo power spectrum in cm**2/(cycle/km) # for L >= 100 km pswt = 3.156 * 1e-05 * freq**(-8 / 3) # Wet tropo power spectrum in cm**2/(cycle/km) for L < 100 km mask = freq > 1e-2 pswt[mask] = 1.4875 * 1e-4 * freq[mask]**(-2.33) self.pswt = pswt self.freq = freq self.fminx = 1 / self.len_repeat self.ps2d, self.f = utils.gen_ps2d(freq, pswt, fminx=self.fminx, fminy=1 / self.LC_MAX, fmax=self.F_MAX, alpha=self.ALPHA, lf_extpl=True, hf_extpl=True) def _radiometer_error(self, x_al: np.ndarray) -> Tuple[np.ndarray, np.ndarray]: # Define radiometer error power spectrum for a beam # High frequencies are cut to filter the associated error: # during the reconstruction of the wet trop signal psradio = 9.5 * 1e-5 * self.freq**-1.79 psradio[self.freq < 1e-3] = 9.5 * 1e-5 * (1e-3)**-1.79 mask = (self.freq > 0.0023) & (self.freq <= 0.0683) psradio[mask] = 0.036 * self.freq[mask]**-0.814 psradio[self.freq > 0.0683] = 0.32 # Compute random coefficients (1D) for the radiometer error # power spectrum for right and left beams hrad = utils.gen_signal_1d(self.freq, psradio, x_al, fmin=1 / self.len_repeat, fmax=1 / (2 * self.delta_al), alpha=10, nseed=self.nseed + 100, hf_extpl=True, lf_extpl=True) radio_r = hrad * 1e-2 hrad = utils.gen_signal_1d(self.freq, psradio, x_al, fmin=1 / self.len_repeat, fmax=1 / (2 * self.delta_al), alpha=10, nseed=self.nseed + 200, hf_extpl=True, lf_extpl=True) radio_l = hrad * 1e-2 return radio_r, radio_l def generate(self, x_al: np.array, x_ac: np.array) -> Dict[str, np.ndarray]: """ Generate wet troposphere errors Args: x_al (numpy.ndarray): Along track distance x_ac (numpy.ndarray): Across track distance Returns: dict: variable name and errors simulated. """ num_lines = x_al.shape[0] num_pixels = x_ac.shape[0] # Initialization of radiometer error in right and left beam radio_r, radio_l = self._radiometer_error(x_al) # Initialization of swath matrices and large swath matrices (which # include wet tropo data around the nadir and outside the swath) start_x = -2 * self.sigma / self.delta_ac + x_ac[0] stop_x = (2 * self.sigma / self.delta_ac + x_ac[-1] + self.delta_ac) # x_ac_large and wt_large are necessary to compute the gaussian # footprint of a beam on the nadir or near the edge of the swath x_ac_large = np.arange(start_x, stop_x, self.delta_ac) naclarge = np.shape(x_ac_large)[0] # Compute path delay error due to wet tropo and radiometer error # using random coefficient initialized with power spectrums wt = utils.gen_signal_2d_rectangle(self.ps2d, self.f, x_al, x_ac, fminx=self.fminx, fminy=1 / self.LC_MAX, fmax=self.F_MAX, alpha=self.ALPHA, nseed=self.nseed) wt = wt.T * 1e-2 wt_large = utils.gen_signal_2d_rectangle(self.ps2d, self.f, x_al, x_ac_large, fminx=self.fminx, fminy=1 / self.LC_MAX, fmax=self.F_MAX, alpha=self.ALPHA, nseed=self.nseed) wt_large = wt_large.T * 1e-2 # Compute Residual path delay error after a 1-beam radiometer # correction if self.nbeam == 1: beam = _calculate_path_delay(self.sigma, radio_l, x_al, x_ac_large, wt_large) beam = scipy.ndimage.filters.gaussian_filter( beam, 30. / self.delta_al) beam2d = np.vstack(num_pixels * (beam, )).T # Compute residual path delay wet_tropo = wt - beam2d wet_tropo_nadir = wt_large[:, naclarge // 2] - beam # Compute Residual path delay error after a 2-beams radiometer # correction elif self.nbeam == 2: beam_r, beam_l = _calculate_path_delay_lr(self.beam_positions, self.sigma, radio_r, radio_l, x_al, x_ac_large, wt_large) # Filtering beam signal to cut frequencies higher than 125 km beam_r = scipy.ndimage.filters.gaussian_filter( beam_r, 30 / self.delta_al) beam_l = scipy.ndimage.filters.gaussian_filter( beam_l, 30 / self.delta_al) # Compute residual path delay (linear combination of left # and right path delay) polyfit = np.polynomial.polynomial.polyfit pol = polyfit([self.beam_positions[0], self.beam_positions[1]], [beam_l, beam_r], 1) beam = (np.array(num_pixels * [pol[0]]).T + np.array(num_lines * [x_ac]) * np.array(num_pixels * [pol[1]]).T) wet_tropo = wt - beam wet_tropo_nadir = wt_large[:, naclarge // 2] - beam[:, num_pixels // 2] else: raise ValueError("nbeam must be in [1, 2]") # wt_nadir = wt_large[:, naclarge // 2] return { "wet_troposphere": wet_tropo, "wet_troposphere_nadir": wet_tropo_nadir } if nb is not None: _calculate_path_delay = nb.njit(cache=True)(calculate_path_delay) _calculate_path_delay_lr = nb.njit(cache=True)(calculate_path_delay_lr) _meshgrid = nb.njit(cache=True)(meshgrid) else: _calculate_path_delay = calculate_path_delay _calculate_path_delay_lr = calculate_path_delay_lr _meshgrid = meshgrid
12,414
41.958478
79
py
skimulator
skimulator-master/skimulator/error/rain.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import sys import numpy import pickle from typing import Dict, Tuple from scipy import interpolate from . import utils import logging logger = logging.getLogger(__name__) def compute_rain(time: float, sx_al: numpy.ndarray, sx_al_nadir: numpy.ndarray, sx_ac: numpy.ndarray, dic: dict, nseed: int )-> Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray, numpy.ndarray]: hour = int((time - numpy.floor(time))*24) size_dic = len(dic['xal'][hour]) numpy.random.seed(nseed) rr_ind = int(numpy.random.random_sample() * size_dic) xal = dic['xal'][hour][rr_ind] var = dic['rr'][hour][rr_ind] var2 = dic['pia'][hour][rr_ind] xac = dic['xac'][hour][rr_ind] x_al_g_tot = + sgrid.x_al for i in range(numpy.shape(x_al)[1]): x_al_g_tot[:, i] = sx_al[:, i] + sx_al_nadir xal_g = numpy.mod(x_al_g_tot - numpy.min(x_al_g_tot), numpy.max(xal)) interp = interpolate.RectBivariateSpline _Teval = interp(xal, xac, numpy.isnan(var), kx=1, ky=1, s=0) Teval = _Teval.ev(xal_g, sx_ac) # Trick to avoid nan in interpolation var_mask = + var var_mask[numpy.isnan(var_mask)] = 0. var2_mask = + var2 var2_mask[numpy.isnan(var2_mask)] = 0. # Interpolate variable _var_out = interp(xal, xac, var_mask, kx=1, ky=1, s=0) var_out = _var_out.ev(xal_g, sx_ac) _var2_out = interp(xal, xac, var2_mask, kx=1, ky=1, s=0) var2_out = _var2_out.ev(xal_g, sx_ac) # Mask variable with Teval var2_out[Teval > 0] = numpy.nan var_out[Teval > 0] = numpy.nan xal_n = numpy.mod(sx_al_nadir - numpy.min(sx_al_nadir), numpy.max(xal)) var_nad = numpy.zeros(numpy.shape(xal_n)) return var_out, var_nad, var2_out, var_nad def load_rain(rain_file: str) -> dict: with open(rain_file, 'rb') as frain: dic = pickle.load(frain) return dic class Rain(): '''Class that simulate random rain and pia from pyo scenes for all beams Args: parameters (nseed, list_angle, rain_file, vsat)''' def __init__(self, parameters): if parameters.rain_file is not None: self.rain_dic = load_rain(parameters.rain_file) else: self.rain_dic = {} self.nseed = parameters.nseed + 3 self.beam_angle = parameters.list_angle self.nbeam = len(self.beam_angle) self.vsat = parameters.vsat def generate(self, time: numpy.ndarray, x_al: numpy.ndarray, x_al_nadir: numpy.ndarray, x_ac: numpy.ndarray, ac_angle: numpy.ndarray) -> Dict[str, numpy.ndarray]: ''' Generate rain and gpia for an ensemble of beams as well as nadir Args: time: time # TODO check format x_al: along track distance from nadir x_al_nadir: along track distance since beginning of pass x_ac: across track distance from nadir ac_angle: Angle from the across track direction ''' mean_time = numpy.mean(time) rain_all = [] gpia_err_all = [] if self.rain_dic: rain, rain_nad, gpia, gpia_nad = compute_rain(mean_time, x_al, x_al_nadir, x_ac, self.rain_dic, self.nseed) rain_all.append(rain_nad) gpia_err_all.append(gpia_nad) for i in range(self.nbeam): beam_angle = self.list_angle[i] gpia_err = utils.convert_dbkm2ms(gpia, ac_angle[i], self.list_angle[i], self.vsat) rain_all.append(rain[:, i]) gpia_err_all.append(gpia_err) rain_all = numpy.transpose(numpy.array(rain_all)) gpia_err_all = numpy.transpose(numpy.array(gpia_err_all)) return {'rain': rain_all, 'gsig_atm_err': gpia_err_all} else: return {} def mask(var: numpy.ndarray, rain: numpy.ndarray, gpia: numpy.ndarray, rain_thresh: float) -> numpy.ndarray: gpia_thresh = 1 var[rain > rain_thresh] = numpy.nan var[abs(gpia) > gpia_thresh] = numpy.nan return var
4,964
38.094488
79
py
skimulator
skimulator-master/skimulator/error/instrument.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import numpy import logging from typing import Dict logger = logging.getLogger(__name__) def instr_stream(radial_angle: numpy.ndarray, ac_angle: numpy.ndarray, wind_norm: numpy.ndarray, wind_dir:numpy.ndarray, cycle: float)-> numpy.ndarray: # - Compute NRCS from wind: diff_angle = radial_angle - wind_dir nrcs = (-47.5 + 17.5 * (wind_norm - 1)**0.3 - numpy.sqrt(2)/2 *(1 - abs(numpy.cos(2 * diff_angle))) * (4 + 0.1* wind_norm)) # NRCS at 0db is -29 db: snr = nrcs + 29 instr = numpy.full(numpy.shape(wind_norm), numpy.nan) # Decifer formulae depending on SNR snr2 = + snr snr[snr==0] = 1 N = 24 / (snr * 3.7 + 12) al_angle = ac_angle + numpy.pi/2 instr = (abs(numpy.sin(al_angle)) * (10 + 4*numpy.tanh(snr/6)) + (4 + 12 * (1 - numpy.tanh(snr / 6))) * abs(numpy.sin(al_angle))**N + 17 * (1 - numpy.tanh(snr / 6)))* 10**(-2) instr[snr>19] = (14.5 * abs(numpy.sin(al_angle[snr>19])) + 4 * abs(numpy.sin(al_angle[snr>19]))**0.2 + 0.4) * 10**(-2) instr[snr2==0] = (abs(numpy.sin(al_angle[snr2==0])*10) + 16 * abs(numpy.sin(al_angle[snr2==0]))**2 + 17) * 10**(-2) # - Knowledge of HH on top of VV will decrease error (about 20%) # Formulaes have been computed for 64 pulses, rms_instr factor will adapt # noise if number of pulses changes (sqrt(factor_of_pulses)) instr = 0.8 * instr * numpy.sqrt(64/12000/2/cycle) return instr def instr_skim_old(file_rms_instr: str, radial_angle: numpy.ndarray, rms_instr_factor:float): # - Load instrumental rms file into dictionnary theta, rms = numpy.loadtxt(file_rms_instr, usecols=(0, 1), unpack=True) dic_rms = {} with open(file_rms_instr) as f: for line in f: (key, val) = line.split() dic_rms[int(key)] = float(val) # - Compute rms corresponding to each radial angle # Refer radial angle to along track direction radial_angle_along = numpy.degrees((radial_angle + math.pi / 2) % math.pi) # Interpolate radial_angle with theta values step = theta[1]-theta[0] if((theta[1:]-theta[:-1]) != step).any(): logger.error('instrumental rms file noise has not a constant step') sys.exit(1) thetaprev = numpy.floor(radial_angle_along/step) fac = radial_angle_along - thetaprev rms_theta = radial_angle * 0 i = 0 for itheta, ifac in zip(thetaprev, fac): keytheta = list(dic_rms.keys())[int(itheta)] keythetanext = list(dic_rms.keys())[int(itheta + step) % 180] rms_theta[i] = (dic_rms[keytheta] * ifac / step + (step - ifac) / step * dic_rms[keythetanext]) i += 1 # - Errors array are the same size as the swath size instr = rms_theta * rms_instr_factor * 10**(-2) return instr def instr_skim(beam_angle:float, ac_angle: numpy.ndarray, sigma0: numpy.array, cycle: float, instr_configuration: str)-> numpy.ndarray: if beam_angle == 12: if instr_configuration == 'A': _co = (-15.998, 0.657, -4.174, 15.260) if instr_configuration == 'B': _co = (16.409, 0.963, -6.232, 2.823) coeff = _co[0] * numpy.sin(ac_angle * _co[1] + _co[2])**2 + _co[3] sigma_ref = 10**(0.66) elif beam_angle == 6: if instr_configuration == 'A': _co = (17.272, -1.105, -2.977, 6.108) if instr_configuration == 'B': _co = (12.299, 1.667, -1.045, 14.071) coeff = _co[0] * numpy.sin(ac_angle * _co[1] + _co[2]) + _co[3] #igma_ref = 9.9 sigma_ref = 10**(0.99) else: logger.error('Unknown instrumental parametrisation for {}' ' angle'.format(beam_angle)) coeff_random = (coeff * 10**(-2) * sigma0 / sigma_ref) if cycle < 0.0368: coeff_random = coeff_random * numpy.sqrt(0.0368/cycle) coeff_random[sigma0 < 10**-5] = numpy.nan return coeff_random class Instrument(): '''Class instrument defines the instrumental error for STREAM or SKIM concepts Args: parameters: Simulation settings (nseed, type_satellite, instr_configuration and cycle parameters) ''' def __init__(self, parameters): self.nseed = parameters.nseed + 1 self.type_satellite = parameters.type_satellite self.cycle = parameters.cycle self.instr_configuration = parameters.instr_configuration self.beam_angle = parameters.list_angle def generate(self, radial_angle: numpy.array, angle: numpy.array, wind_norm: numpy.array, wind_dir: numpy.array, sigma0: numpy.array) -> Dict[str, numpy.ndarray]: ''' Build errors corresponding to the instrument on the satellite swath. Args: radial_angle: Angle refered to the East angle: Angle refered to the across track direction wind_norm: wind velocity wind_dir: wind direction refered to the East ''' instr_all = [] for b in range(len(self.beam_angle)): if self.type_satellite == 'stream': instr = instr_stream(radial_angle[:, b], angle[:, b], wind_norm[:, b], wind_dir[:, b], self.cycle) elif self.type_satellite == 'skim': instr = instr_skim(self.beam_angle[b], angle[:, b], sigma0[:, b], self.cycle, self.instr_configuration) else: logger.error(f'Unknown concept {self.type_satellite}') sys.exit(1) # Seed to make reproducible noise numpy.random.seed(self.nseed) error = numpy.random.normal(0.0, abs(instr), numpy.shape(instr)[0]) instr_all.append(error) return{"instr": numpy.transpose(numpy.array(instr_all))}
6,800
39.96988
79
py
skimulator
skimulator-master/skimulator/error/dsigma.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """ Error due to Sigma gradient within a beam simulation""" import sys import numpy from typing import Dict, Tuple from . import utils import logging logger = logging.getLogger(__name__) def compute_pdf_dsigma(beam_angle: float)-> Tuple[float, float]: # Compute sigma for pdf altika inc = numpy.deg2rad(beam_angle) sig_altika = numpy.arange(7, 15) siglin = 10.**(sig_altika / 10) R = 0.55 mss = R**2 / siglin dmssdsig0 = -R**2 / siglin**2 xe = numpy.linspace(-0.02, 0.02, 1001) dxe = xe[1] - xe[0] # Compute sum of pdfs a = 19.5 nu = 4.635 * numpy.exp(-0.03 * (sig_altika - 12)**2) nu_m = numpy.array([nu, ]* len(xe)) c = 0.000045 / (10**((sig_altika + 1.5)**2 / 100)) + 4.9e-7 * sig_altika c_m = numpy.array([c, ]*len(xe)) xe_m = numpy.array([xe, ] * len(sig_altika)).transpose() sigbeam = (R**2.*numpy.exp(-(numpy.tan(inc)**2 / mss)) / (mss * numpy.cos(inc)**4)) dmss = 0.0001 dsigdbeam = (R**2*numpy.exp(-(numpy.tan(inc)**2./(mss + dmss))) / ((mss + dmss)*numpy.cos(inc)**4) - sigbeam) / dmss stretched_fac = siglin * dmssdsig0 * dsigdbeam / sigbeam stretched_fac_m = numpy.array([stretched_fac, ]*len(xe)) dxe = xe[1] - xe[0] mat_pdf = 10**((a - (3 + nu_m) * numpy.log(1 + xe_m**2 / ((1 + nu_m) * c_m)))/10) sum_pdf = numpy.array([numpy.sum(mat_pdf, axis=1), ] * len(mss)).transpose() norm_pdf0 = mat_pdf / (sum_pdf) # * dxe) stretched_pdf = abs(1/stretched_fac)*10**((a - (3 + nu_m) * numpy.log(1 + (xe_m/stretched_fac)**2 / ((1 + nu_m) * c_m)))/10) #/sum_pdf return xe, stretched_pdf class DSigma(): '''Class that simulates variation of sigma within a footprint. It is derived from Altika observation Args: parameters (nseed, sat_elev) ''' def __init__(self, parameters): self.nseed = parameters.nseed self.sat_elev = parameters.sat_elev / 10**3 self.beam_angle = parameters.list_angle self.vsat = parameters.vsat def generate(self, sigma0: numpy.ndarray, ac_angle: numpy.ndarray)-> Dict[str, numpy.ndarray]: '''Generate statistical dsigma error within a footprint for one beam Args: beam_angle: angle view from beam sigma0: sigma0 for one beam ac_angle: angle from across track direction ''' dsigma_all = [] for b in range(len(self.beam_angle)): xe, pdf = compute_pdf_dsigma(self.beam_angle[b]) inc = numpy.deg2rad(self.beam_angle[b]) # ratio altika - skim r = 0.66 cshape = numpy.shape(sigma0[:, b]) dsigma = numpy.full(cshape, numpy.nan) # to change, very slow for i in range(cshape[0]): if not numpy.isfinite(sigma0[i, b]): continue if numpy.isnan(sigma0[i, b]): continue siglog = 10 * numpy.log10(sigma0[i, b]) ind_sig = numpy.floor(siglog) - 7 ind_sig = max(ind_sig, 7) ind_sig = int(min(ind_sig, 0)) pdf_stretched = pdf[:, ind_sig] distribution = pdf_stretched / numpy.sum(pdf_stretched) try: res = numpy.random.choice(xe, size=1, p=distribution) dsigma[i] = res[0] except: dsigma[i] = numpy.nan dsigma = (r * dsigma / (10*numpy.log(10)) * self.sat_elev * numpy.sin(inc)) dsigma = utils.convert_dbkm2ms(dsigma, ac_angle[:, b], self.beam_angle[b], self.vsat) dsigma_all.append(dsigma) return {"dsigma": numpy.transpose(numpy.array(dsigma_all))}
4,605
36.754098
80
py
skimulator
skimulator-master/skimulator/error/utils.py
# Copyright (c) 2020 CNES/JPL # # All rights reserved. Use of this source code is governed by a # BSD-style license that can be found in the LICENSE file. """ Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """ Random signal generation utilities ---------------------------------- """ from typing import Optional, Tuple import warnings try: import numba as nb except ImportError: nb = None import numpy import scipy.interpolate try: import mkl_fft IFFT = mkl_fft.ifft IFFT2 = mkl_fft.ifft2 except ImportError: IFFT = numpy.fft.ifft IFFT2 = numpy.fft.ifft2 def gen_signal_1d(fi: numpy.ndarray, psi: numpy.ndarray, x: numpy.ndarray, nseed: int = 0, fmin: Optional[float] = None, fmax: Optional[float] = None, alpha: int = 10, lf_extpl: bool = False, hf_extpl: bool = False) -> numpy.ndarray: """Generate 1d random signal using Fourier coefficient""" # Make sure fi, PSi does not contain the zero frequency: psi = psi[fi > 0] fi = fi[fi > 0] # Adjust fmin and fmax to fi bounds if not specified fmin = fmin or fi[0] fmax = fmax or fi[-1] # Go alpha times further in frequency to avoid interpolation aliasing. fmaxr = alpha * fmax # Interpolation of the non-zero part of the spectrum f = numpy.arange(fmin, fmaxr + fmin, fmin) mask = psi > 0 ps = numpy.exp(numpy.interp(numpy.log(f), numpy.log(fi[mask]), numpy.log(psi[mask]))) # lf_extpl=True prolongates the PSi as a plateau below min(fi). # Otherwise, we consider zeros values. same for hf ps[f < fi[0]] = psi[0] if lf_extpl else 0 ps[f > fi[-1]] = psi[-1] if hf_extpl else 0 ps[f > fmax] = 0 # Detect the sections (if any) where PSi==0 and apply it to PS mask = numpy.interp(f, fi, psi) ps[mask == 0] = 0 f_size = f.size phase = numpy.empty((2 * f_size + 1)) numpy.random.seed(nseed) phase[1:(f_size + 1)] = numpy.random.random(f_size) * 2 * numpy.pi phase[0] = 0 phase[-f_size:] = -phase[1:(f_size + 1)][::-1] fft1a = numpy.concatenate((numpy.array([0]), 0.5 * ps, 0.5 * ps[::-1]), axis=0) fft1a = numpy.sqrt(fft1a) * numpy.exp(1j * phase) / fmin**0.5 yg = 2 * fmaxr * numpy.real(IFFT(fft1a)) xg = numpy.linspace(0, 0.5 / fmaxr * yg.shape[0], yg.shape[0]) return numpy.interp(numpy.mod(x, xg.max()), xg, yg) def calculate_ps2d(f: numpy.ndarray, f2: numpy.ndarray, ps1d: numpy.ndarray, dfx: numpy.ndarray, dfy: numpy.ndarray) -> numpy.ndarray: result = numpy.zeros(f2.shape) view = result.ravel() dfx_2 = dfx * 0.5 dfx_y = dfx * dfy for idx in range(-1, -f.size - 1, -1): item = f[idx] mask = (f2 >= (item - dfx_2)) & (f2 < (item + dfx_2)) amount = numpy.sum(result[:, idx]) * dfx_y miss = ps1d[idx] * dfx - amount view[mask.ravel()] = 0 if miss <= 0 else miss * 0.5 / dfx_y return result def calculate_signal(rectangle, x, y, xgmax, ygmax): result = numpy.zeros((len(y), len(x))) xn = (x.max() - x[0]) // xgmax yn = (y.max() - y[0]) // ygmax dx = x - x[0] dy = y - y[0] for ix_n in range(int(xn + 1)): ix0 = numpy.where((dx >= (ix_n * xgmax)) & (dx < ((ix_n + 1) * xgmax)))[0] for iy_n in range(int(yn + 1)): iy0 = numpy.where((dy >= (iy_n * ygmax)) & (dy < ((iy_n + 1) * ygmax)))[0] result[iy0[0]:iy0[-1] + 1, ix0[0]:ix0[-1] + 1] = rectangle[:len(iy0), :len(ix0)] return result def convert_dbkm2ms(var_out: numpy.array, ac_angle: numpy.array, vsat: float, beam_angle: float)-> numpy.ndarray: # Computation of lobe antenna th3 = numpy.deg2rad(0.65) an = th3 / numpy.sqrt(4 * numpy.log(2)) # Computation of mispointing delta = an**2 / numpy.sin(numpy.deg2rad(beam_angle))**2 * var_out out = delta * vsat * numpy.cos(ac_angle) return out def proj_radial(u: numpy.array, v: numpy.array, radial_angle: numpy.array ) -> numpy.ndarray: ur = u * numpy.cos(radial_angle) + v * numpy.sin(radial_angle) return ur def dist_sphere(lon1, lon2, lat1, lat2, R): radlat1 = numpy.deg2rad(lat1) radlat2 = numpy.deg2rad(lat2) raddifflon = numpy.deg2rad(lon2 - lon1) d = numpy.arccos(numpy.sin(radlat2) * numpy.sin(radlat1) + numpy.cos(radlat2) * numpy.cos(radlat1) * numpy.cos(raddifflon)) return d * R / 1000. # in km def cross_product(mat: numpy.array, ncoeff: int, cshape: numpy.array )-> numpy.ndarray: ncross = sum(range(ncoeff + 1)) + ncoeff cross = numpy.full((cshape, ncross), numpy.nan) cross[:, 0:ncoeff] = mat[:, :ncoeff] niter = ncoeff for k in range(ncoeff): for l in range(ncoeff - k): cross[:, niter] = mat[:, k]*mat[:, k + l] niter += 1 return cross def reconstruct_var(ncoeff, cshape, f1, b1, cross, xlabel): shape_f1 = numpy.shape(f1)[0] shape_b1 = numpy.shape(b1)[0] #f1 = f1.reshape(int(shape_f1//float(shape_b1)), shape_f1) proba = numpy.exp(numpy.dot(cross, f1) + b1) proba_tot = numpy.tile(numpy.nansum(proba, axis=1), (shape_b1, 1)) proba = proba / numpy.transpose(proba_tot) var = numpy.sum(proba * numpy.tile(xlabel, (cshape, 1)),axis=1) b1 = numpy.tile(b1, (1, 1)) train_weight = numpy.concatenate((f1,b1),axis=0) l_train = numpy.zeros((cshape, ncoeff + 1)).astype('float64') l_train[:, : ncoeff] = cross l_train[:, ncoeff] = 1. res = numpy.dot(l_train, train_weight) xx2D = numpy.tile(xlabel, (cshape, 1)) var = (numpy.sum(xx2D * numpy.exp(res), axis=1) / numpy.sum(numpy.exp(res),axis=1)) return var if nb is not None: _calculate_ps2d = nb.njit("(float64[:, ::1]) (float64[::1]," "float64[:, ::1], float64[::1], float64," "float64)", cache=True, nogil=True)(calculate_ps2d) _calculate_signal = nb.njit("(float64[:, ::1]) (float64[:, ::1]," "float64[::1], float64[::1], float64," "float64)", cache=True, nogil=True)(calculate_signal) else: _calculate_ps2d = calculate_ps2d _calculate_signal = calculate_signal
7,169
34.147059
76
py
skimulator
skimulator-master/skimulator/error/altimeter.py
# Copyright (c) 2020 CNES/JPL # # All rights reserved. Use of this source code is governed by a # BSD-style license that can be found in the LICENSE file. """ Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """ Altimeter instrument error -------------------------- """ from typing import Dict import numpy from . import utils class Altimeter: """Altimeter instrument error. Args: parameters (settings.Parameters): Simulation settings """ def __init__(self, parameters) -> None: # Store the generation parameters of the random signal. self.delta_al = 2 * parameters.delta_al self.nseed = parameters.nseed + 4 self.len_repeat = parameters.len_repeat # Define the sepctrum of the nadir instrument error freq = numpy.arange(1 / 3000, 1 / self.delta_al, 1 / 3000) psd = 8 + 1.05 * 1e-4 * freq**(-2.2) psd[freq < 0.00023627939582672978] = 1e4 # Convert spectrum in m2/cy self.psd = psd * 1e-4 self.freq = freq def generate(self, x_al: numpy.array) -> Dict[str, numpy.ndarray]: """Generate altimeter instrument error. Args: x_al (numpy.ndarray): Along track distance Returns: dict: variable name and errors simulated. """ # Compute random noise of 10**2 cm**2/(km/cycle) # Compute the correspond error on the nadir in m error = utils.gen_signal_1d(self.freq, self.psd, x_al, nseed=self.nseed, fmin=1 / self.len_repeat, fmax=1 / self.delta_al, alpha=10) return {"altimeter": error.reshape(error.shape[0], -1)}
2,323
33.686567
79
py
skimulator
skimulator-master/skimulator/error/simulate_spectrum.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import numpy import math import logging from typing import Dict, Tuple import pkg_resources from . import utils # Define logger level for debug purposes logger = logging.getLogger(__name__) def load_mss_noise() -> Dict[str, numpy.array]: mss_path = pkg_resources.resource_filename('skimulator', 'share/noise_pdf_mss1d.npy') mssr_noise = numpy.load(mss_path, allow_pickle=True)[()] return mssr_noise def combine_usr(lon: numpy.array, lat: numpy.array, usr: numpy.array, mssx: numpy.array, mssy: numpy.array, mssxy: numpy.array, dazi: float, angle: numpy.array, incl: numpy.array, wnd_dir: numpy.array, nseed: int=1, Rearth: float=6378. * 10**3 ) -> Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]: ''' PURPOSE : recombine measurements of various azimuths and ambiguity removal using wind direction Provides a single estimate (not applicable to a set of points) INPUT : - usrazi : radial Stokes drift for each azimuth in azibin [m/s], len = nazi dazi : azimuth increment [degrees] azimuthat : along track azimuth [radians] thwnd : direction of the wind [radians], without ambiguity. Must be in the [0;2*pi[ interval OUTPUT : usrhat : estimate of the radial stokes drift [m/s] ''' mssnoise = load_mss_noise() #Define ensemble of angle where observations are required azibin = numpy.arange(0, 180, dazi) # Ambiguity from SKIM measurment # TOREMOVE usrabs = + numpy.abs(usr) #usrabs = + usr # numpy.abs(usr) nsample, nbeam = numpy.shape(usrabs) # From rad to deg and considering pi ambiguity, # Radial angle is refered to the East degangle = numpy.rad2deg(numpy.mod(angle, numpy.pi)) # Inclination of pass (along track angle) inclination = numpy.transpose(([incl, ]* nbeam)) angle_al = numpy.mod(inclination - angle - numpy.pi/2, 2*numpy.pi) mssr = (numpy.cos(angle_al)**2 * mssx + numpy.sin(angle_al)**2 * mssy + numpy.sin(2*angle_al) * mssxy) numpy.random.seed(nseed) mssr_noise = numpy.random.choice(mssnoise['bins'], numpy.shape(mssr), p=mssnoise['hist']) mssr = mssr + mssr_noise az_al = numpy.mod(inclination - numpy.pi, numpy.pi) degangle = numpy.mod(numpy.rad2deg(angle_al), 180) az_al = numpy.mod(numpy.rad2deg(az_al), 180) # Initialize the reconstruction of ussr usr_comb = numpy.full((nsample, nbeam), numpy.nan) mssr_comb = numpy.full((nsample, nbeam), numpy.nan) usp_comb = numpy.full((nsample, nbeam), numpy.nan) # Distance max to take neighbours dmax = 200 for ibeam in range(nbeam): for isample in range(nsample): if not numpy.isfinite(usrabs[isample, ibeam]): continue plon = lon[isample, ibeam] plat = lat[isample, ibeam] pangle = degangle[isample, ibeam] rangle = angle_al[isample, ibeam] iincl = az_al[isample, ibeam] #dist = mod_tools.dist_sphere(plon, lon, plat, lat) dist = 110*numpy.sqrt((numpy.cos(numpy.deg2rad(lat)) * (lon - plon))**2 + (lat-plat)**2) Idist0 = numpy.where(dist < dmax) azimuthr = angle_al[Idist0].ravel() lonr = lon[Idist0].ravel() latr = lat[Idist0].ravel() usrr = usrabs[Idist0].ravel() mssrr = mssr[Idist0].ravel() distr = dist[Idist0].ravel() ang_azi = numpy.digitize(degangle[Idist0].ravel(), azibin) - 1 Ifinite = numpy.logical_and(numpy.isfinite(usrr), numpy.isfinite(mssrr)) # for each bin, find the closest point and append radial Stokes # drift with ambiguity on sign usrazi = numpy.full(len(azibin), numpy.nan) mssrazi = numpy.full(len(azibin), numpy.nan) thazi = numpy.full(len(azibin), numpy.nan) for iazi in range(len(azibin)): I1 = numpy.where(numpy.logical_and(ang_azi==iazi, Ifinite)) distazi = utils.dist_sphere(plon, lonr[I1], plat, latr[I1], Rearth) # TODO: detect if distazi is empty try: Idist = numpy.argmin(distazi) usrazi[iazi] = usrr[I1][Idist] mssrazi[iazi] = mssrr[I1][Idist] thazi[iazi] = numpy.mod(azimuthr[I1][Idist], numpy.pi) except: usrazi[iazi] = numpy.nan mssrazi[iazi] = numpy.nan try: iaziat = int(numpy.where(numpy.logical_and(iincl > (azibin), iincl < (azibin + dazi)))[0]) Iaziatnot = numpy.logical_not(numpy.arange(len(azibin))==iaziat) except: iaziat = None if iaziat: # Interpolate input data in the along track direction (will be to # avoid as the noise cnnot be removed in that direction) usrazi[iaziat] = numpy.interp(azibin[iaziat], azibin[Iaziatnot], usrazi[Iaziatnot], period=180) mssrazi[iaziat] = numpy.interp(azibin[iaziat], azibin[Iaziatnot], mssrazi[Iaziatnot], period=180) # 3. Perform an estimate of the Stokes drift direction and magnitude angle_usr = thazi - rangle _usr_comb, _usp_comb = proj_uss(usrazi, azibin, angle_usr, dazi, wnd_dir[isample, ibeam]) usr_comb[isample, ibeam] = _usr_comb usp_comb[isample, ibeam] = _usp_comb mssr_comb[isample, ibeam] = numpy.nanmean(mssrazi) #mssr_comb[isample, ibeam] = (numpy.nansum(mssrazi * numpy.sin(angle_usr)) * 2 * dazi / 180.) return usr_comb, usp_comb, mssr_comb def proj_uss(usrazi: numpy.array, azibin: numpy.array, angle_usr: numpy.array, dazi: float, wnd_dir: numpy.array): ''' Project uss on spectrum point ARGS: usrazi : radial Stokes drift for each azimuth in azibin [m/s] azibin: discretized azimuth angle_usr: angle of measurement dazi: increment angle wnd_dir: Wind direction from the ... RETURN: _usr_comb: Radial component of Stockes drift _usp_comb: Perpendicular component of Stockes drift ''' c = numpy.fft.fft(usrazi)/len(usrazi) rd2 = -0.5 * numpy.angle(c[1]) # estimate of Us direction thrbeam = numpy.mod(rd2, 2 * numpy.pi) if numpy.cos(wnd_dir - thrbeam) < 0.: thrbeam = numpy.arctan2(-numpy.sin(thrbeam), -numpy.cos(thrbeam)) # 4. Ambiguity removal using closeness to estimated direction radazibin = numpy.deg2rad(azibin) cond = numpy.where(numpy.cos(thrbeam - radazibin) > 0)[0] usfull = - usrazi * 1 #TOREMOVE #usfull = usrazi if cond.any(): usfull[(cond)] = usrazi[(cond)] _usr_comb = (numpy.nansum(usfull * numpy.cos(angle_usr)) * 2 * dazi / 180.) _usp_comb = (numpy.nansum(usfull * numpy.sin(angle_usr)) * 2 * dazi / 180.) return _usr_comb, _usp_comb def find_closest(lon: numpy.array, lat: numpy.array, mss: numpy.array, mss_est: numpy.array, ice: numpy.array, hs: numpy.array, beam_angle: list, Rearth: float=6378. * 10**3 ) -> Tuple[numpy.ndarray, numpy.ndarray]: ''' Find closest mss at 6degree or nadir, closest hs at nadir. It assumes that there is a nadir measurement. ARGS: lon: longitude for all beams, nadir included lat: latitude for all beams, nadir included mss: mss measurement for all beams mss_est: ice: Percentage of ice for all beams hs: significant height at nadir beam_angle: list anf beam angle RETURNS: mssclose: closest mss measurement found hsclose: closest hs nadir measurement ''' lon_nadir = lon[:, 0] lat_nadir = lat[:, 0] lon = lon[:, 1:] lat = lat[:, 1:] if numpy.min(abs(lon)) <1 and numpy.max(abs(lon))>359: lon = numpy.mod(lon + 180, 360) - 180 lon_nadir = numpy.mod(lon_nadir + 180, 360) - 180 nsample, nbeam = numpy.shape(mss[:, 1:]) mssclose = numpy.full((nsample, nbeam), numpy.nan) hsclose = numpy.full((nsample, nbeam), numpy.nan) ind_6 = numpy.where(numpy.array(beam_angle) == 6)[0] mss6 = mss[:, (ind_6 + 1)].ravel() mss_nadir = mss[:, 0] ice_nadir = ice[:, 0] mss_extr = + mss[:, 1:] ice_extr = + ice[:, 1:] mss_est[numpy.where(ice_extr>0)] = mss_extr[numpy.where(ice_extr>0)] ice6 = ice[:, (ind_6 + 1)].ravel() lon_nadir_ocean = +lon_nadir lon_nadir_ocean[numpy.where((ice_nadir>0))] = numpy.nan # | ~numpy.isfinite(mss_nadir))] = numpy.nan lon_nadir_ice = +lon_nadir lon_nadir_ice[numpy.where((ice_nadir==0))] = numpy.nan # | ~numpy.isfinite(mss_nadir))] = numpy.nan lon_6_ocean = + lon[:, (ind_6)].ravel() lon_6_ocean[numpy.where((ice6 > 0))] = numpy.nan # | ~numpy.isfinite(mss6))] = numpy.nan lon_6_ice = + lon[:, (ind_6)].ravel() lon_6_ice[numpy.where((ice6 == 0))] = numpy.nan #| ~numpy.isfinite(mss6))] = numpy.nan for isample in range(nsample): for ibeam in range(nbeam): if not numpy.isfinite(mss[isample, ibeam + 1]): continue plon = lon[isample, ibeam] plat = lat[isample, ibeam] if ice[isample, ibeam + 1] > 0: _lon = lon_nadir_ice else: _lon = lon_nadir_ocean dist_nadir = utils.dist_sphere(plon, _lon, plat, lat_nadir, Rearth) #dist_nadir[numpy.where(numpy.isnan(mss_nadir))] = numpy.nan # TODO remove ugly try except try: inadir = numpy.nanargmin(dist_nadir) except: continue mss6[mss6==0] = numpy.nan mss_nadir[mss_nadir==0] = numpy.nan dnadir = dist_nadir[inadir] if ice[isample, ibeam + 1] > 0: _lon = lon_6_ice else: _lon = lon_6_ocean dist_6 = utils.dist_sphere(plon, _lon, plat, lat[:, (ind_6)].ravel(), Rearth) try: i6 = numpy.nanargmin(dist_6) except: continue d6 = dist_6[i6] if d6 > 15 and dnadir > 15: mssclose[isample, ibeam] = mss_est[isample, ibeam] ind_dist_nadir = numpy.where(dist_nadir < 150) ind_dist_6 = numpy.where(dist_6 < 50) _hs = numpy.nanmean(hs[ind_dist_nadir]) # if numpy.isnan(_hs): # continue if abs(hs[inadir] - _hs) > 1: continue #if abs(hs[inadir] - _hs) > 0.4: # hsclose[isample, ibeam] = _hs #else: hsclose[isample, ibeam] = hs[inadir] _mss = numpy.nanmean(mss6[ind_dist_6]) if abs(mss6[i6] - _mss) > 0.01: continue #if abs(mss6[i6] - _mss) > 0.005: # mssclose[isample, ibeam] = _mss #else: if d6 < dnadir: mssclose[isample, ibeam] = mss6[i6] else: mssclose[isample, ibeam] = mss_nadir[inadir] if mss[isample, ibeam + 1] == 0: mssclose[isample, ibeam] = 0 return mssclose, hsclose def simulate_spect_skim(sgrid: dict, dic_input: dict, beam_angle: list, radial_angle: numpy.array, delta_azim: float=15, fc: int=1, nseed: int=1 )-> Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]: lon = numpy.transpose(numpy.array(sgrid['lon'])) lat = numpy.transpose(numpy.array(sgrid['lat'])) _angle = numpy.array(sgrid['angle']) if (sgrid['ipass'] %2) != 0: _angle = _angle + numpy.pi incl = numpy.transpose(numpy.array(sgrid['incl'])) list_compulsory = ['uuss', 'vuss', 'mssx', 'mssy', 'mssxy', 'uwnd', 'vwnd', ] wnd_dir = numpy.arctan2(dic_input['vwnd'], dic_input['uwnd']) wnd_dir = numpy.mod(wnd_dir[:, fc:], 2*numpy.pi) usr = utils.proj_radial(dic_input['uuss'][:, fc:], dic_input['vuss'][:, fc:], radial_angle) if 'ice' in dic_input.keys(): ice = dic_input['ice'] else: ice = dic_input['uuss'] * 0 if 'hs' in dic_input.keys(): hs = dic_input['hs'][:, 0] else: hs = dic_input['uuss'][:, 0] * 0 mss = dic_input['mssu'] + dic_input['mssc'] _res = combine_usr(lon[:, fc:], lat[:, fc:], usr, dic_input['mssx'][:, fc:], dic_input['mssy'][:, fc:], dic_input['mssxy'][:, fc:], delta_azim, _angle, incl, wnd_dir, nseed=nseed) usr_comb, usp_comb, mssr_comb = _res mssclose, hsclose = find_closest(lon, lat, mss, mssr_comb, ice, hs, beam_angle) return usr_comb, mssclose, hsclose
13,999
43.444444
105
py
skimulator
skimulator-master/skimulator/error/wave_doppler.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ import sys import numpy from typing import Dict from . import utils import logging import pkg_resources logger = logging.getLogger(__name__) def wd_stream(radial_angle: numpy.array, wndir: numpy.array) -> numpy.ndarray: wd = 6 * numpy.cos(radial_angle - wndir) * 10**(-2) return wd class WaveDoppler(): '''Class wave_doppler defines the computation of the wave doppler that is seen by the instrument for STREAM and SKIM concepts Args: parameters: Simulation settings ''' def __init__(self, parameters): # Seed is not used ? self.nseed = parameters.nseed + 2 self.type_satellite = parameters.type_satellite self.beam_angle = parameters.list_angle def generate(self, wind_norm: numpy.array, wind_dir: numpy.array, windr: numpy.array, mss: numpy.array, hs: numpy.array, ice: numpy.array, sigma0: numpy.array, radial_angle: numpy.array, uuss: numpy.array, vuss: numpy.array) -> Dict[str, numpy.array]: ''' Build errors corresponding to the wave Doppler seen by the instrument on the satellite swath. Args: radial_angle: Angle refered to the across track direction wind_dir: Wind direction refered to the East wind_norm: Wind velocity mss: Mean square slope (mssu + mssc or mssclose) hs: significant wave height ice: ice sigma0: sigma0 uuss: stokes drift Eastward componant vuss: stokes drift Northward componant radial_angle: angle refered to the East ''' if self.type_satellite == 'stream': uwd = wd_stream(radial_angle, wind_dir) ussr = numpy.full(numpy.shape(uwd), numpy.nan) elif self.type_satellite == 'skim': ussr = utils.proj_radial(uuss, vuss, radial_angle) uwd_all = [] _comp = compute_wd_ai_par for b in range(len(self.beam_angle)): uwd = _comp(wind_norm[:, b], windr[:, b], mss[:, b], hs[:, b], ice[:, b], sigma0[:, b], ussr[:, b], radial_angle[:, b], self.beam_angle[b]) uwd_all.append(uwd) uwd_all = numpy.transpose(numpy.array(uwd_all)) else: logger.error(f'Unknown concept {self.type_satellite}') sys.exit(1) return {"wd": uwd_all, "ur_uss": ussr} def correct(self, sgrid:dict, dic_input:dict, wind_norm: numpy.array, windr:numpy.array, ice: numpy.array, sigma0: numpy.array, radial_angle: numpy.array, dazim: float)-> Dict[str, numpy.ndarray]: if self.type_satellite == 'stream': return {} elif self.type_satellite == 'skim': from . import simulate_spectrum _sim = simulate_spectrum.simulate_spect_skim usr_est, mss_est, hs_est = _sim(sgrid, dic_input, self.beam_angle, radial_angle, delta_azim=dazim, fc=1, nseed=self.nseed) uwd_all = [] _comp = compute_wd_ai_par for b in range(len(self.beam_angle)): uwd = _comp(wind_norm[:, b], windr[:, b], mss_est[:, b], hs_est[:, 0], ice[:, b], sigma0[:, b], usr_est[:, b], radial_angle[:, b], self.beam_angle[b]) uwd_all.append(uwd) uwd_all = numpy.transpose(numpy.array(uwd_all)) usr_est = usr_est return {"wd_est": uwd_all, "ur_uss_est": usr_est} def compute_wd_old_par(uuss: numpy.array, vuss:numpy.array, ucur: numpy.array, vcur: numpy.array, uwnd: numpy.array, vwnd: numpy.array, radial_angle:numpy.array, beam_angle: float ) -> Dict[str, numpy.ndarray]: ''' Compute Wave doppler for SKIM instrument using old parametrisation uwd = Gr * ussr and Gr = a * log(b + c/nwr) * (1- tanh(angle)) Args: uuss: Stokes drift eastward componant vuss: Stokes drift northward componant ucur: Current eastward componant vcur: Current northward componant uwnd: Wind eastward componant vwnd: Wind northward componant radial_angle: Angle refered to the East beam_angle: View angle of beam ''' ur_uss = utils.proj_radial(uuss, vuss, radial_angle) nwr = numpy.sqrt((uwnd - ucur)**2 + (vwnd - vcur)**2) nwr[nwr==0] = numpy.nan _angle = numpy.deg2rad((beam_angle - 25) / 10) GR = 25 * (0.82 * numpy.log(0.2 + 7/nwr)) * (1 - numpy.tanh(_angle)) GP = 0 uwd_noerr = GR * ur_uss cshape = numpy.shape(uwd_noerr) noise = numpy.random.normal(0, abs(uwd_noerr) * 0.25, cshape[0]) wd = uwd_noerr + noise return {"ur_uss": ur_uss, "wd_noerr": uwd_noerr, "wd": wd} def compute_wd_ai_par(nwnd: numpy.array, wndr: numpy.array, mss: numpy.array, hs: numpy.array, ice: numpy.array, sigma0: numpy.array, ussr: numpy.array, radial_angle: numpy.array, beam_angle: float) -> Dict[str, numpy.array]: ''' Compute wave doppler using coefficients learned from ww3 data Args: nwmd: Wind velocity wndr: Wind direction refered to the East mss: Mean square slope (mssu + mssc or mssclose) hs: significant wave height ice: ice sigma0: sigma0 ussr: stokes radial componant radial_angle: angle refered to the East beam_angle: beam view incidence angle ''' # Load Coefficents coeff_path = pkg_resources.resource_filename('skimulator', 'share/coeffr.npy') coeff = numpy.load(coeff_path, allow_pickle=True)[()] coeff_f1ur = coeff['f1UWDRglob{:d}deg'.format(int(beam_angle))] coeff_b1ur = coeff['b1UWDRglob{:d}deg'.format(int(beam_angle))] coeff_m = coeff['x_4dmean_{:d}deg'.format(int(beam_angle))] coeff_std = coeff['x_4dstd_{:d}deg'.format(int(beam_angle))] xlabel = coeff['xx{:d}'.format(int(beam_angle))] ncoeffur = 7 #len(coeff_4dm) # -- Compute norm -- # Construct matrix of input data cshape = numpy.shape(wndr) ussr_max = Usr_95(wndr) ussr_min = Usr_5(wndr) diff_max = ussr_max - ussr diff_min = ussr_min - ussr ussr[numpy.where(diff_max < 0)] = ussr_max[numpy.where(diff_max < 0)] ussr[numpy.where(diff_min > 0)] = ussr_min[numpy.where(diff_min > 0)] wndr_min = + wndr wndr_min[numpy.where(abs(wndr) >= 8.)] = 8. nwnd_min = + nwnd nwnd_min[numpy.where(nwnd >= 8.)] = 8. mss_max = mss_95(nwnd) mss_min = mss_5(nwnd) diff_max = (mss_max - mss) diff_min = (mss_min - mss) mss[numpy.where(diff_max < 0)] = mss_max[numpy.where(diff_max < 0)] mss[numpy.where(diff_min > 0)] = mss_min[numpy.where(diff_min > 0)] # noise_nwnd = numpy.random.normal(0, 10, cshape[0]) # wndr = wndr + noise_nwnd _ind = numpy.where(mss <= 8e-4) mss[_ind] = 8e-4 mat_noerr = numpy.full((cshape[0], ncoeffur), numpy.nan) mat_noerr[:, 0] = ussr mat_noerr[:, 1] = wndr # mat_noerr[:, 2] = numpy.sign(wndr) * wndr_min mat_noerr[:, 2] = nwnd mat_noerr[:, 3] = nwnd_min mat_noerr[:, 4] = hs mat_noerr[:, 5] = 1. / mss mat_noerr[:, 6] = 1. / (mss + numpy.log(nwnd + 0.7) * 0.009) # Compute cross product cross = utils.cross_product(mat_noerr, ncoeffur, cshape[0]) ncoeffcross = len(coeff_m) # Normalize with coefficents for i in range(ncoeffcross): cross[:, i] = (cross[:, i] - coeff_m[i]) / coeff_std[i] shape_b1 = numpy.shape(coeff_b1ur)[0] Uwd = utils.reconstruct_var(ncoeffcross, cshape[0], coeff_f1ur, coeff_b1ur, cross, xlabel) if beam_angle == 6: sigma_ice = 2.5 else: sigma_ice = 1. Uwd = Uwd * (1 - sigma_ice * ice / sigma0) Uwd[_ind] = 0.001 Uwd[numpy.where(ice==1)] = 0.001 return Uwd def Usr_95(wndr): # wndr : radial wind speed [m/s] # provides 95th percentile of USR (max tolerable) wndr = wndr - 0.75 # middle point to lower bound mawndr = numpy.maximum(wndr, -8) miwndr = numpy.minimum(wndr, 8) res = 0.0200072*wndr - 0.0073556*miwndr - 0.00290607*mawndr + 0.03929183 return res def Usr_5(wndr): # wndr : radial wind speed [m/s] # provides 95th percentile of USR (min tolerable) wndr = wndr - 0.75 # middle point to lower bound mawndr = numpy.maximum(wndr, -8) miwndr = numpy.minimum(wndr, 8) res = 0.01896863*wndr - 0.00267442*miwndr - 0.00657048*mawndr - 0.02238212 return res def mss_5(wnd): # wnd : wind speed [m/s] # provides 5th percentile of mss (min tolerable) wnd = wnd - 0.5 # middle point to lower bound mawnd = numpy.maximum(wnd, 5) miwnd = numpy.minimum(wnd, 20) res = (0.00152612277*wnd - 4.73687795e-06*mawnd + 2.14390064e-04*wnd**2 - 2.28370767e-04*wnd*mawnd) return res def mss_95(wnd): # wnd : wind speed [m/s] # provides 95th percentile of mss (max tolerable) mawnd = numpy.maximum(wnd, 5) miwnd = numpy.minimum(wnd, 20) wnd = wnd - 0.5 # middle point to lower bound res = (2.71004327e-03 * wnd + 1.27761483e-03*mawnd - 1.68132538e-03*miwnd - 4.400380528e-5*wnd**2 - 1.35712627e-05) return res
10,280
38.694981
79
py
skimulator
skimulator-master/skimulator/error/__init__.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ from .altimeter import Altimeter from .wave_doppler import WaveDoppler from .attitude import Attitude from .dsigma import DSigma from .instrument import Instrument from .rain import Rain #from .wet_troposphere import WetTroposphere
917
35.72
68
py
skimulator
skimulator-master/skimulator/error/attitude.py
""" Copyright (C) 2017-2021 OceanDataLab This file is part of skiMulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skiMulator. If not, see <http://www.gnu.org/licenses/>. """ """ Attitude (yaw and aocs) signal""" import sys import numpy from scipy import interpolate from typing import Dict, Tuple import netCDF4 import datetime from .. import fitspline2d from .. import const import logging logger = logging.getLogger(__name__) def load_yaw_aocs(yaw_file: str) -> Tuple[numpy.ndarray, numpy.ndarray]: fid = netCDF4.Dataset(yaw_file, 'r') time = fid.variables['time'][:] time = time / 86400. yaw_microrad = fid.variables['yaw1'][:] fid.close() return time, yaw_microrad def make_yaw_aocs(time_yaw: numpy.array, vac_yaw: numpy.array, time: numpy.array) -> numpy.ndarray: max_yaw = numpy.max(time_yaw) # ind_time = numpy.where(time > max_yaw) # time[ind_time] = time[ind_time] - max_yaw time = time * 86400 time = numpy.mod(time, max_yaw) f = interpolate.interp1d(time_yaw, vac_yaw) yaw_aocs = f(time) return yaw_aocs def make_yaw_ted(time: numpy.array, cycle: float, angle: numpy.array, first_time: datetime.datetime, beam: float, instr_conf: str) -> numpy.array: import pkg_resources nxspline, nyspline, thedeg = (128, 64, 12) bn = f'Spline_{nxspline:d}_{nyspline:d}_TED_' if instr_conf == 'A': fname = f'{bn}TAS_{beam:d}_DEG.npy' else: fname = '{bn}CB_{beam:d}_DEG.npy' coeff_path = pkg_resources.resource_filename('skimulator', 'share/{}'.format(fname)) wres = numpy.load(coeff_path) param = fitspline2d.ted_tas(wres, nxspline, nyspline, thedeg) # Convert time in seconds time = numpy.mod(time, cycle) * 86400 # Normalize time between 0 and 1 by dividing by the total seconds in # 1 orbit and shift for 0s at 0deglat ascending max_time_orbit = 6083. # max_time_orbit = max(6083, numpy.max(time)) tdec = 5.622404310127427501e-02 * 86400 t_orbit = numpy.mod(time - tdec, max_time_orbit) / max_time_orbit # Shift angle to across track, clockwise az = numpy.mod(-1 * numpy.rad2deg(angle), 360) # Normalize date between 0 and 1 for seasonal cycle date_start = datetime.datetime(first_time.year, 1, 1) time_d = (first_time - date_start).total_seconds() date_end = datetime.datetime(first_time.year, 12, 31, 23, 59, 59) time_total = (date_end - date_start).total_seconds() time_d = numpy.mod(time_d + time, time_total) / time_total # RESULTAT EN ARCSEC # if (t_orbit < 0).any() or (t_orbit < 1).any() or (az < 0).any() # or (az > 360).any() or (time_d <0).any() or (time_d>1).any(): # print(t_orbit, az, time_d) yaw_ted = (param.transform(t_orbit, az, time_d)).astype('float32') # conversion in rad yaw_ted = yaw_ted * numpy.pi * 10**6 / (180 * 3600) return yaw_ted class Attitude(): '''Class attitude defines the yaw from ted and aocs Args: parameters () ''' def __init__(self, parameters): import pkg_resources self.coeff_path_avv6 = pkg_resources.resource_filename('skimulator', 'share/AVV_6deg.npy') self.coeff_path_avv12 = pkg_resources.resource_filename('skimulator', 'share/AVV_12deg.npy') self.coeff_path_std6 = pkg_resources.resource_filename('skimulator', 'share/AVV_6deg.npy') self.coeff_path_std12 = pkg_resources.resource_filename('skimulator', 'share/AVV_6deg.npy') self.DT_ORBIT = 6083 self.n12deg = parameters.list_angle.count(12) self.nbeam = (self.n12deg + parameters.list_angle.count(6)) self.instr_configuration = parameters.instr_configuration self.time_yaw, self.vac_yaw = load_yaw_aocs(parameters.yaw_file) l_angle = [[0,], parameters.list_angle] # Flatten list self.beam_angle = [item for sublist in l_angle for item in sublist] def load_coefficient(): '''not used''' self.avv6 = numpy.load(coeff_path_avv6) self.std6 = numpy.load(coeff_path_std6) self.avv12 = numpy.load(coeff_path_avv12) self.std12 = numpy.load(coeff_path_std12) def generate(self, time:numpy.array, ti_cycle: float, ac_angle: numpy.array, first_time: datetime.datetime)-> Dict[str, numpy.ndarray]: '''Generate AOCS and TED attitude signal''' yaw_aocs_all = [] yaw_ted_all = [] yaw_total_all = [] for b in range(len(self.beam_angle)): yaw_aocs = make_yaw_aocs(self.time_yaw, self.vac_yaw, time[:, b]) if b !=0: yaw_ted = make_yaw_ted(time[:, b], ti_cycle, ac_angle[:, b-1], first_time, self.beam_angle[b], self.instr_configuration) else: yaw_ted = 0 * yaw_aocs # Conversion from microrad to m/s yaw_total = ((yaw_aocs + yaw_ted) * const.vsat * 10**(-6) * numpy.cos(ac_angle[:, b - 1])) yaw_aocs_all.append(yaw_aocs) yaw_ted_all.append(yaw_ted) yaw_total_all.append(yaw_total) yaw_aocs_all = numpy.transpose(numpy.array(yaw_aocs_all)) yaw_ted_all = numpy.transpose(numpy.array(yaw_ted_all)) yaw_total_all = numpy.transpose(numpy.array(yaw_total_all)) return {"yaw" : yaw_total_all, "yaw_aocs": yaw_aocs_all, "yaw_ted": yaw_ted_all} def calc(self, radial_angle, time): '''not finished''' pidx = ((128 * radial_angle / numpy.pi / 2 + 0.49999).astype('int')) % 128 oidx = ((64*(3600*24*(time)) / self.DT_ORBIT + 0.49999).astype('int')) % 64 nx, ny = radial_angle.shape res = numpy.zeros([nx,ny]) noise = numpy.random.randn(nx,ny)
6,582
38.419162
80
py
skimulator
skimulator-master/skimulator/error/generator.py
# Copyright (c) 2020 CNES/JPL # # All rights reserved. Use of this source code is governed by a # BSD-style license that can be found in the LICENSE file. """ Copyright (C) 2017-2021 OceanDataLab This file is part of skimulator. skimulator is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. skimulator is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with skimulator. If not, see <http://www.gnu.org/licenses/>. """ """ Generate instrumental errors ---------------------------- """ from typing import Dict import numpy import datetime from . import (Altimeter, WaveDoppler, Instrument, Rain, Attitude, DSigma) from . import utils # WetTroposphere class Generator: """Instrumental error generator. Args: parameters (settings.Parameters): Simulation settings first_date (numpy.datetime64): Date of the first simulated measurement. """ def __init__(self, parameters): #: The list of user-defined error generators self.generators = [] for item in parameters.noise: if item == Altimeter.__name__: self.generators.append(Altimeter(parameters)) elif item == Attitude.__name__: self.generators.append(Attitude(parameters)) elif item == DSigma.__name__: self.generators.append(DSigma(parameters)) elif item == Instrument.__name__: self.generators.append(Instrument(parameters)) elif item == Rain.__name__: self.generators.append(Rain(parameters)) elif item == WaveDoppler.__name__: self.generators.append(WaveDoppler(parameters)) # elif item == WetTroposphere.__name__: # self.generators.append(WetTroposphere(parameters)) else: # A new error generation class has been implemented but it is # not handled by this object. raise ValueError(f"unknown error generation class: {item}") def generate_dask(self, cycle_number: int, curvilinear_distance: float, time: numpy.array, x_al: numpy.array, x_ac: numpy.array, swh: numpy.array) -> Dict[str, numpy.ndarray]: """Generate errors Args: cycle_number (int): Cycle number. curvilinear_distance (float): Curvilinear distance covered by the satellite during a complete cycle. time (numpy.ndarray): Date of measurements. x_al (numpy.ndarray): Along track distance. x_ac (numpy.ndarray): Across track distance. Returns: dict: Associative array between error variables and simulated values. """ result = {} if not self.generators or x_al.shape[0] == 0: return result futures = [] with dask.distributed.worker_client() as client: for item in self.generators: if isinstance(item, Altimeter): futures.append(client.submit(item.generate, x_al)) elif isinstance(item, BaselineDilation): futures.append(client.submit(item.generate, x_al, x_ac)) elif isinstance(item, CorrectedRollPhase): futures.append(client.submit(item.generate, time, x_ac)) elif isinstance(item, Karin): futures.append( client.submit(item.generate, x_al, x_ac, curvilinear_distance, cycle_number, swh)) elif isinstance(item, RollPhase): futures.append(client.submit(item.generate, x_al, x_ac)) elif isinstance(item, Timing): futures.append(client.submit(item.generate, x_al, x_ac)) elif isinstance(item, WetTroposphere): futures.append(client.submit(item.generate, x_al, x_ac)) for future in dask.distributed.as_completed(futures): result.update(future.result()) return result def generate(self, x_al: numpy.array, dic_input: dict, ac_angle: numpy.array, time:numpy.array, ti_cycle: float, first_time: datetime.datetime, radial_angle: numpy.array, x_al_nadir: numpy.array, x_ac: numpy.array ) -> Dict[str, numpy.ndarray]: res = {} fc = 1 for item in self.generators: if isinstance(item, Altimeter): res.update(item.generate(x_al[:, 0])) elif isinstance(item, DSigma): sigma0 = dic_input['sigma0'] res.update(item.generate(sigma0, ac_angle)) elif isinstance(item, Attitude): res.update(item.generate(time, ti_cycle, ac_angle, first_time)) elif isinstance(item, Instrument): sigma0 = dic_input['sigma0'] wind_dir = numpy.arctan2(dic_input['vwnd'][:, fc:], dic_input['uwnd'][:, fc:]) wind_norm = numpy.sqrt(dic_input['uwnd'][:, fc:]**2 + dic_input['vwnd'][:, fc:]**2) res.update(item.generate(radial_angle, ac_angle, wind_norm, wind_dir, sigma0)) elif isinstance(item, Rain): res.update(item.generate(time, x_al, x_al_nadir, x_ac, ac_angle)) elif isinstance(item, WaveDoppler): sigma0 = dic_input['sigma0'] wind_norm = numpy.sqrt(dic_input['uwnd'][:, fc:]**2 + dic_input['vwnd'][:, fc:]**2) windr = utils.proj_radial(dic_input['uwnd'][:, fc:], dic_input['vwnd'][:, fc:], radial_angle) wind_dir = numpy.arctan2(dic_input['vwnd'][:, fc:], dic_input['uwnd'][:, fc:]) if 'mssu' in dic_input.keys(): mss = dic_input['mssu'][:, fc:] + dic_input['mssc'][:, fc:] else: logger.error('mss variable not found') sys.exit(1) if 'hs' in dic_input.keys(): hs = dic_input['hs'][:, fc:] else: hs = sigma0 * 0 if 'uuss' in dic_input.keys(): uuss = dic_input['uuss'][:, fc:] vuss = dic_input['vuss'][:, fc:] else: uuss = sigma0 * 0 vuss = sigma0 * 0 if 'ice' in dic_input.keys(): ice = dic_input['ice'][:, fc:] else: ice = sigma0 * 0 res.update(item.generate(wind_norm, wind_dir, windr, mss, hs, ice, sigma0, radial_angle, uuss, vuss)) return res def correct(self, grid_dict: dict, dic_input: dict, radial_angle: numpy.array, dazim: float, fc: int=1 ) -> Dict[str, numpy.ndarray]: res = {} for item in self.generators: if isinstance(item, WaveDoppler): sigma0 = + dic_input['sigma0'] wind_norm = numpy.sqrt(dic_input['uwnd'][:, fc:]**2 + dic_input['vwnd'][:, fc:]**2) windr = utils.proj_radial(dic_input['uwnd'][:, fc:], dic_input['vwnd'][:, fc:], radial_angle) if 'ice' in dic_input.keys(): ice = dic_input['ice'][:, fc:] else: ice = sigma0 * 0 res.update(item.correct(grid_dict, dic_input, wind_norm, windr, ice, sigma0, radial_angle, dazim)) return res def generate_obs(self, dic_input: dict, error:dict, corr: dict, fc: int=1, ) -> Dict[str, numpy.ndarray]: ur_obs = + dic_input['ur_true'] nsample = numpy.shape(dic_input['ssh_true'])[0] ssh_obs = + dic_input['ssh_true'][:, 0].reshape(nsample, -1) for item in self.generators: if isinstance(item, Altimeter): fc = 1 ssh_obs += error['altimeter'] elif isinstance(item, DSigma): ur_obs += error['dsigma'] elif isinstance(item, Attitude): # TODO add attitude corrected error or yaw total att = 1 elif isinstance(item, Instrument): ur_obs += error['instr'] elif isinstance(item, Rain): # TODO mask data mask = fc elif isinstance(item, WaveDoppler): if 'wd_est' in dic_input.keys(): wd_corr = (error['wd'] - corr['wd_est']) else: wd_corr = error['wd'] ur_obs += wd_corr return {"ur_obs": ur_obs, "ssh_obs": ssh_obs}
9,631
42.981735
80
py
skimulator
skimulator-master/doc/source/conf.py
# -*- coding: utf-8 -*- # # S4 documentation build configuration file, created by # sphinx-quickstart on Thu Jul 10 16:54:19 2014. # # This file is execfile()d with the current directory set to its # containing dir. # # Note that not all possible configuration values are present in this # autogenerated file. # # All configuration values have a default; values that are commented out # serve to show the default. import sys import os from os.path import expanduser home = expanduser("~") # If extensions (or modules to document with autodoc) are in another directory, # add these directories to sys.path here. If the directory is relative to the # documentation root, use os.path.abspath to make it absolute, like shown here. #sys.path.insert(0, os.path.abspath('.')) sys.path.insert(0, os.path.abspath('../../'+'skimulator')) sys.path.insert(0, os.path.abspath('../../')) #sys.path.append(os.getcwd()+'/../../'+'skimulator') #sys.path.append(os.getcwd()+'/../../') #os.path.join(os.path.dirname(__file__), os.path.pardir) # -- General configuration ------------------------------------------------ # If your documentation needs a minimal Sphinx version, state it here. #needs_sphinx = '1.0' # Add any Sphinx extension module names here, as strings. They can be # extensions coming with Sphinx (named 'sphinx.ext.*') or your custom # ones. extensions = [ 'sphinx.ext.autodoc', 'sphinx.ext.todo', 'sphinx.ext.mathjax', 'sphinx.ext.ifconfig', 'sphinx.ext.viewcode', ] # Add any paths that contain templates here, relative to this directory. templates_path = ['_templates'] # The suffix of source filenames. source_suffix = '.rst' # The encoding of source files. #source_encoding = 'utf-8-sig' # The master toctree document. master_doc = 'index' # General information about the project. project = u'skimulator' copyright = u'2017, Lucile Gaultier' # The version info for the project you're documenting, acts as replacement for # |version| and |release|, also used in various other places throughout the # built documents. # # The short X.Y version. version = '1.1' # The full version, including alpha/beta/rc tags. release = '1.1.0' # The language for content autogenerated by Sphinx. Refer to documentation # for a list of supported languages. #language = None # There are two options for replacing |today|: either, you set today to some # non-false value, then it is used: #today = '' # Else, today_fmt is used as the format for a strftime call. #today_fmt = '%B %d, %Y' # List of patterns, relative to source directory, that match files and # directories to ignore when looking for source files. exclude_patterns = [] # The reST default role (used for this markup: `text`) to use for all # documents. #default_role = None # If true, '()' will be appended to :func: etc. cross-reference text. #add_function_parentheses = True # If true, the current module name will be prepended to all description # unit titles (such as .. function::). #add_module_names = True # If true, sectionauthor and moduleauthor directives will be shown in the # output. They are ignored by default. #show_authors = False # The name of the Pygments (syntax highlighting) style to use. pygments_style = 'sphinx' # A list of ignored prefixes for module index sorting. #modindex_common_prefix = [] # If true, keep warnings as "system message" paragraphs in the built documents. #keep_warnings = False # -- Options for HTML output ---------------------------------------------- # The theme to use for HTML and HTML Help pages. See the documentation for # a list of builtin themes. #html_theme = 'default' html_theme = 'sphinxdoc' # Theme options are theme-specific and customize the look and feel of a theme # further. For a list of options available for each theme, see the # documentation. #html_theme_options = {} # Add any paths that contain custom themes here, relative to this directory. #html_theme_path = [] # The name for this set of Sphinx documents. If None, it defaults to # "<project> v<release> documentation". #html_title = None # A shorter title for the navigation bar. Default is the same as html_title. #html_short_title = None # The name of an image file (relative to this directory) to place at the top # of the sidebar. #html_logo = None # The name of an image file (within the static path) to use as favicon of the # docs. This file should be a Windows icon file (.ico) being 16x16 or 32x32 # pixels large. #html_favicon = None # Add any paths that contain custom static files (such as style sheets) here, # relative to this directory. They are copied after the builtin static files, # so a file named "default.css" will overwrite the builtin "default.css". html_static_path = ['_static'] # Add any extra paths that contain custom files (such as robots.txt or # .htaccess) here, relative to this directory. These files are copied # directly to the root of the documentation. #html_extra_path = [] # If not '', a 'Last updated on:' timestamp is inserted at every page bottom, # using the given strftime format. #html_last_updated_fmt = '%b %d, %Y' # If true, SmartyPants will be used to convert quotes and dashes to # typographically correct entities. #html_use_smartypants = True # Custom sidebar templates, maps document names to template names. #html_sidebars = {} # Additional templates that should be rendered to pages, maps page names to # template names. #html_additional_pages = {} # If false, no module index is generated. #html_domain_indices = True # If false, no index is generated. #html_use_index = True # If true, the index is split into individual pages for each letter. #html_split_index = False # If true, links to the reST sources are added to the pages. #html_show_sourcelink = True # If true, "Created using Sphinx" is shown in the HTML footer. Default is True. #html_show_sphinx = True # If true, "(C) Copyright ..." is shown in the HTML footer. Default is True. #html_show_copyright = True # If true, an OpenSearch description file will be output, and all pages will # contain a <link> tag referring to it. The value of this option must be the # base URL from which the finished HTML is served. #html_use_opensearch = '' # This is the file name suffix for HTML files (e.g. ".xhtml"). #html_file_suffix = None # Output file base name for HTML help builder. htmlhelp_basename = 'skimulator' # -- Options for LaTeX output --------------------------------------------- latex_elements = { # The paper size ('letterpaper' or 'a4paper'). #'papersize': 'letterpaper', # The font size ('10pt', '11pt' or '12pt'). #'pointsize': '10pt', # Additional stuff for the LaTeX preamble. #'preamble': '', } # Grouping the document tree into LaTeX files. List of tuples # (source start file, target name, title, # author, documentclass [howto, manual, or own class]). latex_documents = [ ('index', 'skimulator.tex', 'SKIM Simulator Documentation', u'Lucile Gaultier', 'manual'), ] # The name of an image file (relative to this directory) to place at the top of # the title page. #latex_logo = None # For "manual" documents, if this is true, then toplevel headings are parts, # not chapters. #latex_use_parts = False # If true, show page references after internal links. #latex_show_pagerefs = False # If true, show URL addresses after external links. #latex_show_urls = False # Documents to append as an appendix to all manuals. #latex_appendices = [] # If false, no module index is generated. #latex_domain_indices = True # -- Options for manual page output --------------------------------------- # One entry per manual page. List of tuples # (source start file, name, description, authors, manual section). man_pages = [ ('index', 'SKIM Simulator', u'SKIM Simulator Documentation', [u'Lucile Gaultier'], 1) ] # If true, show URL addresses after external links. #man_show_urls = False # -- Options for Texinfo output ------------------------------------------- # Grouping the document tree into Texinfo files. List of tuples # (source start file, target name, title, author, # dir menu entry, description, category) texinfo_documents = [ ('index', 'SKIM Simulator', u'SKIM Simulator Documentation', u'Lucile Gaultier', 'SKIM Simulator', 'Open Source SKIM Simulator.', 'Miscellaneous'), ] # Documents to append as an appendix to all manuals. #texinfo_appendices = [] # If false, no module index is generated. #texinfo_domain_indices = True # How to display URL addresses: 'footnote', 'no', or 'inline'. #texinfo_show_urls = 'footnote' # If true, do not generate a @detailmenu in the "Top" node's menu. #texinfo_no_detailmenu = False # -- Options for Epub output ---------------------------------------------- # Bibliographic Dublin Core info. epub_title = u'skimulator' epub_author = u'Lucile Gaultier' epub_publisher = u'Lucile Gaultier' epub_copyright = u'2017, Lucile Gaultier' # The basename for the epub file. It defaults to the project name. #epub_basename = u'SKIM Simulator' # The HTML theme for the epub output. Since the default themes are not optimized # for small screen space, using the same theme for HTML and epub output is # usually not wise. This defaults to 'epub', a theme designed to save visual # space. #epub_theme = 'epub' # The language of the text. It defaults to the language option # or en if the language is not set. #epub_language = '' # The scheme of the identifier. Typical schemes are ISBN or URL. #epub_scheme = '' # The unique identifier of the text. This can be a ISBN number # or the project homepage. #epub_identifier = '' # A unique identification for the text. #epub_uid = '' # A tuple containing the cover image and cover page html template filenames. #epub_cover = () # A sequence of (type, uri, title) tuples for the guide element of content.opf. #epub_guide = () # HTML files that should be inserted before the pages created by sphinx. # The format is a list of tuples containing the path and title. #epub_pre_files = [] # HTML files shat should be inserted after the pages created by sphinx. # The format is a list of tuples containing the path and title. #epub_post_files = [] # A list of files that should not be packed into the epub file. epub_exclude_files = ['search.html'] # The depth of the table of contents in toc.ncx. #epub_tocdepth = 3 # Allow duplicate toc entries. #epub_tocdup = True # Choose between 'default' and 'includehidden'. #epub_tocscope = 'default' # Fix unsupported image types using the PIL. #epub_fix_images = False # Scale large images. #epub_max_image_width = 0 # How to display URL addresses: 'footnote', 'no', or 'inline'. #epub_show_urls = 'inline' # If false, no index is generated. #epub_use_index = True
10,718
30.342105
80
py
skimulator
skimulator-master/doc/source/params.py
# -----------------------# # Files and directories # -----------------------# ## -- Get the user home directory from os.path import expanduser import os import math home = expanduser("~") # ------ Name of the configuration (to build output files names) config = [yourconfig] # ------ Directory that contains orbit file: dir_setup = os.path.join([yourpath], 'skimulator', 'data') # ------ Directory that contains your own inputs: indatadir = '[yourpath_to_yourdata]/' # ------ Directory that contains your outputs: outdatadir = '[yourpath_to_outputs]/' # ------ Orbit file: satname = [chosenorbit] filesat = os.path.join(dir_setup, [chosenorbit]) # ------ Number of days in orbit (optional if specified in orbit file) satcycle = 29 # ------ Satellite elevation (optional if specified in orbit file) sat_elev = 817 * 10**3 # ------ Order of columns (lon, lat, time) in orbit file # (default is (0, 1, 2) with order_orbit_col = None) order_orbit_col = None # , dir_setup+os.sep+'orbjason.txt', dir_setup+os.sep+'orbaltika.txt' ] # ------ Number of processors for parallelisation purposes proc_number = [number of processor (integer)] # ------ Deactivate printing of progress bar to avoid huge log progress_bar = True or False # -----------------------# # SKIM swath parameters # -----------------------# # ------ Satellite grid file root name: # (Final file name is root_name_[numberofpass].nc) filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) or filesgrid = os.path.join(outdatadir, '[your_grid_root_name]') # ------ Force the computation of the satellite grid: makesgrid = True or False # ------ Give a subdomain if only part of the model is needed: # (modelbox=[lon_min, lon_max, lat_min, lat_max]) # (If modelbox is None, the whole domain of the model is considered) modelbox = None or [yourlon_min, yourlon_max, yourlat_min, yourlat_max] #------- Rotation speed of the antenna (in tr/min) rotation_speed = rotation depends on the chosen config #------- List of position of beams: list_pos = (0, [angle_in_rad], [angle_in_rad] ...) #------- List of angle of beams in degrees: list_angle = ([incidence], [incidence], [incidence] ...) #------- List of timeshift as regard to nadir for 12 degree beams: list_shift = (1, 3, 2 ...) #------- Cycle duration cycle = 0.0096 # ------ Shift longitude of the orbit file if no pass is in the domain # (in degree): Default value is None (no shift) shift_lon = 0 # ------ Shift time of the satellite pass (in day): # Default value is None (no shift) shift_time = None # -----------------------# # Model input parameters # -----------------------# # ------ List of model files: # (The first file contains the grid and is not considered as model data) # To generate the noise alone, file_input=None and specify region # in modelbox file_input = os.path.join(indatadir, [your_list_of_file_name.txt]' or None # ------ Type of model data: # (Optional, default is NETCDF_MODEL and reads netcdf3 and netcdf4 files) # (Other option is WW3) model = 'WW3' or 'NETCDF_MODEL' # ------ First time of the model first_time = 'yyyy-mm-ddTHH:MM:SSZ' # ------ Grid file name file_grid_model = (os.path.join(indatadir, [yourgridfileu]), os.path.join(indatadir, [yourgridfilev]),) # ------ Specify if there is a ice mask for high latitudes # (if true, mask is recomputed at each cycle) ice_mask = False or True # ------ Type of grid: # 'regular' or 'irregular', if 'regular' only 1d coordinates # are extracted from model grid = 'regular' # ------ Specify list of variable: list_input_var = {'ucur': [[u_var], [vel_ext], 0], 'vcur': [[v_var], [v_ext], 1], 'uuss': [[uuss_var], [uss_ext], 0], 'vuss': [[vuss_var], [uss_ext], 1], 'ice': [[ice_var], [ice_ext], 0], 'mssd': [[mssd_var], [msd_ext], 0], 'mssx': [[mssx_var], [mss_ext], 0], 'mssy':[[mssy_var], [mss_ext]], 'ssh': [[ssh_var], [ssh_ext], 0], 'uwnd': [[uwnd_var], [wnd_ext], 0], 'vwnd': [[vwnd_var], [wnd_ext], 1]} # ------ Specify longitude variable: lon = ('longitude', 'longitude') # ------ Specify latitude variable: lat = ('latitude', 'latitude') # ------ Specify number of time in file: dim_time = 24 # ------ Time step between two model outputs (in days): timestep = 1/24. # ------ Number of outputs to consider: # (timestep*nstep=total number of days) nstep = 35*24 # ------ Not a number value: model_nan = -32767. # -----------------------# # SKIM output files # -----------------------# # ------ Output file root name: # (Final file name is root_name_c[cycle].nc file_output = os.path.join(outdatadir, config) # ------ Interpolation of the SSH from the model (if grid is irregular and # pyresample is not installed: # (either 'linear' or 'nearest', use 'nearest' for large region # as it is faster and use less memory.) interpolation = 'nearest' or 'linear' # ------ List of output variables: list_output = ['ssh_obs', 'ur_true', 'ucur', 'vcur', 'uuss', 'vuss', 'instr', 'radial_angle', 'vwnd', 'mssx', 'mssy', 'mssxy', 'uwb', 'ssh_true', 'ssh', 'ice', 'mssd', 'vindice', 'ur_obs', 'uwnd', 'sigma0'] # -----------------------# # SKIM error parameters # -----------------------# # ------ File containing random coefficients to compute and save # random error coefficients so that runs are reproducible: # If file_coeff is specified and does not exist, file is created # If you don't want runs to be reproducible, file_coeff is set to None file_coeff = None or os.path.join(outdatadir, 'Random_coeff.nc') # Compute instrumental nadir noise: nadir = True # ------ Number of random realisations for instrumental and geophysical error # (recommended ncomp=2000), ncomp1d is used for 1D spectrum, and ncomp2d # is used for 2D spectrum (wet troposphere computation): ncomp1d = 3000 ncomp2d = 2000 # ------- Instrument white noise error instr = True or False # ------- Choice of instrument configuration instr_configuration = 'A' or 'B' # ------- Attitude error attitude = True or False # ------- File which provide the AOCS error: yaw_file = os.path.join(dir_setup, 'sample_req1.nc') # ------- Wave bias uwb = True or False ## -- Geophysical error ## ---------------------- # ------ Consider ice in sigma0 computation ice = True or False # ------ Rain error (True to compute it): rain = True or False # ------ Rain file containing scenarii (python file): rain_file = os.path.join(dir_setup, [yourrainscenarii]) # ------ Threshold to flag data: rain_threshold = 0.15 # -----------------------# # L2C computation # -----------------------# # config name for L2c: config_l2c = '[yourl2cconfig]' # Length resolution to select neighbors (in km): resol = 40 # Grid resolution for l2c (alongtrack, acrosstrack) grid (in km): posting = 5 # Remove noisy data around nadir (in km): ac_threshold = 20 # List of variables to be interpolated on the swath: list_input_var_l2c = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 1]} # -----------------------# # L2D computation # -----------------------# # config name for L2d: config_l2d = '' # Length resolution to select neighbors (multiplication factor): resol_spatial_l2d = 1 # Temporal resolution to select neighbors (multiplication factor): resol_temporal_l2d = 1 # Grid resolution for l2d (lat, lon) grid (in degrees): posting_l2d = (0.1, 0.1) # Time domain: (start_time, end_time, dtime) in days: time_domain = (5, 25, 1) # Spatial domain (lon_min, lon_max, lat_min, lat_max): spatial_domain = [0, 360, -90, 90] # List of variables to be interpolated on the grid: list_input_var_l2d = {'ucur': ['ucur', 'cur', 0], 'vcur': ['vcur', 'cur', 1]}
7,779
38.492386
89
py
skimulator
skimulator-master/doc/images/code_image/Fig4.py
''' FIG. 4: Radial currents and instrumental noise. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] # List files indatadir = '/mnt/data/project/' indatadir = '/tmp/key/project/' indatadir = os.path.join(indatadir, 'skim', 'skim_output') config="WW3_GS_6b108az" config="WW3_GS_8b105az" filesgrid = os.path.join(indatadir, '{}_'.format(config)) ipass = 59 indatapath = '{}c01_p{:03d}.nc'.format(filesgrid, ipass) listfile = glob.glob(indatapath) listfile = sorted(listfile) outdatadir = '../' modelbox = [-73, -71.0, 34.0, 36.0] scale = 11 # Prepare figure pyplot.figure(figsize=(10, 15)) projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() projection = transform ax1 = pyplot.subplot(111, projection=projection) ax1.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') ax1.set_extent(modelbox, projection) gl = ax1.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False data = netCDF4.Dataset(indatapath, 'r') indatapath = '{}grid_p{:03d}.nc'.format(filesgrid, ipass) datag = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] ur = data['ur_model'][:] corrangle = datag['radial_angle'][:] urinstr = data['instr'][:] urnoise = ur + urinstr uur = ur * numpy.cos(corrangle) vur = ur * numpy.sin(corrangle) uurnoise = urnoise * numpy.cos(corrangle) vurnoise = urnoise * numpy.sin(corrangle) for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) #pyplot.plot(lon[:, i], lat[:, i], style_color, # transform=cartopy.crs.PlateCarree()) pyplot.quiver(lon[:, i], lat[:, i], uurnoise[:, i], vurnoise[:, i], color='green', scale=scale, transform=transform) pyplot.quiver(lon[:, i], lat[:, i], uur[:, i], vur[:, i], color='red', scale=scale, transform=transform, alpha=0.5) pyplot.savefig(os.path.join(outdatadir, 'Fig4.png'))
2,193
30.797101
74
py
skimulator
skimulator-master/doc/images/code_image/Fig6.py
''' FIG. 4: Radial currents and instrumental noise. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] # List files indatadir = '/mnt/data/project/' indatadir = '/tmp/key/project/' indatadir = os.path.join(indatadir, 'skim', 'skim_output') config="WW3_GS_6b108az" config="WW3_GS_8b105az" filesgrid = os.path.join(indatadir, '{}_'.format(config)) ipass = 59 indatapath = '{}c01_p{:03d}.nc'.format(filesgrid, ipass) listfile = glob.glob(indatapath) listfile = sorted(listfile) outdatadir = '../' modelbox = [-73, -71.0, 34.0, 36.0] scale = 11 # Prepare figure pyplot.figure(figsize=(10, 15)) projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() projection = transform ax1 = pyplot.subplot(111, projection=projection) ax1.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') ax1.set_extent(modelbox, projection) gl = ax1.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False data = netCDF4.Dataset(indatapath, 'r') indatapath = '{}grid_p{:03d}.nc'.format(filesgrid, ipass) datag = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] ur = data['ur_model'][:] corrangle = datag['radial_angle'][:] ur_noise = data['uss_err'][:] + data['instr'][:] urnoise = ur + ur_noise uur = ur * numpy.cos(corrangle) vur = ur * numpy.sin(corrangle) uurnoise = urnoise * numpy.cos(corrangle) vurnoise = urnoise * numpy.sin(corrangle) for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) #pyplot.plot(lon[:, i], lat[:, i], style_color, # transform=cartopy.crs.PlateCarree()) pyplot.quiver(lon[:, i], lat[:, i], uurnoise[:, i], vurnoise[:, i], color='green', scale=scale, transform=transform) pyplot.quiver(lon[:, i], lat[:, i], uur[:, i], vur[:, i], color='red', scale=scale, transform=transform, alpha=0.5) pyplot.savefig(os.path.join(outdatadir, 'Fig6.png'))
2,216
31.130435
74
py
skimulator
skimulator-master/doc/images/code_image/Fig2.py
''' FIG. 2: scheme of the SKIM geometry with 4 beams at 12 degrees and 1 beam at 6 degree and 5 beams at 12 degrees and 2 beams at 6 degree. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] # List files config="WW3_AT_metop_2018_8b" indatadir = '/mnt/data/project/' indatadir = '/tmp/key/data/skim_at_output/{}'.format(config) #indatadir = os.path.join(indatadir, 'skim', 'skim_output') filesgrid = os.path.join(indatadir, '{}_grid'.format(config)) ipass = 58 indatapath = '{}_p{:03d}.nc'.format(filesgrid, ipass) outdatadir = '../' modelbox = [-90, -70., 32., 40.] #modelbox = [270, 290., 32., 40.] # Prepare figure pyplot.figure(figsize=(10, 15)) projection = cartopy.crs.Mercator() transform = cartopy.crs.PlateCarree() projection = transform ax1 = pyplot.subplot(121, projection=projection) #ax.coastlines() ax1.add_feature(cartopy.feature.OCEAN, zorder=1) ax1.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') #ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], # projection) ax1.set_extent([-74., -68., 34, 37], projection) gl = ax1.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False ax1.set_title('(a) 6 beams configuration') data = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] lon_nadir = data['lon_nadir'][:] lon_nadir[lon_nadir > 180] = lon_nadir[lon_nadir > 180] - 360 lat_nadir = data['lat_nadir'][:] pyplot.plot(lon_nadir, lat_nadir, 'k+', transform=transform) for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) pyplot.plot(lon[:, i], lat[:, i], style_color, transform=cartopy.crs.PlateCarree()) ax2 = pyplot.subplot(122, projection=projection) #ax.coastlines() ax2.add_feature(cartopy.feature.OCEAN, zorder=1) ax2.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') #ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], # projection) #config="WW3_GS_8b105az" config="WW3_AT_metop_2018_6a" indatadir = '/tmp/key/data/skim_at_output/{}'.format(config) filesgrid = os.path.join(indatadir, '{}_grid'.format(config)) indatapath = '{}_p{:03d}.nc'.format(filesgrid, ipass) print(indatapath) ax2.set_extent([-74., -68., 34, 37], projection) gl = ax2.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False ax2.set_title('(b) 8 beams configuration') data = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] lon_nadir = data['lon_nadir'][:] lon_nadir[lon_nadir > 180] = lon_nadir[lon_nadir > 180] - 360 lat_nadir = data['lat_nadir'][:] pyplot.plot(lon_nadir, lat_nadir, 'k+', transform=transform) for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) pyplot.plot(lon[:, i], lat[:, i], style_color, transform=cartopy.crs.PlateCarree()) pyplot.savefig(os.path.join(outdatadir, 'Fig2.png'))
3,290
35.164835
139
py
skimulator
skimulator-master/doc/images/code_image/Fig3.py
''' FIG. 3: Model interpolated currents and the corresponding radial currents. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] # List files config="WW3_AT_metop_2018_8b" indatadir = '/mnt/data/project/' indatadir = '/tmp/key/data/skim_at_output/{}'.format(config) modeldatapath = '/tmp/key/data/model/ww3_gs/ww3.201109_cur.nc' filesgrid = os.path.join(indatadir, '{}_'.format(config)) ipass = 58 indatapath = '{}c01_p{:03d}.nc'.format(filesgrid, ipass) listfile = glob.glob(indatapath) listfile = sorted(listfile) outdatadir = '../' modelbox = [-71, -69.0, 34.0, 36.0] scale = 11 is_cartopy = True # Prepare figure pyplot.figure(figsize=(10, 15)) projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() if is_cartopy is True: #projection = transform ax1 = pyplot.subplot(121, projection=projection) else: ax1 = pyplot.subplot(121) if is_cartopy is True: ax1.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') ax1.set_extent(modelbox, projection) gl = ax1.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False ax1.set_title('(a) Model, "true" velocity') data = netCDF4.Dataset(indatapath, 'r') datam = netCDF4.Dataset(modeldatapath, 'r') iifile = data['index'][:, 0] lon = datam['longitude'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = datam['latitude'][:] mlon, mlat = numpy.meshgrid(lon, lat) mu = datam['ucur'][int(iifile[0]),:, :] mv = datam['vcur'][int(iifile[0]),:, :] mu=numpy.ma.masked_invalid(mu) numpy.ma.masked_where(mu == datam['ucur']._FillValue, mu, copy=False) mu[mu.mask] = 0 numpy.ma.masked_where(mv == datam['vcur']._FillValue, mv, copy=False) mv[mv.mask] = 0 s0 = 10 if is_cartopy is True: pyplot.pcolor(mlon, mlat, numpy.sqrt(mu**2 + mv**2), vmax=2.0, cmap='jet', transform=transform) pyplot.quiver(mlon[::s0, ::s0], mlat[::s0, ::s0], mu[::s0, ::s0], mv[::s0, ::s0], scale=scale, transform=transform) else: pyplot.pcolor(mlon, mlat, numpy.sqrt(mu**2 + mv**2), vmax=2.0, cmap='jet') pyplot.quiver(mlon[::s0, ::s0], mlat[::s0, ::s0], mu[::s0, ::s0], mv[::s0, ::s0]) indatapath = '{}grid_p{:03d}.nc'.format(filesgrid, ipass) datag = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] ax2 = pyplot.subplot(122, projection=projection) ax2.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') print(indatapath) ax2.set_extent(modelbox, projection) gl = ax2.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False ax2.set_title('(b) Interpolated (green) and projected (red) velocity') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] ur = data['ur_true'][:] corrangle = datag['radial_angle'][:] datau = data['u_model'][:] datav = data['v_model'][:] uur = ur * numpy.cos(corrangle) uur[uur.mask] = 0 vur = ur * numpy.sin(corrangle) vur[vur.mask] = 0 for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) # pyplot.plot(lon[:, i], lat[:, i], style_color, # transform=transform) pyplot.quiver(lon[:, i], lat[:, i], datau[:, i], datav[:, i], color='green', scale=scale, transform=transform) pyplot.quiver(lon[:, i], lat[:, i], uur[:, i], vur[:, i], color='red', scale=scale, transform=transform) pyplot.savefig(os.path.join(outdatadir, 'Fig3.png'))
3,784
32.794643
78
py
skimulator
skimulator-master/doc/images/code_image/Fig1.py
''' FIG. 1: 5-day worth of SKIM simulated data in a global configuration with the science orbit. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # List files outdatadir = '/tmp/key' outdatadir = '/mnt/data/project/' outdatadir = os.path.join(outdatadir, 'skim', 'skim_output') config="WW3_GLOB_6b108az" filesgrid = os.path.join(outdatadir, '{}_grid'.format(config)) indatapath = '{}_*'.format(filesgrid) listfile = glob.glob(indatapath) listfile = sorted(listfile) #.sort() outdatadir = '../' modelbox = [0., 360., -90., 90.] # Prepare figure pyplot.figure(figsize=(10, 15)) ax = pyplot.axes(projection=cartopy.crs.Orthographic(0, 0)) #ax.set_extent([modelbox[0], modelbox[1], modelbox[2], modelbox[3]], # crs=cartopy.crs.Mercator()) #ax.coastlines() ax.add_feature(cartopy.feature.OCEAN, zorder=1) ax.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') ax.set_global() #ax.gridlines() for ifile in listfile[:90]: data = netCDF4.Dataset(ifile, 'r') lon = data['lon'][:, 0] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:, 0] lon_nadir = data['lon_nadir'][:] lon_nadir[lon_nadir > 180] = lon_nadir[lon_nadir > 180] - 360 lat_nadir = data['lat_nadir'][:] pyplot.plot(lon[:], lat[:], '.', color='#4EE2EC', markersize=0.3, transform=cartopy.crs.PlateCarree()) pyplot.plot(lon_nadir, lat_nadir, '.', color='#565051', markersize=0.3, transform=cartopy.crs.PlateCarree()) pyplot.savefig(os.path.join(outdatadir, 'Fig1.png'))
1,647
31.96
92
py
skimulator
skimulator-master/doc/images/code_image/Fig5.py
''' FIG. 4: Radial currents and instrumental noise. ''' import netCDF4 import numpy import matplotlib matplotlib.use('Agg') from matplotlib import pyplot import skimulator.rw_data as rw import glob import cartopy import os # Initialize color listcolor = ['c', 'y', 'b', 'g', 'k', 'r', 'c', 'y'] # List files indatadir = '/mnt/data/project/' indatadir = '/tmp/key/project/' indatadir = os.path.join(indatadir, 'skim', 'skim_output') config="WW3_GS_6b108az" config="WW3_GS_8b105az" filesgrid = os.path.join(indatadir, '{}_'.format(config)) ipass = 59 indatapath = '{}c01_p{:03d}.nc'.format(filesgrid, ipass) listfile = glob.glob(indatapath) listfile = sorted(listfile) outdatadir = '../' modelbox = [-73, -71.0, 34.0, 36.0] scale = 11 # Prepare figure pyplot.figure(figsize=(10, 15)) projection = cartopy.crs.PlateCarree() transform = cartopy.crs.PlateCarree() projection = transform ax1 = pyplot.subplot(111, projection=projection) ax1.add_feature(cartopy.feature.LAND, zorder=1, edgecolor='black') ax1.set_extent(modelbox, projection) gl = ax1.gridlines(crs=transform, draw_labels=True, color='gray', linestyle='--', alpha=0.5) gl.xlabels_top = False gl.ylabels_left = False data = netCDF4.Dataset(indatapath, 'r') indatapath = '{}grid_p{:03d}.nc'.format(filesgrid, ipass) datag = netCDF4.Dataset(indatapath, 'r') lon = data['lon'][:] lon[lon > 180] = lon[lon > 180] - 360 lat = data['lat'][:] ur = data['ur_model'][:] corrangle = datag['radial_angle'][:] ur_stroke = data['uss_err'][:] urnoise = ur + ur_stroke uur = ur * numpy.cos(corrangle) vur = ur * numpy.sin(corrangle) uurnoise = urnoise * numpy.cos(corrangle) vurnoise = urnoise * numpy.sin(corrangle) for i in range(numpy.shape(lon)[1]): style_color = '{}+'.format(listcolor[i]) #pyplot.plot(lon[:, i], lat[:, i], style_color, # transform=cartopy.crs.PlateCarree()) pyplot.quiver(lon[:, i], lat[:, i], uurnoise[:, i], vurnoise[:, i], color='green', scale=scale, transform=transform) pyplot.quiver(lon[:, i], lat[:, i], uur[:, i], vur[:, i], color='red', scale=scale, transform=transform, alpha=0.5) pyplot.savefig(os.path.join(outdatadir, 'Fig5.png'))
2,199
30.884058
74
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/setup.py
#!/usr/bin/env python # -*- coding: utf-8 -*- """ compositionspace : add licence and such """ from setuptools import setup, find_packages, Extension with open('README.md') as readme_file: readme = readme_file.read() setup_requirements = ['pytest-runner', ] test_requirements = ['pytest>=3', ] setup( author="Alaukik Saxena, Sarath Menon, Mariano Forti", author_email='s.menon@mpie.de', python_requires='>=3.8', classifiers=[ 'Development Status :: 2 - Pre-Alpha', 'Intended Audience :: Developers', 'License :: OSI Approved :: GNU General Public License v3 (GPLv3)', 'Natural Language :: English', 'Programming Language :: Python :: 3', 'Programming Language :: Python :: 3.5', 'Programming Language :: Python :: 3.6', 'Programming Language :: Python :: 3.7', ], description="APT analysis tools", install_requires = ['numpy', 'matplotlib', 'pandas', 'h5py', 'scikit-learn', 'tqdm', 'pyevtk', 'pyyaml', 'pyvista'], #license="GNU General Public License v3", long_description=readme, long_description_content_type='text/markdown', include_package_data=True, keywords='compositionspace', name='compositionspace', packages=find_packages(include=['compositionspace', 'compositionspace.*']), setup_requires=setup_requirements, test_suite='tests', tests_require=test_requirements, url='', version='0.0.9', zip_safe=False, #entry_points={ # 'console_scripts': [ # 'calphy = calphy.kernel:main', # 'calphy_kernel = calphy.queuekernel:main', # ], #} )
1,643
29.444444
80
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/datautils.py
import pandas as pd import re import os import yaml from tqdm.notebook import tqdm import matplotlib.pyplot as plt import numpy as np import pickle import time import h5py import warnings import compositionspace.paraprobe_transcoder as paraprobe_transcoder #really check this! pd.options.mode.chained_assignment = None class DataPreparation: def __init__(self, inputfile): if isinstance(inputfile, dict): self.params = inputfile else: with open(inputfile, "r") as fin: params = yaml.safe_load(fin) self.params = params self.version = "1.0.0" if not os.path.exists(self.params['output_path']): os.mkdir(self.params['output_path']) def get_label_ions(self, pos, rrngs): pos['comp'] = '' pos['colour'] = '#FFFFFF' pos['nature'] = '' count=0; for n,r in rrngs.iterrows(): count= count+1; pos.loc[(pos.Da >= r.lower) & (pos.Da <= r.upper),['comp','colour', 'nature']] = [r['comp'],'#' + r['colour'],count] return pos def atom_filter(self, x, atom_range): """ Get a list of atom species and their counts Parameters ---------- Returns ------- Notes ----- Assuming all the data """ atom_total = [] x_arr = x.values mass = x_arr[:,-1] x_columns = x.columns for i in range(len(atom_range)): low_bound = atom_range['lower'][i] upper_bound = atom_range['upper'][i] id_ = np.where(np.logical_and(mass>=low_bound, mass<=upper_bound)) atom = x_arr[id_] atom_total.append(atom) atom_total = tuple(atom_total) atom_total = np.vstack(atom_total) atom_total = pd.DataFrame(atom_total, columns = x_columns) count_atom= len(atom_total) return atom_total, count_atom def get_pos(self, file_name): """ Read the pos file Parameters ---------- file_name: string Name of the input file Returns ------- pos: np structured array The atom positions and ---- ratio Notes ----- Assumptions Examples -------- Raises ------ FileNotFoundError: describe """ if not os.path.exists(file_name): raise FileNotFoundError(f"filename {file_name} does not exist") with open(file_name, 'rb') as f: dt_type = np.dtype({'names':['x', 'y', 'z', 'm'], 'formats':['>f4', '>f4', '>f4', '>f4']}) pos = np.fromfile(f, dt_type, -1) pos = pos.byteswap().newbyteorder() return pos def get_rrng(self, file_name): """ Read the data Parameters ---------- Returns ------- Notes ----- """ if not os.path.exists(file_name): raise FileNotFoundError(f"filename {file_name} does not exist") patterns = re.compile(r'Ion([0-9]+)=([A-Za-z0-9]+).*|Range([0-9]+)=(\d+.\d+) +(\d+.\d+) +Vol:(\d+.\d+) +([A-Za-z:0-9 ]+) +Color:([A-Z0-9]{6})') ions = [] rrngs = [] with open(file_name, "r") as rf: for line in rf: m = patterns.search(line) if m: if m.groups()[0] is not None: ions.append(m.groups()[:2]) else: rrngs.append(m.groups()[2:]) ions = pd.DataFrame(ions, columns=['number','name']) ions.set_index('number',inplace=True) rrngs = pd.DataFrame(rrngs, columns=['number','lower','upper','vol','comp','colour']) rrngs.set_index('number',inplace=True) rrngs[['lower','upper','vol']] = rrngs[['lower','upper','vol']].astype(float) rrngs[['comp','colour']] = rrngs[['comp','colour']].astype(str) return ions, rrngs def get_apt(self, file_name): """ Read the apt file Parameters ---------- file_name: string Name of the input file Returns ------- pos: np structured array The atom positions and ---- ratio Notes ----- Assumptions Examples -------- Raises ------ FileNotFoundError: describe """ if not os.path.exists(file_name): raise FileNotFoundError(f"filename {file_name} does not exist") apt = paraprobe_transcoder.paraprobe_transcoder(file_name) apt.read_cameca_apt() POS = apt.Position MASS = apt.Mass POS_MASS = np.concatenate((POS,MASS),axis = 1) return POS_MASS def get_apt_dataframe(self): """ Read the data Parameters ---------- Returns ------- Notes ----- """ df_Mass_POS_lst = [] file_name_lst=[] ions = None rrngs = None pbar = tqdm(os.listdir(self.params["input_path"]), desc="Reading files") for filename in pbar: if filename.endswith(".POS"): #print(filename) path = os.path.join(self.params["input_path"], filename) pos = self.get_pos(path) df_POS_MASS = pd.DataFrame({'x':pos['x'],'y': pos['y'],'z': pos['z'],'Da': pos['m']}) df_Mass_POS_lst.append(df_POS_MASS) file_name_lst.append(filename) if filename.lower().endswith(".pos"): path = os.path.join(self.params["input_path"], filename) pos = self.get_pos(path) df_POS_MASS = pd.DataFrame({'x':pos['x'],'y': pos['y'],'z': pos['z'],'Da': pos['m']}) df_Mass_POS_lst.append(df_POS_MASS) file_name_lst.append(filename) if filename.endswith(".apt"): path = os.path.join(self.params["input_path"], filename) POS_MASS = self.get_apt(path) df_POS_MASS = pd.DataFrame(POS_MASS, columns = ["x","y","z","Da"]) df_Mass_POS_lst.append(df_POS_MASS) file_name_lst.append(filename) if filename.lower().endswith(".rrng"): path = os.path.join(self.params["input_path"], filename) ions,rrngs = self.get_rrng(path) if filename.endswith(".RRNG"): path = os.path.join(self.params["input_path"], filename) ions,rrngs = self.get_rrng(rrange_file) return (df_Mass_POS_lst, file_name_lst, ions, rrngs) def get_big_slices(self): """ Cut the data into specified portions Parameters ---------- Returns ------- Notes ----- """ #df_lst, files, ions, rrngs= read_apt_to_df(folder) df_lst, files, ions, rrngs= self.get_apt_dataframe() filestrings = [] prefix = self.params['output_path'] for idx, file in enumerate(files): org_file = df_lst[idx] atoms_spec = [] c = np.unique(rrngs.comp.values) for i in range(len(c)): range_element = rrngs[rrngs['comp']=='{}'.format(c[i])] total, count = self.atom_filter(org_file, range_element) name = i total["spec"] = [name for j in range(len(total))] atoms_spec.append(total) df_atom_spec = pd.concat(atoms_spec) x_wu=df_atom_spec sorted_df = x_wu.sort_values(by=['z']) filestring = "file_{}_large_chunks_arr.h5".format(file.replace(".","_")) filestring = os.path.join(prefix, filestring) filestrings.append(filestring) hdf = h5py.File(filestring, "w") group1 = hdf.create_group("group_xyz_Da_spec") group1.attrs["columns"] = ["x","y","z","Da","spec"] group1.attrs["spec_name_order"] = list(c) sublength_x= abs((max(sorted_df['z'])-min(sorted_df['z']))/self.params["n_big_slices"]) start = min(sorted_df['z']) end = min(sorted_df['z']) + sublength_x pbar = tqdm(range(self.params["n_big_slices"]), desc="Creating chunks") for i in pbar: temp = sorted_df[sorted_df['z'].between(start, end, inclusive="both")] group1.create_dataset("chunk_{}".format(i), data = temp.values) start += sublength_x end += sublength_x hdf.close() self.chunk_files = filestrings def get_big_slices_molecules(self): df_lst, files,ions,rrngs = self.get_apt_dataframe() for file_idx in range(len(files)): Org_file =df_lst[file_idx] atoms_spec = [] c = np.unique(rrngs.comp.values) for i in range(len(c)): print(c[i]) range_element = rrngs[rrngs['comp']=='{}'.format(c[i])] total, Count = atom_filter(Org_file,range_element) name = c[i] #name = i total["spec"] = [name for j in range(len(total))] atoms_spec.append(total) Df_atom_spec = pd.concat(atoms_spec) ############################# #check molecules: print("MOLECULE CHECK") molecule_check = np.array([len(c[i].split(" ")) for i in range(len(c))]) molecules = c[np.argwhere(molecule_check == 2)] Df_lst = [] for mol in molecules: spec_mol = mol[0].split(" ") Df = Df_atom_spec.loc[Df_atom_spec['spec'] == mol[0]].copy() Df["spec"] = [spec_mol[1]]*len(Df) Df_atom_spec.loc[Df_atom_spec['spec'] == mol[0], ['spec'] ] = spec_mol[0] Df_lst.append(Df) Df_lst.append(Df_atom_spec) Df_atom_spec = pd.concat(Df_lst) SpecSemi = np.unique(Df_atom_spec.spec.values) #check doubles print(SpecSemi) mol_doub_check = np.array([int(SpecSemi[j].split(":")[1]) for j in range(len(SpecSemi))]) mol_doub = SpecSemi[np.argwhere(mol_doub_check == 2)] Df_lst = [] for mol in mol_doub: spec_mol = mol[0].split(":")[0]+":1" Df = Df_atom_spec.loc[Df_atom_spec['spec'] == mol[0]].copy() Df["spec"] = [spec_mol]*len(Df) Df_atom_spec.loc[Df_atom_spec['spec'] == mol[0], ['spec'] ] = spec_mol Df_lst.append(Df) Df_lst.append(Df_atom_spec) Df_atom_spec = pd.concat(Df_lst) SpecFinal = np.unique(Df_atom_spec.spec.values) Df_spec_lst = [] for spec_ID in range(len(SpecFinal)): print( SpecFinal[spec_ID]) Df = Df_atom_spec.loc[Df_atom_spec['spec'] == SpecFinal[spec_ID]].copy() name = spec_ID Df["spec"] = [name for j in range(len(Df))] Df_spec_lst.append(Df) Df_atom_spec = pd.concat(Df_spec_lst) ############################ x_wu=Df_atom_spec sort_x = x_wu.sort_values(by=['z']) ## open hdf5 file #hdf = pd.HDFStore("./file_{}_large_chunks.h5".format(files[file_idx].replace(".","_"))) output_path = self.params['output_path'] + "/Output_big_slices.h5" hdf = h5py.File(output_path, "w") G1 = hdf.create_group("Group_xyz_Da_spec") G1.attrs["columns"] = ["x","y","z","Da","spec"] G1.attrs["spec_name_order"] = list(SpecFinal) ##end sublength_x= abs((max(sort_x['z'])-min(sort_x['z']))/self.params["n_big_slices"]) print(sublength_x) start = min(sort_x['z']) end = min(sort_x['z']) +sublength_x for i in tqdm(range(self.params["n_big_slices"])): #temp = sort_x.iloc[start:end] print(start) print(end) temp = sort_x[sort_x['z'].between(start, end, inclusive=True)] #temp.to_csv('B3_Hi_ent_cubes/{}.csv'.format(i), index=False) #hdf.put("chunk_{}".format(i), temp, format = "table", data_columns= True) ##Put data into hdf5 file G1.create_dataset("chunk_{}".format(i), data = temp.values) ##end start += sublength_x end += sublength_x #print(end) hdf.close() def get_voxels(self): """ Parameters ---------- Returns ------- Notes ----- """ filestrings = [] prefix = self.params['output_path'] size = self.params["voxel_size"] for filename in self.chunk_files: hdfr = h5py.File(filename, "r") filestring = filename.replace("large", "small") #filestring = os.path.join(prefix, filestring) filestrings.append(filestring) with h5py.File(filestring, "w") as hdfw: group_r = hdfr.get("group_xyz_Da_spec") group_keys = list(group_r.keys()) columns_r = list(list(group_r.attrs.values())[0]) group1 = hdfw.create_group("0") prev_attri =list(list(group_r.attrs.values())[0]) prev_attri.append("vox_file") group1.attrs["columns"] = prev_attri group1.attrs["spec_name_order"] = list(list(group_r.attrs.values())[1]) name_sub_file = 0 step = 0 m=0 pbar = tqdm(group_keys, desc="Getting Voxels") for key in pbar: read_array = np.array(group_r.get(key)) s= pd.DataFrame(data = read_array, columns = columns_r) x_min = round(min(s['x'])) x_max = round(max(s['x'])) y_min = round(min(s['y'])) y_max = round(max(s['y'])) z_min = round(min(s['z'])) z_max = round(max(s['z'])) p=[] x=[] for i in range(z_min, z_max, size): cubic = s[s['z'].between(i, i+size, inclusive="both")] for j in range(y_min, y_max, size): p = cubic[cubic['y'].between(j, j+size, inclusive="both")] for k in range(x_min, x_max, size): x = p[p['x'].between(k, k+size, inclusive="both")] if len(x['x'])>20: #warnings.warn("I am running some code with hardcoded numbers. Really recheck what's up here!") name ='cubes_z{}_x{}_y{}'.format(i,j,k).replace("-","m") if step>99999: step=0 m=m+1 group1 = hdfw.create_group("{}".format(100000*m)) x["vox_file"] = [name_sub_file for n_file in range(len(x))] group1.create_dataset("{}".format(name_sub_file), data = x.values) name_sub_file = name_sub_file+1 step=step+1 group1 = hdfw.get("0") group1.attrs["total_voxels"]="{}".format(name_sub_file) self.voxel_files = filestrings def calculate_voxel_composition(self, fileindex=0, outfilename="output_vox_ratio_composition.h5"): vox_ratio_files = [] for voxel_file in self.voxel_files: small_chunk_file_name = self.voxel_files[fileindex] hdf_sm_r = h5py.File(small_chunk_file_name, "r") group = hdf_sm_r.get("0") total_voxels =list(list(group.attrs.values())[2]) spec_lst_len = len(list(list(group.attrs.values())[1])) items = list(hdf_sm_r.items()) item_lst = [] ### CHECK THESE NUMBERS! for item in range(len(items)): item_lst.append([100000*(item), 100000*(item+1)]) item_lst = np.array(item_lst) total_voxels_int ="" for number in total_voxels: total_voxels_int = total_voxels_int + number total_voxels_int = int(total_voxels_int) files = [file_num for file_num in range(total_voxels_int)] spec_names = np.arange(spec_lst_len) dic_ratios = {} for spec_name in spec_names: dic_ratios["{}".format(spec_name)] = [] dic_ratios["Total_no"]=[] dic_ratios["file_name"]=[] dic_ratios["vox"] = [] ratios = [] f_count = 0 pbar = tqdm(files, desc="Calculating voxel composition") for filename in pbar: group = np.min(item_lst[[filename in range(j[0],j[1]) for j in item_lst]]) arr = np.array(hdf_sm_r.get("{}/{}".format(group,filename))[:,4]) N_x = len(arr) for spec in (spec_names): ratio = (len(np.argwhere(arr==spec)))/N_x dic_ratios["{}".format(spec)].append(ratio) dic_ratios["file_name"].append(filename) dic_ratios["vox"].append(f_count) dic_ratios["Total_no"].append(N_x) f_count = f_count+1 df = pd.DataFrame.from_dict(dic_ratios) output_path = os.path.join(self.params["output_path"], outfilename) with h5py.File(output_path, "w") as hdfw: hdfw.create_dataset("vox_ratios", data =df.drop("file_name", axis = 1).values ) hdfw.attrs["what"] = ["All the Vox ratios for a given APT smaple"] hdfw.attrs["howto_Group_name"] = ["Group_sm_vox_xyz_Da_spec/"] df_actual = df.drop("file_name", axis = 1) df_columns = list(df_actual.columns) hdfw.attrs["columns"]= df_columns hdf_sm_r.close() self.voxel_ratio_file = output_path
19,133
34.108257
151
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/segmentation.py
from compositionspace.datautils import DataPreparation from compositionspace.models import get_model from sklearn.decomposition import PCA from sklearn.mixture import GaussianMixture import json import h5py import numpy as np import pandas as pd import matplotlib.pylab as plt from tqdm.notebook import tqdm import os from pyevtk.hl import pointsToVTK from pyevtk.hl import gridToVTK#, pointsToVTKAsTIN import yaml import pyvista as pv class CompositionClustering(): def __init__(self, inputfile): if isinstance(inputfile, dict): self.params = inputfile else: with open(inputfile, "r") as fin: params = yaml.safe_load(fin) self.params = params self.version = "1.0.0" def get_PCA_cumsum(self, vox_ratio_file, vox_file): with h5py.File(vox_file,"r") as hdf: group = hdf.get("Group_sm_vox_xyz_Da_spec") group0 = hdf.get("0") spec_lst = list(list(group0.attrs.values())[1]) with h5py.File(vox_ratio_file , "r") as hdfr: ratios = np.array(hdfr.get("vox_ratios")) ratios_columns = list(list(hdfr.attrs.values())[0]) group_name = list(list(hdfr.attrs.values())[1]) print(len(ratios)) print((ratios_columns)) ratios = pd.DataFrame(data=ratios, columns=ratios_columns) X_train=ratios.drop(['Total_no','vox'], axis=1) PCAObj = PCA(n_components = len(spec_lst)) PCATrans = PCAObj.fit_transform(X_train) PCACumsumArr = np.cumsum(PCAObj.explained_variance_ratio_) plt.figure(figsize=(5,5)) plt.plot( range(1,len(PCACumsumArr)+1,1),PCACumsumArr,"-o") plt.ylabel("Explained Variance") plt.xlabel('Dimensions') plt.grid() output_path = os.path.join(self.params["output_path"], "PCA_cumsum.png") plt.savefig(output_path) plt.show() return PCACumsumArr, ratios def get_bics_minimization(self, vox_ratio_file, vox_file): with h5py.File(vox_file,"r") as hdf: group = hdf.get("Group_sm_vox_xyz_Da_spec") group0 = hdf.get("0") spec_lst = list(list(group0.attrs.values())[2]) with h5py.File(vox_ratio_file , "r") as hdfr: ratios = np.array(hdfr.get("vox_ratios")) ratios_columns = list(list(hdfr.attrs.values())[0]) group_name = list(list(hdfr.attrs.values())[1]) ratios = pd.DataFrame(data=ratios, columns=ratios_columns) gm_scores=[] aics=[] bics=[] X_train=ratios.drop(['Total_no','vox'], axis=1) n_clusters=list(range(1,self.params["bics_clusters"])) pbar = tqdm(n_clusters, desc="Clustering") for n_cluster in pbar: gm = GaussianMixture(n_components=n_cluster,verbose=0) gm.fit(X_train) y_pred=gm.predict(X_train) #gm_scores.append(homogeneity_score(y,y_pred)) aics.append(gm.aic(X_train)) bics.append(gm.bic(X_train)) output_path = os.path.join(self.params["output_path"], "bics_aics.png") plt.plot(n_clusters, aics, "-o",label="AIC") plt.plot(n_clusters, bics, "-o",label="BIC") plt.legend() plt.savefig(output_path) plt.show() return self.params["bics_clusters"], aics, bics def calculate_centroid(self, data): """ Calculate centroid Parameters ---------- data: pandas DataFrame or numpy array Returns ------- centroid """ if isinstance(data, pd.DataFrame): length = len(data['x']) sum_x = np.sum(data['x']) sum_y = np.sum(data['y']) sum_z = np.sum(data['z']) return sum_x/length, sum_y/length, sum_z/length else: length = len(data[:,0]) sum_x = np.sum(data[:,0]) sum_y = np.sum(data[:,1]) sum_z = np.sum(data[:,2]) return sum_x/length, sum_y/length, sum_z/length def get_voxel_centroid(self, vox_file, files_arr): with h5py.File(vox_file, "r") as hdf: items = list(hdf.items()) item_lst = [] for item in range(len(items)): item_lst.append([100000*(item),100000*(item+1)]) item_lst = np.array(item_lst) dic_centroids = {} dic_centroids["x"]=[] dic_centroids["y"]=[] dic_centroids["z"] = [] dic_centroids["file_name"] = [] df_centroids = pd.DataFrame(columns=['x', 'y', 'z','filename']) for filename in files_arr: group = np.min(item_lst[[filename in range(j[0],j[1]) for j in item_lst]]) xyz_Da_spec_atoms = np.array(hdf.get("{}/{}".format(group, filename))) x, y, z = self.calculate_centroid(xyz_Da_spec_atoms) dic_centroids["x"].append(x) dic_centroids["y"].append(y) dic_centroids["z"].append(z) dic_centroids["file_name"].append(filename) return dic_centroids def get_composition_cluster_files(self, vox_ratio_file, vox_file, n_components): ml_params = self.params["ml_models"] with h5py.File(vox_file,"r") as hdf: group = hdf.get("Group_sm_vox_xyz_Da_spec") group0 = hdf.get("0") spec_lst = list(list(group0.attrs.values())[2]) with h5py.File(vox_ratio_file , "r") as hdfr: ratios = np.array(hdfr.get("vox_ratios")) ratios_columns = list(list(hdfr.attrs.values())[0]) group_name = list(list(hdfr.attrs.values())[1]) ratios = pd.DataFrame(data=ratios, columns=ratios_columns) X_train=ratios.drop(['Total_no','vox'], axis=1) gm = get_model(ml_params=ml_params) gm.fit(X_train) y_pred=gm.predict(X_train) cluster_lst = [] for phase in range(n_components): cluster_lst.append(np.argwhere(y_pred == phase).flatten()) df_lst = [] for cluster in cluster_lst: df_lst.append(ratios.iloc[cluster]) #sorting cluster_lst_sort = [] len_arr = np.array([len(x) for x in cluster_lst]) sorted_len_arr = np.sort(len_arr) for length in sorted_len_arr: cluster_lst_sort.append(cluster_lst[np.argwhere(len_arr == length)[0,0]]) #print([len(x) for x in cluster_lst_sort]) cluster_lst = cluster_lst_sort return cluster_lst, ratios def get_composition_clusters(self, vox_ratio_file, vox_file, outfile="vox_centroid_file.h5"): voxel_centroid_output_file = [] n_components = self.params["n_phases"] ml_params = self.params["ml_models"] cluster_lst, ratios = self.get_composition_cluster_files(vox_ratio_file, vox_file, n_components) plot_files = [] for phase in range(len(cluster_lst)): cluster_files = [] cluster = cluster_lst[phase] for voxel_id in cluster: cluster_files.append(ratios['vox'][voxel_id]) plot_files.append(cluster_files) plot_files_group = [] for cluster_files in plot_files: plot_files_group.append([int(file_num) for file_num in cluster_files ]) with h5py.File(vox_file,"r") as hdf_sm_r: hdf_sm_r = h5py.File(vox_file,"r") group = hdf_sm_r.get("0") #total_voxels =list(list(group.attrs.values())[0]) total_voxels =list(list(group.attrs.values())[2]) total_voxels_int ="" for number in total_voxels: total_voxels_int = total_voxels_int + number total_voxels_int = int(total_voxels_int) hdf_sm_r.close() plot_files_cl_All_group = [file_num for file_num in range(total_voxels_int)] plot_files_group.append(plot_files_cl_All_group) output_path = os.path.join(self.params["output_path"], outfile) with h5py.File(output_path,"w") as hdfw: for cluster_file_id in range(len(plot_files_group)): G = hdfw.create_group(f"{cluster_file_id}") G.attrs["what"] = ["Centroid of voxels"] G.attrs["howto_Group_name"] = ["Group_sm_vox_xyz_Da_spec/"] G.attrs["colomns"] = ["x","y","z","file_name"] CentroidsDic = self.get_voxel_centroid(vox_file, plot_files_group[cluster_file_id]) G.create_dataset(f"{cluster_file_id}", data = pd.DataFrame.from_dict(CentroidsDic).values) self.voxel_centroid_output_file = output_path def generate_plots(self): vtk_files = [] with h5py.File(self.voxel_centroid_output_file, "r") as hdfr: groups =list(hdfr.keys()) for group in range(len(groups)-1): phase_arr = np.array(hdfr.get(f"{group}/{group}")) phase_columns = list(list(hdfr.get(f"{group}").attrs.values())[0]) phase_cent_df =pd.DataFrame(data=phase_arr, columns=phase_columns) image = phase_cent_df.values file_path = self.voxel_centroid_output_file + f"_{group}" vtk_files.append(file_path + ".vtu") x = np.ascontiguousarray(image[:,0]) y= np.ascontiguousarray(image[:,1]) z = np.ascontiguousarray(image[:,2]) label = np.ascontiguousarray( image[:,3]) pointsToVTK(file_path, x, y, z, data = {"label" : label} ) self.vtk_files = vtk_files def plot3d(self, **kwargs): self.generate_plots() for file in self.vtk_files: grid = pv.read(file) grid.plot(**kwargs, jupyter_backend="panel")
10,121
35.541516
106
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/paraprobe_transcoder.py
# -*- coding: utf-8 -*- """ Reader for the APSuite6/IVAS4 *.APT file format MK::GPLV3, 03/09/2020, Markus K\"uhbach, m.kuehbach@mpie.de """ import numpy as np #https://www.python-kurs.eu/numpy_dtype.php class APTFileBranches(): def __init__(self): self.dict_kwnsect = { 1: 'tof', 2: 'pulse', 3: 'freq', 4: 'tElapsed', 5: 'erate', 6: 'tstage', 7: 'TargetErate', 8: 'TargetFlux', 9: 'pulseDelta', 10: 'Pres', 11: 'VAnodeMon', 12: 'Temp', 13: 'AmbTemp', 14: 'FractureGuard', 15: 'Vref', 16: 'Noise', 17: 'Uniformity', 18: 'xstage', 19: 'ystage', 20: 'zstage', 21: 'z', 22: 'tofc', 23: 'Mass', 24: 'tofb', 25: 'xs', 26: 'ys', 27: 'zs', 28: 'rTip', 29: 'zApex', 30: 'zSphereCorr', 31: 'XDet_mm', 32: 'YDet_mm', 33: 'Multiplicity', 34: 'Vap', 35: 'Detector Coordinates', 36: 'Position'} self.dict_sectionid = { 'Failure': 0, 'tof': 1, 'pulse': 2, 'freq': 3, 'tElapsed': 4, 'erate': 5, 'tstage': 6, 'TargetErate': 7, 'TargetFlux': 8, 'pulseDelta': 9, 'Pres': 10, 'VAnodeMon': 11, 'Temp': 12, 'AmbTemp': 13, 'FractureGuard': 14, 'Vref': 15, 'Noise': 16, 'Uniformity': 17, 'xstage': 18, 'ystage': 19, 'zstage': 20, 'z': 21, 'tofc': 22, 'Mass': 23, 'tofb': 24, 'xs': 25, 'ys': 26, 'zs': 27, 'rTip': 28, 'zApex': 29, 'zSphereCorr': 30, 'XDet_mm': 31, 'YDet_mm': 32, 'Multiplicity': 33, 'Vap': 34, 'Detector Coordinates': 35, 'Position': 36 } self.dict_iHeaderSize = { 1: 148, 2: 148, 3: 148, 4: 148, 5: 148, 6: 148, 7: 148, 8: 148, 9: 148, 10: 148, 11: 148, 12: 148, 13: 148, 14: 148, 15: 148, 16: 148, 17: 148, 18: 148, 19: 148, 20: 148, 21: 148, 22: 148, 23: 148, 24: 148, 25: 148, 26: 148, 27: 148, 28: 148, 29: 148, 30: 148, 31: 148, 32: 148, 33: 148, 34: 148, 35: 148, 36: 148+6*4 } self.dict_iHeaderVersion = { 1: 2, 2: 2, 3: 2, 4: 2, 5: 2, 6: 2, 7: 2, 8: 2, 9: 2, 10: 2, 11: 2, 12: 2, 13: 2, 14: 2, 15: 2, 16: 2, 17: 2, 18: 2, 19: 2, 20: 2, 21: 2, 22: 2, 23: 2, 24: 2, 25: 2, 26: 2, 27: 2, 28: 2, 29: 2, 30: 2, 31: 2, 32: 2, 33: 2, 34: 2, 35: 2, 36: 2 } self.dict_iSectionVersion = { 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1, 7: 1, 8: 1, 9: 1, 10: 1, 11: 1, 12: 1, 13: 1, 14: 1, 15: 1, 16: 1, 17: 1, 18: 1, 19: 1, 20: 1, 21: 1, 22: 1, 23: 1, 24: 1, 25: 1, 26: 1, 27: 1, 28: 1, 29: 1, 30: 1, 31: 1, 32: 1, 33: 1, 34: 1, 35: 1, 36: 1 } self.dict_eRelationshipType = { 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1, 7: 1, 8: 1, 9: 1, 10: 1, 11: 1, 12: 1, 13: 1, 14: 1, 15: 1, 16: 1, 17: 1, 18: 1, 19: 1, 20: 1, 21: 1, 22: 1, 23: 1, 24: 1, 25: 1, 26: 1, 27: 1, 28: 1, 29: 1, 30: 1, 31: 1, 32: 1, 33: 1, 34: 1, 35: 1, 36: 1 } self.dict_eRecordType = { 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1, 7: 1, 8: 1, 9: 1, 10: 1, 11: 1, 12: 1, 13: 1, 14: 1, 15: 1, 16: 1, 17: 1, 18: 1, 19: 1, 20: 1, 21: 1, 22: 1, 23: 1, 24: 1, 25: 1, 26: 1, 27: 1, 28: 1, 29: 1, 30: 1, 31: 1, 32: 1, 33: 1, 34: 1, 35: 1, 36: 1 } self.dict_eRecordDataType = { 1: 3, 2: 3, 3: 3, 4: 3, 5: 3, 6: 2, 7: 3, 8: 3, 9: 1, 10: 3, 11: 3, 12: 3, 13: 3, 14: 2, 15: 3, 16: 3, 17: 3, 18: 1, 19: 1, 20: 1, 21: 1, 22: 3, 23: 3, 24: 3, 25: 3, 26: 3, 27: 3, 28: 3, 29: 3, 30: 3, 31: 3, 32: 3, 33: 1, 34: 3, 35: 3, 36: 3} self.dict_iDataTypeSize = { 1: 32, 2: 32, 3: 32, 4: 32, 5: 32, 6: 16, 7: 32, 8: 32, 9: 16, 10: 32, 11: 32, 12: 32, 13: 32, 14: 16, 15: 32, 16: 32, 17: 32, 18: 32, 19: 32, 20: 32, 21: 64, 22: 32, 23: 32, 24: 32, 25: 32, 26: 32, 27: 32, 28: 32, 29: 32, 30: 32, 31: 32, 32: 32, 33: 32, 34: 32, 35: 32, 36: 32 } self.dict_iRecordSize = { 1: 4, 2: 4, 3: 4, 4: 4, 5: 4, 6: 2, 7: 4, 8: 4, 9: 2, 10: 4, 11: 4, 12: 4, 13: 4, 14: 2, 15: 4, 16: 4, 17: 4, 18: 4, 19: 4, 20: 4, 21: 8, 22: 4, 23: 4, 24: 4, 25: 4, 26: 4, 27: 4, 28: 4, 29: 4, 30: 4, 31: 4, 32: 4, 33: 4, 34: 4, 35: 8, 36: 12 } self.dict_iElements = dict() for i in range(1,len(self.dict_kwnsect)+1): self.dict_iElements[i] = int(self.dict_iRecordSize[i] / (self.dict_iDataTypeSize[i]/8)) class APTFileHeader(): def __init__(self, fn, *args, **kwargs): self.known_sections = APTFileBranches() self.healthy = True self.aptfn = fn self.cSignature = "" self.iHeaderSize = 0 self.iHeaderVersion = 0 self.wcFilename = "" self.ftCreationTime = 0 self.llIonCount = 0 def read_cameca_apt_file_header(self, fid): ht = np.dtype([('cSignature', np.int8, (4,)), ('iHeaderSize', np.int32), ('iHeaderVersion', np.int32), ('wcFilename', np.uint16, 256), ('ftCreationTime', np.uint64), ('llIonCount', np.uint64)]) tmp = np.fromfile( fid , ht, count = 1 ) #unicode_, 256), str_parse = "" for i in tmp['cSignature'].flatten(): if chr(i) != '\x00': str_parse = str_parse + chr(i) self.cSignature = str_parse if self.cSignature != 'APT': # print('The file header indicates the file is not a valid *.APT file!') self.healthy = False return None self.iHeaderSize = tmp['iHeaderSize'].flatten()[0] if self.iHeaderSize != 540: # print('The iHeaderSize is unexpectedly different!') self.healthy = False return None self.iHeaderVersion = tmp['iHeaderVersion'].flatten()[0] if self.iHeaderVersion != 2: # print('The iHeaderVersion is not 2 but only 2 is supported by this implementation!') self.healthy = False return None ###MK::parsing UTF-16 works currently only for the lower bit, for the #APTV2 draft specification this is not a problem because currently the internal ##section identifiers use only UTF-8 characters str_parse = "" for i in tmp['wcFilename'].flatten(): if chr(i) != '\x00': str_parse = str_parse + chr(i) self.wcFilename = str_parse self.ftCreationTime = tmp['ftCreationTime'].flatten()[0] self.llIonCount = tmp['llIonCount'].flatten()[0] if self.llIonCount < 1: # print('llIonCount < 1 means there are no ions in the dataset which is weird!') self.healthy = False return None # print('*.APT file ' + self.aptfn + ' successfully read ' + str(self.llIonCount) + ' ions') tmp = [] str_parse = "" def print(self): print('APTFileHeader') print('healthy ' + str(self.healthy)) print('aptfn ' + self.aptfn) print('cSignature ' + self.cSignature) print('iHeaderSize ' + str(self.iHeaderSize)) print('iHeaderVersion ' + str(self.iHeaderVersion)) print('wcFilename ' + self.wcFilename) print('ftCreationTime ' + str(self.ftCreationTime)) print('llIonCount ' + str(self.llIonCount)) #def get_offset(self): # return 540 class APTSectionHeader(): def __init__(self, *args, **kwargs): #fn self.healthy = True self.known_sections = APTFileBranches() #self.aptfn = fn self.cSignature = "" self.iHeaderSize = 0 self.iHeaderVersion = 0 self.wcSectionType = "" self.iSectionVersion = 0 self.eRelationshipType = 0 self.eRecordType = 0 self.eRecordDataType = 0 self.iDataTypeSize = 0 self.iRecordSize = 0 self.wcDataUnit = "" self.llRecordCount = 0 self.llByteCount = 0 def read_cameca_apt_section_header_auto(self, fid ): ht = np.dtype([('cSignature', np.int8, (4,)), ('iHeaderSize', np.int32), ('iHeaderVersion', np.int32), ('wcSectionType', np.uint16, 32), ('iSectionVersion', np.int32), ('eRelationshipType', np.uint32), ('eRecordType', np.uint32), ('eRecordDataType', np.uint32), ('iDataTypeSize', np.int32), ('iRecordSize', np.int32), ('wcDataUnit', np.uint16, 16), ('llRecordCount', np.uint64), ('llByteCount', np.uint64)]) #skip the APTFileHeader and start reading the first section #f = open( self.aptfn, "rb" ) #f.seek(offset, os.SEEK_SET) #seek set is relative to beginning of the file tmp = np.fromfile( fid , ht, count = 1 ) #unicode_, 256), str_parse = "" for i in tmp['cSignature'].flatten(): if chr(i) != '\x00': str_parse = str_parse + chr(i) self.cSignature = str_parse if self.cSignature != 'SEC': # print('The section header is faulty, it misses the required SEC keyword!') self.healthy = False return None ###MK::parsing UTF-16 works currently only for the lower bit, for the #APTV2 draft specification this is not a problem because currently the internal ##section identifiers use only UTF-8 characters str_parse = "" for i in tmp['wcSectionType'].flatten(): if chr(i) != '\x00': str_parse = str_parse + chr(i) self.wcSectionType = str_parse # print('__' + self.wcSectionType + '__') sectionid = self.known_sections.dict_sectionid.get(self.wcSectionType) if sectionid == None: # print('Encountering an unknown section!') self.healthy = False return None ###MK::here HeaderSize and Version tested after reading wcSectionType self.iHeaderSize = tmp['iHeaderSize'].flatten()[0] if self.iHeaderSize != self.known_sections.dict_iHeaderSize[sectionid]: # print('iHeaderSize is ' + str(self.iHeaderSize) + ' this does not match the expectation!') self.healthy = False return None self.iHeaderVersion = tmp['iHeaderVersion'].flatten()[0] if self.iHeaderVersion != self.known_sections.dict_iHeaderVersion[sectionid]: # print('iHeaderVersion is ' + str(self.iHeaderVersion) + ' this does not match the expectation!') self.healthy = False return None self.iSectionVersion = tmp['iSectionVersion'].flatten()[0] if self.iSectionVersion != self.known_sections.dict_iSectionVersion[sectionid]: # print('iSectionVersion is ' + str(self.iSectionVersion) + ' this does not match the expectation!') self.healthy = False return None self.eRelationshipType = tmp['eRelationshipType'].flatten()[0] if self.eRelationshipType != self.known_sections.dict_eRelationshipType[sectionid]: # print('eRelationshipType is ' + str(self.eRelationshipType) + ' this does not match the expectation!') self.healthy = False return None self.eRecordType = tmp['eRecordType'].flatten()[0] if self.eRecordType != self.known_sections.dict_eRecordType[sectionid]: # print('eRecordType is ' + str(self.eRecordType) + ' this does not match the expectation!') self.healthy = False return None self.eRecordDataType = tmp['eRecordDataType'].flatten()[0] if self.eRecordDataType != self.known_sections.dict_eRecordDataType[sectionid]: # print('eRecordDataType is ' + str(self.eRecordDataType) + ' this does not match the expectation!') self.healthy = False return None self.iDataTypeSize = tmp['iDataTypeSize'].flatten()[0] if self.iDataTypeSize != self.known_sections.dict_iDataTypeSize[sectionid]: # print('iDataTypeSize is ' + str(self.iDataTypeSize) + ' this does not match the expectation!') self.healthy = False return None self.iRecordSize = tmp['iRecordSize'].flatten()[0] if self.iRecordSize != self.known_sections.dict_iRecordSize[sectionid]: # print('iRecordSize is ' + str(self.iRecordSize) + ' this does not match the expectation!') self.healthy = False return None str_parse = "" for i in tmp['wcDataUnit'].flatten(): if chr(i) != '\x00': str_parse = str_parse + chr(i) self.wcDataUnit = str_parse self.llRecordCount = tmp['llRecordCount'].flatten()[0] ##MK::test implementatio remaining self.llByteCount = tmp['llByteCount'].flatten()[0] # print('*.APT file successfully read ' + str(sectionid) + ' section header') tmp = [] str_parse = "" ##MK::implement checks def print(self): print('APTSectionHeader') #print(self.aptfn) print('cSignature ' + self.cSignature) print('iHeaderSize ' + str(self.iHeaderSize)) print('iHeaderVersion ' + str(self.iHeaderVersion)) print('wcSectionType ' + self.wcSectionType) print('iSectionVersion ' + str(self.iSectionVersion)) print('eRelationshipType ' + str(self.eRelationshipType)) print('eRecordType ' + str(self.eRecordType)) print('eRecordDataType ' + str(self.eRecordDataType)) print('iDataTypeSize ' + str(self.iDataTypeSize)) print('iRecordSize ' + str(self.iRecordSize)) print('wcDataUnit ' + self.wcDataUnit) print('llRecordCount ' + str(self.llRecordCount)) print('llByteCount ' + str(self.llByteCount)) #def get_offset(self): # if self.wcSectionType == "Position": # return 148 #(148+6*4) # else: # return 148 class paraprobe_transcoder(): def __init__(self, aptfn, *args, **kwargs): self.fn = aptfn self.filestream = None self.header = APTFileHeader( self.fn ) self.idtfyd_sections = [] #self.mass_sect = APTSectionHeader( self.fn ) #self.xyz_sect = APTSectionHeader( self.fn ) def read_cameca_apt_section_data_fixed_onetoone(self, fid, sect ): #which datatype to use, interpret from the section header ni = int(self.header.llIonCount) # print('ni ' + str(ni)) nj = int(int(sect.iRecordSize) / (int(sect.iDataTypeSize)/int(8))) # print('nj ' + str(nj)) dtyp = None #eRecordDataType == 3 and iDataTypeSize == 32 if sect.eRecordDataType == 1: if sect.iDataTypeSize/8 == 2: dtyp = np.int16 elif sect.iDataTypeSize/8 == 4: dtyp = np.int32 elif sect.iDataTypeSize/8 == 8: dtyp = np.int64 else: # print('Unknown datatype selection!') return None elif sect.eRecordDataType == 2: ###MK::currently only one type used uint16 dtyp = np.uint16 elif sect.eRecordDataType == 3: dtyp = np.float32 else: # print('Unknown datatype selection!') return None # print(dtyp) return np.reshape(np.fromfile(fid, dtyp, count = int(ni*nj)), (int(ni), int(nj)), order='C') def read_cameca_apt(self): offset = 0 # print('Offset is ' + str(offset)) self.filestream = open(self.fn, "rb") self.header.read_cameca_apt_file_header( self.filestream ) # self.header.print() #'Failure' is sectionid so one more keys, currently 36+1 for sectionid in range(1,len(self.header.known_sections.dict_sectionid)): sect = APTSectionHeader() sect.read_cameca_apt_section_header_auto( self.filestream ) # sect.print() if sect.healthy == True: #read data of the section #large switch ##### #what type of section is this? sectionid = self.header.known_sections.dict_sectionid.get(sect.wcSectionType, 0) #0 is for 'Failure'# # print('sectionid ' + str(sectionid)) self.idtfyd_sections.append( sect ) ###MK implement more elegantly using switchers #MK::here we use the fact that the file pointer gets advanced implicitly so after reading the sect it points to the #first byte of the data section corresponding to the data associated with this section #this works because *.APT files by definition have the data following immediately after each section headers #with ##MK::at least as of 2020/03/20 no gaps in the binary structure if sectionid == 1: self.tof = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 2: self.pulse = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 3: self.freq = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 4: self.tElapsed = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 5: self.erate = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 6: self.tstage = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 7: self.TargetErate = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 8: self.TargetFlux = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 9: self.pulseDelta = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 10: self.Pres = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 11: self.VAnodeMon = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 12: self.Temp = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 13: self.AmbTemp = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 14: self.FractureGuard = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 15: self.Vref = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 16: self.Noise = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 17: self.Uniformity = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 18: self.xstage = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 19: self.ystage = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 20: self.zstage = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 21: self.z = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 22: self.tofc = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 23: self.Mass = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 24: self.tofb = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 25: self.xs = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 26: self.ys = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 27: self.zs = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 28: self.rTip = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 29: self.zApex = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 30: self.zSphereCorr = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 31: self.XDet_mm = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 32: self.YDet_mm = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 33: self.Multiplicity = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 34: self.Vap = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 35: self.DetectorCoordinates = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) elif sectionid == 36: #position has preceeding 6 float32 for bounds self.tipbox = np.reshape(np.fromfile( self.filestream, np.float32, count = int(2*3) ), (int(2), int(3) ), order='C') self.Position = self.read_cameca_apt_section_data_fixed_onetoone( self.filestream, sect ) else: # print('Section is faulty, which is why we stop reading!') self.filestream.close() break else: # print('Reached the end of the file!') self.filestream.close() break #done with all I/O on the *.APT file so close it self.filestream.close() print('Done, reading *.APT file') # self.header.read_cameca_apt_file_header() # offset = offset + self.header.get_offset() # print('Offset is ' + str(offset)) # self.mass_sect.read_cameca_apt_section_header( offset ) # offset = offset + self.mass_sect.get_offset() # print('Offset is ' + str(offset)) # # #load mass # f = open(self.fn, "rb") # f.seek(offset, os.SEEK_SET) # self.mq = [] # self.mq = np.fromfile(f, np.float32, count = int(self.header.llIonCount)*1) # offset = offset + int(self.header.llIonCount) * 1 * 4 # print('Offset is ' + str(offset)) # print('Mass data read successfully!') # print(np.shape(self.mq)) # # self.xyz_sect.read_cameca_apt_section_header( offset ) # offset = offset + self.mass_sect.get_offset() # print('Offset is ' + str(offset)) # #read x,y,z limits # f = open(self.fn, "rb") # f.seek(offset, os.SEEK_SET) # self.tipbox = np.fromfile(f, np.float32, count = 6) # offset = offset + 6*4 # print('Offset is ' + str(offset)) # #read x,y,z coordinates # f = open(self.fn, "rb") # f.seek(offset, os.SEEK_SET) # self.xyz = [] # self.xyz = np.reshape(np.fromfile(f, np.float32, count = int(self.header.llIonCount)*3), # (int(self.header.llIonCount),3), order='C') # print('Position coordinates read successfully') # print(np.shape(self.xyz)) # offset = offset + int(self.header.llIonCount)*3*4 # print('Offset is ' + str(offset)) # def write(self, file_name: str): # #file_path = Path(file_name) # #file_path.write_text(self.to_utf8()) # with open(file_name,'w') as f: # f.write(self.xml) #example how to read a APT file from APSuite6/IVAS4 #minimal example with default output currently implemented after reconstruction wizard from APSuite6/IVAS4 #Tk().withdraw() #fn = askopenfilename() #maximal example with all branches of an *.APT file exported, be careful --- in APSuite6/IVAS4 this is super slow (40min for 8mio atoms! below) #fn = 'Z:/GITHUB/MPIE_APTFIM_TOOLBOX/paraprobe/code/481f2262-b0d5-4dfb-91b7-fb31b57cd1b0.apt' #apt = paraprobe_transcoder( fn ) #apt.read_cameca_apt() #np.shape(apt.Mass) #verbose #apt.header.print() #apt.mass_sect.print() #apt.xyz_sect.print()
26,191
50.66075
161
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/models.py
from ast import Raise from sklearn.ensemble import RandomForestClassifier from sklearn.mixture import GaussianMixture from sklearn.cluster import DBSCAN def get_model(ml_params): """ get machine learning model for clustering """ model_name = ml_params["name"] model_params = ml_params[model_name] _available_models = ["randomforest"] if(model_name == "randomforest"): model = RandomForestClassifier(max_depth=model_params["max_depth"], n_estimators=model_params["n_estimators"]) return model if(model_name == "GaussianMixture"): model = GaussianMixture(n_components = model_params["n_components"], max_iter=model_params["max_iter"],verbose=model_params["verbose"]) return model if(model_name == "DBScan"): model = DBSCAN(eps= self.params["eps"], min_samples= model_params["min_samples"]) return model else: raise ValueError(f"Now implementation is found for the model {model_name}, choose from: {_available_models}")
1,067
37.142857
144
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/__init__.py
0
0
0
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/compositionspace/postprocessing.py
import pandas as pd import os from tqdm.notebook import tqdm import matplotlib.pyplot as plt import numpy as np import h5py from sklearn.cluster import DBSCAN from pyevtk.hl import pointsToVTK from pyevtk.hl import gridToVTK import yaml class DataPostprocess: def __init__(self, inputfile): if isinstance(inputfile, dict): self.params = inputfile else: with open(inputfile, "r") as fin: params = yaml.safe_load(fin) self.params = params self.version = "1.0.0" def get_post_centroids(self, voxel_centroid_phases_files, cluster_id): """ Reads the voxel centroids for a phase Parameters ---------- voxel_centroid_phases_files : str, Voxel cetroids corresponding to each phase cluster_id: int, phase id in Voxel_centroid_phases_files Returns ------- pandas dataframe for voxel cetroids(x,y,z, voxelId), pandas dataframe for voxel cetroids(x,y,z), list of coloumn names (x,y,z, voxelId) Notes ----- input is taken from composition space based segmentation of phases. """ with h5py.File(voxel_centroid_phases_files , "r") as hdfr: group = cluster_id Phase_arr = np.array(hdfr.get(f"{group}/{group}")) Phase_columns = list(list(hdfr.get(f"{group}").attrs.values())[0]) Phase_cent_df =pd.DataFrame(data=Phase_arr, columns=Phase_columns) Df_centroids = Phase_cent_df.copy() Df_centroids_no_files = Df_centroids.drop(['file_name'] , axis=1) files = Df_centroids['file_name'] return Df_centroids_no_files, Df_centroids, Phase_columns def DBSCAN_clustering(self, voxel_centroid_phases_files, cluster_id, plot= False, plot3d = False, save =False): """ Get individual clusters or precipitates corresponding to each phase/ chemical domain. DBSCAN is applied on the centroids of the voxels helping to remove noisy voxels around clusters. Parameters ---------- cluster_id: int,id of the phase/chemical domain in Output_voxel_cetroids_phases.h5 eps: float,epsilon is a hyperparameter for DBSCAN.The maximum distance between two samples for one to be considered as in the neighborhood of the other. This is not a maximum bound on the distances of points within a cluster. min_smaples: int, min_smaples is a hyperparameter for DBSCAN. The number of samples (or total weight) in a neighborhood for a point to be considered as a core point. This includes the point itself. plot: boolean, if true plots the histogram of the cluster for found by the DBSCAN algorithm. plot3d: boolean, if true plots voxel centroids in corresponding to each precipitate and outputs a .vtu file . save: boolean, saves a .h5 file corresponding to a cluster_id containing centroids for each precipitate. Returns ------- Voxel centroids corresponding to each precipitate. Notes ----- input is taken from composition space based segmentation of phases. """ eps = self.params["ml_models"]["DBScan"]["eps"] min_samples = self.params["ml_models"]["DBScan"]["min_samples"] Df_centroids_no_files, Df_centroids, Phase_columns = self.get_post_centroids( voxel_centroid_phases_files ,cluster_id) db = DBSCAN(eps=eps, min_samples= min_samples).fit(Df_centroids_no_files.values) #eps=5., min_samples= 35 core_samples_mask = np.zeros_like(db.labels_, dtype=bool) core_samples_mask[db.core_sample_indices_] = True labels = db.labels_ # Number of clusters in labels, ignoring noise if present. n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0) n_noise_ = list(labels).count(-1) if plot == True: plt.hist(labels,bins = 100); cluster_combine_lst = [] for i in np.unique(labels): if i !=-1: cl_idx =np.argwhere(labels==i).flatten() cl_cent=Df_centroids_no_files.iloc[cl_idx] cl_cent["ID"] = [i]*len(cl_cent) cluster_combine_lst.append(cl_cent) if plot3d == True: OutFile = os.path.join(self.params["output_path"], f"Output_DBSCAN_segmentation_phase{cluster_id}") Df_comb = pd.concat(cluster_combine_lst) image = Df_comb.values x = np.ascontiguousarray(image[:,0]) y= np.ascontiguousarray(image[:,1]) z = np.ascontiguousarray(image[:,2]) label = np.ascontiguousarray( image[:,3]) pointsToVTK(OutFile,x,y,z, data = {"label" : label} ) if save == True: OutFile = os.path.join(self.params["output_path"], f"Output_DBSCAN_segmentation_phase{cluster_id}") with h5py.File(OutFile, "w") as hdfw: G = hdfw.create_group(f"{cluster_id}") G.attrs["columns"] = Phase_columns for i in tqdm(np.unique(labels)): if i !=-1: cl_idx =np.argwhere(labels==i).flatten() cl_cent=Df_centroids.iloc[cl_idx] G.create_dataset("{}".format(i), data = cl_cent.values)
5,481
38.157143
126
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/tests/test_read_files.py
import pytest import numpy as np import os import sys from compositionspace.datautils import DataPreparation def test_file_rrng(): data = DataPreparation("tests/experiment_params.yaml") datarrng = data.get_rrng("tests/data/R31_06365-v02.rrng") assert datarrng[0]["name"].values[0] == "C" def test_file_pos(): data = DataPreparation("tests/experiment_params.yaml") datapos = data.get_pos("tests/data/R31_06365-v02.pos") assert np.isclose(datapos[0][0]+5.3784895, 0) def test_file_df(): data = DataPreparation("tests/experiment_params.yaml") data = data.get_apt_dataframe() assert np.isclose(data[0][0]["x"].values[0]+5.3784895, 0) assert data[1][0] == 'R31_06365-v02.pos' assert data[2]["name"].values[0] == "C" assert np.isclose(data[3]["lower"].values[0]-5.974, 0) def test_chunkify(): data = DataPreparation("tests/experiment_params.yaml") data.get_big_slices() assert os.path.exists(data.chunk_files[0]) == True def test_voxelise(): data = DataPreparation("tests/experiment_params.yaml") data.get_big_slices() data.get_voxels() data.calculate_voxel_composition() assert os.path.exists(data.voxel_ratio_file) == True
1,213
31.810811
61
py
CompositionSpaceNFDI
CompositionSpaceNFDI-main/docs/source/conf.py
# Configuration file for the Sphinx documentation builder. # # For the full list of built-in configuration values, see the documentation: # https://www.sphinx-doc.org/en/master/usage/configuration.html # -- Project information ----------------------------------------------------- # https://www.sphinx-doc.org/en/master/usage/configuration.html#project-information import os import sys import shutil import glob sys.path.insert(0, os.path.abspath('../../compositionspace/')) def skip(app, what, name, obj, would_skip, options): if name in ( '__init__',): return False return would_skip def setup(app): app.connect('autodoc-skip-member', skip) if os.path.exists("example"): shutil.rmtree("example") shutil.copytree("../../example", "example") project = 'compositionspace' copyright = '2022, Alaukik Saxena, Sarath Menon' author = 'Alaukik Saxena, Sarath Menon' extensions = [ 'sphinx.ext.autodoc', 'sphinx.ext.coverage', 'sphinx.ext.mathjax', 'sphinx.ext.viewcode', 'sphinx.ext.githubpages', 'sphinx.ext.napoleon', 'm2r2', 'sphinx_markdown_tables', 'nbsphinx', ] html_theme = 'furo' html_theme_options = { #'logo_only' : True, #'canonical_url' : 'https://calphy.readthedocs.io/', } html_extra_path = ['../_static' ] source_suffix = ['.rst', '.md'] exclude_patterns = []
1,349
23.545455
83
py
fuzzyJoiner
fuzzyJoiner-master/build_model.py
from random import shuffle import pickle import numpy as np # import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics from scipy.stats.mstats import gmean import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_PRECOMPUTED_SPLIT = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False filepath="weights.best.hdf5" output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print(str("failed to find embedding for:" + word).encode('utf-8')) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath, 'r', encoding='utf8') as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r', encoding='utf8') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2, val_split = 0.2, base_file=None): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_test_samples = int(test_split * len(ents)) num_validation_samples = int(val_split * len(ents)) if USE_PRECOMPUTED_SPLIT: train = pickle.load(open((base_file + '.train'), "rb" )) test = pickle.load(open((base_file + '.test'), "rb" )) validation = pickle.load(open((base_file + '.validation'), "rb" )) else: train = ents[:-(num_validation_samples + num_test_samples)] test = ents[-num_test_samples:] validation = ents[-(num_validation_samples + num_test_samples):-num_test_samples] pickle.dump(train, open(base_file + '.train', 'wb')) pickle.dump(test, open(base_file + '.test', 'wb')) pickle.dump(validation, open(base_file + '.validation', 'wb')) return train, test, validation """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = K.square(y_pred[:,0,0]) a_n_distance = K.square(y_pred[:,1,0]) p_n_distance = K.square(y_pred[:,2,0]) phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_semi_hard_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees triplets = {} triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] expected_text = set(entity2same[key]) expected_ids = [entity2unique[i] for i in expected_text] for positive in expected_text: k = entity2unique[positive] nearest = t.get_nns_by_vector(predictions[k], NNlen) dist_k = t.get_distance(index, k) semi_hards = [] for n in nearest: if n == index or n in expected_ids or n == k: continue n_dist = t.get_distance(index, n) if n_dist > dist_k: semi_hards.append(unique_text[n]) # shuffle(semi_hards) # semi_hards = semi_hards[0:20] for i in semi_hards: triplets['anchor'].append(key) triplets['positive'].append(unique_text[k]) triplets['negative'].append(i) return triplets def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = t.get_distance(index, entity2unique[i]) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = t.get_distance(index, entity2unique[j]) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = t.get_distance(index, entity2unique[p]) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 closest_positive_counts.append(closest_pos_count / min(len(expected_text), NNlen - 1)) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) print("Precision at 1: " + str(precise / len(entity2same))) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w', encoding='utf8') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') parser.add_argument('--model', type=str, help='name for model file') parser.add_argument('--use_precomputed_split', action='store_true', help='load precomputed split for test and validation data') parser.add_argument('--data_path', type=str, help='location to store/load data from') args = parser.parse_args() filepath = args.model LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) USE_PRECOMPUTED_SPLIT = args.use_precomputed_split if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test, validation = split(entities, test_split = .20, val_split = 0.20 , base_file=args.data_path) print("TRAIN") print(str(train).encode('utf-8')) print(len(train)) print("TEST") print(str(test).encode('utf-8')) print(len(test)) print("validation") print(str(validation).encode('utf-8')) print(len(validation)) #pickle.dump(test, open(filepath + '.test_data.pickle', 'wb')) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) entity2same_validation = generate_names(validation, people) print(str(entity2same_train).encode('utf-8')) print(str(entity2same_test).encode('utf-8')) # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) unique_text_validation, entity2unique_validation = build_unique_entities(entity2same_validation) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) if not USE_PRECOMPUTED_SPLIT: # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") tokenizer.fit_on_texts(unique_text + unique_text_test + unique_text_validation) pickle.dump( tokenizer, open(args.data_path + '.tokenizer' , "wb" )) else: tokenizer = pickle.load(open(args.data_path + '.tokenizer', 'rb')) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) sequences_validation = tokenizer.texts_to_sequences(unique_text_validation) sequences_validation = pad_sequences(sequences_validation, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: triplets = generate_triplets_from_ANN(embedder_model, sequences, entity2unique, entity2same_train, unique_text, True) print(len(triplets['anchor'])) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) validation_data, val_match_stats = generate_triplets_from_ANN(embedder_model, sequences_validation, entity2unique_validation, entity2same_validation, unique_text_validation, False) validation_seq = get_sequences(validation_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, train_match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T Y_val = np.random.randint(2, size=(1,2,len(validation_data['anchor']))).T checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=2, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_data=([validation_seq['anchor'], validation_seq['positive'], validation_seq['negative']],Y_val)) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
25,195
37.118003
249
py
fuzzyJoiner
fuzzyJoiner-master/preloaded_runner.py
import pickle import numpy as np import tensorflow as tf import random as random import json from keras import backend as K #from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Input, Lambda, GRU from keras.layers import Embedding from keras.models import Model from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from names_cleanser import NameDataCleanser, CompanyDataCleanser import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False filepath="weights.best.hdf5" def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print(str("failed to find embedding for:" + word).encode('utf-8')) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath, 'r', encoding='utf8') as fl: for line in fl: entities.append(line) return entities """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = t.get_distance(index, entity2unique[i]) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = t.get_distance(index, entity2unique[j]) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = t.get_distance(index, entity2unique[p]) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 closest_positive_counts.append(closest_pos_count / min(len(expected_text), NNlen - 1)) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) print("Precision at 1: " + str(precise / len(entity2same))) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w', encoding='utf8') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def build_model_from_weights(weights_file, embbeding_dimensions): embedder = Embedding(embbeding_dimensions, EMBEDDING_DIM, input_length=MAX_SEQUENCE_LENGTH, trainable=False) main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, 2): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') try: model.load_weights(weights_file) except ValueError as e: full = str(e) #https://stackoverflow.com/questions/4289331/python-extract-numbers-from-a-string new_emb = [int(num) for num in full.replace('.', ' ').split() if num.isdigit()] return build_model_from_weights(weights_file, new_emb[2]) model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') try: model.load_weights(weights_file) except ValueError as e: full = str(e) #https://stackoverflow.com/questions/4289331/python-extract-numbers-from-a-string new_emb = [int(num) for num in full.replace('.', ' ').split() if num.isdigit()] return build_model_from_weights(weights_file, new_emb[2]) model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model # First get the arguments output_file_name_for_hpo = "val_dict_list.json" parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--loss_function', type=str, help='triplet loss function type: triplet-loss, improved-loss, angular-loss, adapted-loss') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') parser.add_argument('--model', type=str, help='name for model file') parser.add_argument('--tokenizer', type=str, help='location of tokenizer file') parser.add_argument('--previous_test', type=str, help='use previous test data') args = parser.parse_args() filepath = args.model LOSS_FUNCTION = None if args.loss_function == 'triplet-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'adapted-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) people = 'people' in args.entity_type # Load test data if specified if args.previous_test and args.previous_test.lower() in ("yes", "true", "t", "1"): test = pickle.load(open(args.input, 'rb')) else: test = read_entities(args.input) print("TEST") print(str(test).encode('utf-8')) #encode test data for annoy entity2same_test = generate_names(test, people, limit_pairs=True) print(str(entity2same_test).encode('utf-8')) #load tokenizer tokenizer = pickle.load(open(args.tokenizer, 'rb')) #Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("test text len:" + str(len(unique_text_test))) #use tokenizer to convert to sequences sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build model so we can load weights into it word_index = tokenizer.word_index num_words = len(word_index) + 1 model, test_positive_model, test_negative_model, inter_model = build_model_from_weights(args.model, num_words) current_model = inter_model # print some statistics test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
17,465
35.848101
139
py
fuzzyJoiner
fuzzyJoiner-master/parse_additive.py
0
0
0
py
fuzzyJoiner
fuzzyJoiner-master/Levenstien_Rule_Based.py
from sys import argv import string import Levenshtein import statistics from names_cleanser import NameDataCleanser, CompanyDataCleanser def read_entities(filepath): entities = [] with open(filepath, 'r', encoding='utf8') as fl: for line in fl: entities.append(line) return entities def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def load_buckets(entity2same): #used https://stackoverflow.com/questions/34860982/replace-the-punctuation-with-whitespace translator = str.maketrans(string.punctuation, ' '*len(string.punctuation)) bucket_dict = {} for item in entity2same: for entity in entity2same[item] + [item]: entity_no_punct = entity.translate(translator).lower() split_entity = [item for item in entity_no_punct.split(' ') if item] for word in split_entity: if word not in bucket_dict: bucket_dict[word] = [] bucket_dict[word].append(entity) return bucket_dict def get_stats(entity2same, bucket_dict): def get_closest(items, key, number_to_get): low_key = key.lower() closest = [(x, Levenshtein.distance(low_key, x.lower())) for x in items] closest = list(set(closest)) # print('key') # # print(key.encode('utf-8')) # print('unsorted') # # print([(thing[0].encode('utf-8'), thing[1]) for thing in closest]) # print('sorted') closest = sorted(closest, key=lambda a: a[1]) # print([(thing[0].encode('utf-8'), thing[1]) for thing in closest]) # print('dist removed') closest = [item[0] for item in closest] # print([thing.encode('utf-8') for thing in closest]) # print([thing.encode('utf-8') for thing in closest[:number_to_get]]) return closest[:number_to_get] # predictions = model.predict(sequences) # t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed # t.set_seed(123) # for i in range(len(predictions)): # # print(predictions[i]) # v = predictions[i] # t.add_item(i, v) # t.build(100) # 100 trees match = 0 no_match = 0 lev_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] NNlen = 20 translator = str.maketrans(string.punctuation, ' '*len(string.punctuation)) print_num = 0 skipped = 0 for key in entity2same: # if not key.translate(translator).strip(): # skipped += 1 # continue nearest = [] for word in key.translate(translator).lower().split(): if not word: continue nearest = nearest + bucket_dict[word] #if key == "Abu al - Qasim Muhammad ibn 'Abd Allah ibn 'Abd al - Muttalib ibn Hashim": #print(nearest) if len(nearest) > NNlen: nearest = get_closest(nearest, key, NNlen) #if key == "Abu al - Qasim Muhammad ibn 'Abd Allah ibn 'Abd al - Muttalib ibn Hashim": #print(nearest) nearest_text = set(nearest) expected_text = set(entity2same[key]) if print_num % 1000 == 0: print(print_num) print(key.encode('utf-8')) print([item.encode('utf-8') for item in nearest]) print_num += 1 # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) if len(nearest_text) == 0: skipped += 1 continue # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), len(nearest_text)) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = Levenshtein.distance(key.lower(), j.lower()) pos_distances.append(dist_pos) dist_neg = Levenshtein.distance(key.lower(), i.lower()) neg_distances.append(dist_neg) if dist_pos < dist_neg: lev_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = Levenshtein.distance(key.lower(), i.lower()) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = Levenshtein.distance(key.lower(), j.lower()) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = Levenshtein.distance(key.lower(), p.lower()) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 #print(len(expected_text)) #print(len(nearest_text)) if min(len(expected_text),len(nearest_text)) == 0: print(nearest) print(expected_text) print(nearest_text) print('key' + str(key)) for word in key.translate(translator).lower().split(): if not word: continue print(bucket_dict[word]) print(word) closest_positive_counts.append(closest_pos_count / min(len(expected_text),len(nearest_text))) # for i in negatives: # for j in expected_text: # triplets['anchor'].append(key) # triplets['positive'].append(j) # triplets['negative'].append(i) print('bad entities' + str(skipped)) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(lev_accuracy / total)) print("Precision at 1: " + str(precise / (len(entity2same) - skipped))) # obj = {} # obj['accuracy'] = lev_accuracy / total # obj['steps'] = 1 # with open(output_file_name_for_hpo, 'w', encoding='utf8') as out: # json.dump(obj, out) # if test: return match/(match + no_match) # else: # return triplets, match/(match + no_match) input_file = argv[1] people = True if argv[2][0] is 'p' else False print(argv) print(people) entities = read_entities(input_file) entity2same = generate_names(entities, people) bucket_dict = load_buckets(entity2same) # for key in bucket_dict: # print('key {} value {}'.format(key, bucket_dict[key]).encode('utf-8')) print(get_stats(entity2same, bucket_dict))
8,910
37.409483
120
py
fuzzyJoiner
fuzzyJoiner-master/names_cleanser.py
import re import argparse import Levenshtein from nltk import bigrams from os import listdir from os.path import isfile, join from difflib import SequenceMatcher # from sklearn.metrics import jaccard_similarity_score class GenericDataCleanser(object): name_reject_set = frozenset(['father of', '(', 'author of', 'pope', 'emperor']) company_reject_set = frozenset(["("]) def __init__(self, entity_type, function = None, number=None, pairs=2): if number: self.get=True self.number_of_names = number else: self.get = False function_dictionary = {"nametest":self.test_x_names, "companytest":self.test_x_companies, "name":self.good_name_data, "company":self.good_company_data, "names":self.good_name_data} if entity_type == "name" or entity_type == "names" or entity_type == "company": self.parsing_function = function_dictionary[entity_type] elif function and function == "test": self.parsing_function = function_dictionary[entity_type + "test"] self.pairs = 2 def test_x_names(self, data): self.number_of_names -= 1 if self.number_of_names < 0: return False else: return self.good_name_data(data) def test_x_companies(self, data): self.number_of_names -= 1 if self.number_of_names < 0: return False else: return self.good_company_data(data) def is_english(self, data): try: data.encode('ascii') except UnicodeEncodeError: return False return True def fix_bad_chars(self, data): #used https://www.tutorialspoint.com/python/string_translate.htm #change if we want to preserve any of these things intab = "\n" outtab = " " trantab = str.maketrans(intab, outtab) return data.translate(trantab) #test whether or not a company name should be used def remove_bad(self, data): data = data.replace('"', "(") data = data.replace("\n", "") data = data.replace(" ", "") return data.lstrip() def good_company_data(self, data): if data.startswith("<http://dbpedia.org/resource"): return False if data.startswith("The Master Trust Bank of Japan"): return False if not self.is_english(data): return False data = self.fix_bad_chars(data) data = data.lower() for item in self.company_reject_set: if item in data: return False return True def good_name_data(self, data): if not self.is_english(data): return False data = self.fix_bad_chars(data) data = data.lower() for item in self.name_reject_set: if item in data: return False return True #cleanses the data and returns only good data #if we cannot get 2 peices of good data returns the empty string def cleanse_data(self, dataToCleanse): ret = [] name_array = dataToCleanse.split("|") for part in name_array: if not part: continue if len(ret) >= self.pairs: break if self.parsing_function(part): # if part[-1:] == "\n": # part = part[:-1] if part not in ret: part = self.remove_bad(part) if part: part = part.replace(',', ' , ').replace('-', ' - ').replace('.', ' . ').replace(' ', ' ') ret.append(part) if len(ret) != self.pairs: return None return ret #parses input file and writes an output def parse_file(self,input_file, output_file, output_rejects_file): lines = input_file.readlines() entity_id = 0 print('in parse file') count = 0 for line in lines: ret = self.cleanse_data(line) if ret and len(ret) >= self.pairs: count += 1 # all these entity names are in the exact same set # we can create the number of pairs that we were asked to create # for now, create it with the 'anchor' (first element of the list) # and all other names entity_id += 1 newline = '|'.join(ret) + '\n' output_file.write(newline) else: output_rejects_file.write(line) if self.get: print("ran out of names, proceeding with as many as were available") print('total number of names:' + str(count)) def clean_file(self, filename, output): onlyfiles = [f for f in listdir(filename) if isfile(join(filename, f))] output_file = open(output, "w", encoding='utf-8') output_rejects_file = open('./rejects.txt', 'w', encoding='utf-8') output_file = open(output, "w", encoding='utf-8') for file_path in onlyfiles: input_file = open(filename + "/" + file_path, encoding='utf-8') self.parse_file(input_file, output_file, output_rejects_file) output_rejects_file.close() input_file.close() output_file.close() class CompanyDataCleanser(GenericDataCleanser): def __init__(self, limit_pairs=False): self.limit_pairs = False self.pairs = 2 self.get = False def is_acronym(self, name): p = re.compile('[A-Z0-9]+') if len(name) < 6 and p.match(name): return True return False def has_cyrillic(self, text): return bool(re.search('[а-яА-Я]', text)) def cleanse_data(self, line): if self.has_cyrillic(line): return None # Wikipedia has URLs for the company name first remove trailing '>' and remove disambiguation of type by (company) line = line.replace('(company)', '') line = line.replace('\n', '') arr = line.split('|') url = arr[0].split('/') company_name = url[-1].replace('>', '') # SEC filings have some odd company names, filter these out if re.match('T[0-9]+', company_name): return None if re.match('[0-9 ]+', company_name): return None company_name = company_name.replace('_', ' ') company_name_no_spaces = re.sub('[ .,]', '', company_name).lower().strip() ret_val = [] ret_val.append(company_name) for i in range(1, len(arr)): arr[i] = arr[i].replace('\n', '') name = arr[i].replace(' ', '') name = name.lower().strip() name = re.sub('[,.]', '', name) if name == company_name: continue # check to see if this is an acronym if self.is_acronym(arr[i]): ret_val.append(arr[i]) else: # find longest substring seqMatch = SequenceMatcher(None, company_name_no_spaces, name) # find match of longest sub-string match = seqMatch.find_longest_match(0, len(company_name_no_spaces), 0, len(name)) # ensure that we match the company name somewhere early in the string if match.size > 1 and match.a < 1: ret_val.append(arr[i]) else: print('removing name:' + arr[i] + ' for ' + company_name) return ret_val class NameDataCleanser(GenericDataCleanser): MAX_PAIRS = 6 def __init__(self, number=0, pairs = 2, limit_pairs = True): if number > 0: self.get=True self.number_of_names = number else: self.get = False self.pairs = pairs self.limit_pairs = limit_pairs def create_new_name(self, name, current_name_set): names = name.split() ret = set([]) if len(names) <= 2: ret.add(names[1] + ' , ' + names[0]) # last_name, first_name ret.add(names[0][0] + ' . ' + names[1]) # first_initial. lastname ret.add(names[1] + ' , ' + names[0][0]) # last_name, first_initial. elif len(names) > 2: ret.add(names[0] + ' ' + names[1][0] + ' . ' + names[-1]) # first_name middle initial last_name ret.add(names[-1] + ' , ' + names[0] + ' ' + names[1]) # last_name, first name, middle name ret.add(names[-1] + ' , ' + names[0] + ' ' + names[1][0] + ' . ') # last_name, first name, middle initial ret.add(names[-1] + " " + names[0]) # drop all middle names ret.add(names[0][0] + ' . ' + names[-1]) # first initial. lastname ret.add(names[-1] + ' , ' + names[0][0]) # last_name, first_initial. diff = ret.difference(set(current_name_set)) return diff def cleanse_data(self, line): # skip lines with pope/emperor/queen etc they tend to be useless titles = ['king','queen','emperor'] qualifiers = ['of', 'I','II','III','IV','V','VI','VII','VIII','IX','X','XI','XII','XIII','XIV','XV','XVI','XVII','XVIII'] if 'pope' in line.lower(): return for p in titles: if p in line.lower(): for q in qualifiers: if q in line: return # remove any names with "..." with nothing. also remove all (.*) line = re.sub('["][^"]*["]', '', line) line = re.sub('[(][^)]*[)]', '', line) # remove any non-Latin characters from the line line = re.sub('[^\x00-\x7F\x80-\xFF\u0100-\u017F\u0180-\u024F\u1E00-\u1EFF]','',line) arr = line.split("|") dups = [] for name in arr: dups.append(name.replace(',', ' , ').replace('-', ' - ').replace('.', ' . ').replace(' ', ' ').strip()) if (len(set(dups))) == 1: return # remove silly names cleansed_arr = [] for name in arr: if name.strip() == '' or name.strip() == '.': continue # remove all abbreviated names that end with . if name.endswith('.'): continue # remove all single names if len(name.split(' ')) == 1: continue name = name.replace(',', ' , ').replace('-', ' - ').replace('.', ' . ').replace(' ', ' ').strip() cleansed_arr.append(name.replace('\n', '').replace('"','')) if len(cleansed_arr) == 0 or len(cleansed_arr[0].strip().split(' ')) == 1: return # check if the name is simply some name with all other words is just a title base_parts_set = set(cleansed_arr[0].split()) if len(base_parts_set.intersection(set(qualifiers))) == 1: return # compare each name with every other name in the array to make sure # we have at least one name part in common with the first name, which we assume is # the 'canonical' name. ret = set(cleansed_arr) base = cleansed_arr[0].lower() # print(cleansed_arr[0]) for j in range(1, len(cleansed_arr)): cmp = cleansed_arr[j].lower() # same name just changed case if base == cmp and cleansed_arr[j] in ret: ret.remove(cleansed_arr[j]) continue ratio = Levenshtein.ratio(base, cmp) if (ratio == 0 and cleansed_arr[j] in ret): ret.remove(cleansed_arr[j]) # print('removing due to ratio' + cleansed_arr[j] + "|" + cleansed_arr[0]) continue # string similarity metrics will tell you Richard Phillips is similar to Jack Dowling because they look at it character by character # need something at a word level. Look at all the name parts in the thing we are comparing to base, and make sure at least # one name overlaps match = False for part in cleansed_arr[j].lower().split(' '): part = part.replace('\n','') if part in base: match = True if not match and cleansed_arr[j] in ret: ret.remove(cleansed_arr[j]) # print('removing due to no overlap in names' + cleansed_arr[j] + "|" + cleansed_arr[0]) continue # often we have situations where the first name matches but everything else does not. This happens # when a woman changes names for instance. A complication here is that Japanese names reverse names # so we need to not remove that case match = False s = set(cleansed_arr[0].lower().split()).intersection(set(cleansed_arr[j].lower().split())) if len(s) > 1: match = True rest_of_base = ''.join(cleansed_arr[0].lower().split()[1:]) compare_to = ''.join(cleansed_arr[j].lower().split()[1:]) base_bigrams = set(bigrams(rest_of_base)) cmp_bigrams = set(bigrams(compare_to)) if len(base_bigrams.intersection(cmp_bigrams)) == 0 and not match: # print(base_bigrams) # print(cmp_bigrams) if cleansed_arr[j] in ret: # print('removing due to looks like a female name changed after marriage' + cleansed_arr[j] + base) ret.remove(cleansed_arr[j]) continue anchor = cleansed_arr[0] ret_val = [cleansed_arr[0]] # put anchor in first, then add all other elements for n in ret: if n != cleansed_arr[0]: ret_val.append(n) if len(ret_val) < self.pairs: additional_elements = self.create_new_name(anchor, ret_val) ret_val.extend(additional_elements) if len(ret_val) < self.pairs: return elif len(ret_val) > self.MAX_PAIRS: anchor_parts = set(anchor.split()) def jaccard_sim_function(x): x_parts = set(x.split()) return len(anchor_parts.intersection(x_parts)) / len(anchor_parts.union(x_parts)) sorted(ret_val, key=jaccard_sim_function) # print(ret_val) ret_val = ret_val[0:self.MAX_PAIRS] assert len(set(ret_val)) >= self.pairs return ret_val if __name__ == '__main__': parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) parser.add_argument('-f', dest="input_file", help="file to cleanse file") parser.add_argument('-o', dest="output_file", help="cleansed data file name") parser.add_argument('-t', dest="entity_type", help="names or companies") parser.add_argument('-p', dest="num_pairs", help="how many same pairs should be created (default=2)", nargs='?', default=2, type=int) parser.add_argument('-u', dest="function", help="use test parsing function", nargs='?', default=None) parser.add_argument('-n', dest="number", help="process only this many from the file (useful for debugging)", nargs='?', type=int, default = 0) parser.add_argument('-l', dest="no_limit", nargs='?', type=bool, default=False) args = parser.parse_args() assert args.num_pairs <= 4 print(args.num_pairs) if args.entity_type == 'names': cleaner = NameDataCleanser(args.number, args.num_pairs) elif args.entity_type == 'companies': cleaner = CompanyDataCleanser() # Uncomment next line to test old code # cleaner = GenericDataCleanser(args.entity_type, args.function, args.number, args.num_pairs) cleaner.clean_file(args.input_file, args.output_file)
15,807
37.462287
188
py
fuzzyJoiner
fuzzyJoiner-master/random_test_selecter.py
from sys import argv from random import shuffle input_file = open(argv[1], 'r') output_file = open(argv[2], 'w') lines = input_file.readlines() shuffle(lines) lines = lines[-int(argv[3]):] for line in lines: output_file.write(line) input_file.close() output_file.close()
272
21.75
32
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM.py
import numpy as np import random as random # """ # The below is necessary in Python 3.2.3 onwards to # have reproducible behavior for certain hash-based operations. # See these references for further details: # https://docs.python.org/3.4/using/cmdline.html#envvar-PYTHONHASHSEED # https://github.com/keras-team/keras/issues/2280#issuecomment-306959926 import os os.environ['PYTHONHASHSEED'] = '0' # The below is necessary for starting Numpy generated random numbers # in a well-defined initial state. np.random.seed(42) # The below is necessary for starting core Python generated random numbers # in a well-defined state. # random.seed(12345) # Force TensorFlow to use single thread. # Multiple threads are a potential source of # non-reproducible results. # For further details, see: https://stackoverflow.com/questions/42022950/which-seeds-have-to-be-set-where-to-realize-100-reproducibility-of-training-res # session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1) import tensorflow as tf # from keras import backend as K # The below tf.set_random_seed() will make random number generation # in the TensorFlow backend have a well-defined initial state. # For further details, see: https://www.tensorflow.org/api_docs/python/tf/set_random_seed tf.set_random_seed(1234) sess = tf.Session(graph=tf.get_default_graph()) # """ from keras import backend as K K.set_session(sess) from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Lambda, GRU, Activation from keras.layers import Embedding from keras.models import Model, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=30 USE_GRU=True DEBUG = True DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): # margin = K.constant(MARGIN) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0]) + K.square(margin - y_pred[:,2,0])) # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(MARGIN) lambda_p = 0.02 threshold = 0.1 a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / 2) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def tanhNorm(x): square_sum = K.sum(K.square(x), axis=-1, keepdims=True) dist = K.sqrt(K.maximum(square_sum, K.epsilon())) tanh = K.tanh(dist) scale = tanh / dist return x * scale def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = expected_text negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(accuracy / total)) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, limit_pairs=False): num_names = 4 names_generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) entity2same = {} for entity in entities: ret = names_generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def get_hidden_layer(name, net, is_last): if USE_GRU: if is_last: return GRU(128, activation='relu', name=name)(net) else: return GRU(128, return_sequences=True, activation='relu', name=name)(net) else: return Dense(128, activation='relu', name=name)(net) def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) # net = GRU(128, return_sequences=True, activation='relu', name='embed')(net) #net = GRU(128, return_sequences=True, activation='relu', name='embed2')(net) #net = GRU(128, return_sequences=True, activation='relu', name='embed2a')(net) net = GRU(128, activation='relu', name='embed3')(net) """ for i in range(0, NUM_LAYERS): net = get_hidden_layer('embed' + str(i), net, False) net = get_hidden_layer('embed_last', net, True) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) """ base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=triplet_loss, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='loss function type: triplet-loss, improved-triplet-loss, modified-loss, or angular-loss') parser.add_argument('--use_l2_norm', type=bool, help='whether to add a l2 norm') parser.add_argument('--test_neighbor_len', type=int, help='size of the neighborhood for testing') parser.add_argument('--train_neighbor_len', type=int, help='size of the neighborhood for training') parser.add_argument('--embedding_type', type=str, help='encoding type to use for input: Kazuma (for Kazuma character embedding) or one-hot') parser.add_argument('--use_GRU', type=bool, help='use GRU or default to MLP') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') args = parser.parse_args() """ LOSS_FUNCTION = None if args.loss_function == 'triplet-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved_triplet_loss': LOSS_FUNCTION=improved_triplet_loss elif args.loss_function == 'modified_loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = true LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) MARGIN = args.margin print('Margin:' + str(MARGIN)) TRAIN_NEIGHBOR_LEN = args.train_neighbor_len TEST_NEIGHBOR_LEN = args.test_neighbor_len print('Train neighbor length: ' + str(TRAIN_NEIGHBOR_LEN)) print('Test neighbor length: ' + str(TEST_NEIGHBOR_LEN)) USE_L2_NORM = args.use_l2_norm print('Use L2Norm: ' + str(USE_L2_NORM)) EMBEDDING_TYPE = args.embedding_type print('Embedding type: ' + EMBEDDING_TYPE) USE_GRU = args.use_GRU print('Use GRU: ' + str(args.use_GRU)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) """ # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train) entity2same_test = generate_names(test, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) match_stats = 0 # num_iter = 100 num_iter = 1 counter = 0 current_model = embedder_model prev_match_stats = 0 while test_match_stats < .9 and counter < num_iter: counter += 1 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats)) K.clear_session()
21,712
37.227113
167
py
fuzzyJoiner
fuzzyJoiner-master/old/matcher_functions.py
import sqlalchemy from sqlalchemy.sql import select from sqlalchemy import Table, Column, Integer, String, MetaData, ForeignKey def connect(user, password, db, host='localhost', port=5432): '''Returns a connection and a metadata object''' # We connect with the help of the PostgreSQL URL # postgresql://federer:grandestslam@localhost:5432/tennis url = 'postgresql://{}:{}@{}:{}/{}' url = url.format(user, password, host, port, db) # The return value of create_engine() is our connection object con = sqlalchemy.create_engine(url, client_encoding='utf8') # We then bind the connection to MetaData() meta = sqlalchemy.MetaData(bind=con, reflect=True) return con, meta # create a set of tuples for the machine learning model where each entity name is matched # with at least 2 other names of the same entity, and each entity is paired with 2 other # names of a different entity. The different pairs are chosen based on the fact that the have the # same name part, or are completely different. If the name is unique then the 2 different pairs # are just 2 random pairs. def create_tuples_for_siamese_network(): aliases_m = meta.tables['aliases'] word2entities_m= meta.tables['words2entities'] #this should combine several functions and return alisases and bucket lists. def condensed_start(): pass def load_good_buckets(table_string1, table_string2, dictionary, con, meta): table = meta.tables[table_string1] words1 = [list(row) for row in con.execute(select([table]))] table = meta.tables[table_string2] words2 = [list(row) for row in con.execute(select([table]))] buckets = [] bucket_words = {} temp_list1 = [] temp_list2 = [] x = 0 y = 0 while x != None and y != None: if words1[x][1] < words2[y][1]: x = find_next_bucket(words1, x) elif words1[x][1] > words2[y][1]: y = find_next_bucket(words2, y) else: word = words1[x][1] temp_list1, x = load_bucket(words1, x, dictionary) temp_list2, y = load_bucket(words2, y, dictionary) pair = [temp_list1, temp_list2] buckets.append(pair) bucket_words[word] = buckets[-1] return buckets, bucket_words def find_next_bucket(table, position): prev_word = table[position][1] while position < len(table): if table[position][1] != prev_word: return position position += 1 return None def load_bucket(table, position, dictionary): bucket = [] prev_word = table[position][1] while position < len(table): if table[position][1] != prev_word: return (bucket, position) bucket.append([dictionary[table[position][0]], table[position][0]]) position += 1 return bucket, None def create_double_num_dicts(aliases): serial_num = 0 num_to_word = {} word_to_num = {} for pair in aliases: for name in pair: num_to_word[serial_num] = name word_to_num[name] = serial_num serial_num += 1 return num_to_word, word_to_num def fscore(true_items, test_dict, beta): true_positives = 0.0 false_positive = 0.0 false_negative = 0.0 not_indexed = 0 for key in test_dict: if test_key(true_items, test_dict, key): true_positives += 1 else: false_positive += 1 for pair in true_items: if pair[0] not in test_dict: false_negative += 1 not_indexed += 1 elif pair[1] not in test_dict[pair[0]]: false_negative += 1 print( "total names: " + str(len(true_items))) print( "not indexed: " + str(not_indexed)) print ("wrongly indexed: " + str(false_negative - not_indexed)) print ("all false negitives: " + str(false_negative)) print ("true positives: " + str(true_positives)) print ("false positives: " + str(false_positive)) temp = ((1 + (beta * beta)) * true_positives) return temp / (temp + false_positive + ((beta * beta) * false_negative)) def test_key(true_items, test_dict, key): for answer in test_dict[key]: if (key, answer) in true_items: return True return False def make_test_dict(items, k): dictionary = {} overflow = 0 for pair in items: if pair[0] in dictionary: if len(dictionary[pair[0]]) < k: dictionary[pair[0]].append(pair[1]) else: overflow += 1 else: dictionary[pair[0]] = [pair[1]] print( "overflow: " + str(overflow)) return dictionary def get_aliases(con, meta): table = meta.tables['aliases'] aliases = con.execute(select([table])) dictionary = set([]) for row in aliases: dictionary.add((row[0], row[1])) return dictionary def run_test(pre_procces, test, num_to_word, bucket_list): bucket_list = pre_proccess_words(num_to_word, bucket_list, pre_procces) matches = set([]) for pair in bucket_list: for name1 in pair[0]: for name2 in pair[1]: if test(name1[1], name2[1]): matches.add((num_to_word[name1[0]], num_to_word[name2[0]])) return matches def create_alias_dict(con, meta): table = meta.tables['aliases'] aliases = con.execute(select([table])) dictionary = {} for row in aliases: dictionary[row[0]] = row[1] return dictionary def get_missed(aliases, test_dict): missed = set([]) for pair in aliases: if pair[0] not in test_dict or pair[1] not in test_dict[pair[0]]: missed.add(pair) return missed def export_unbucketed(impossible, con, meta): execute_pairs = [] if 'unbucketed' in meta.tables: meta.tables['unbucketed'].drop(con) unbucketed = Table('unbucketed', meta, Column('name1', String), Column('name2', String), extend_existing=True) zipping_string = ('name1', 'name2') for key in impossible: for name in impossible[key]: execute_pairs.append(dict(zip(zipping_string, (key, name)))) meta.create_all(con) con.execute(unbucketed.insert(), execute_pairs) def export_missed(aliases, test_dict, con, meta): missed_items = get_missed(aliases, test_dict) execute_pairs = [] if 'missed' in meta.tables: meta.tables['missed'].drop(con) missed = Table('missed', meta, Column('name1', String), Column('name2', String), extend_existing=True) zipping_string = ('name1', 'name2') for pair in missed_items: execute_pairs.append(dict(zip(zipping_string, pair))) meta.create_all(con) con.execute(missed.insert(), execute_pairs) def get_impossible(aliases, bucket_list, num_to_word): testing_dict = {} for pair in aliases: if pair[0] in testing_dict: testing_dict[pair[0]].append(pair[1]) else: testing_dict[pair[0]] = [pair[1]] bucket_list = pre_proccess_words(num_to_word, bucket_list, lambda x : x) for bucket_pair in bucket_list: other_bucket = set([bucket_pair[1][x][1] for x in range(len(bucket_pair[1]))]) for name in bucket_pair[0]: other_name = 0 while other_name < len(testing_dict[name[1]]): try: if testing_dict[name[1]][other_name] in other_bucket: testing_dict[name[1]].pop(other_name) except ValueError: pass other_name += 1 for key in testing_dict.copy(): if not testing_dict[key]: del testing_dict[key] return testing_dict def pre_proccess_words(num_to_word, bucket_list, function): for pair in bucket_list: for bucket in pair: for name in bucket: name[1] = function(num_to_word[name[0]]) return bucket_list def run_special_test(bucket_list, num_to_word): matches = set([]) for pair in bucket_list: if len(pair[0]) == 1 and len(pair[1]) == 1: matches.add((num_to_word[pair[0][0][0]], num_to_word[pair[1][0][0]])) return matches
8,149
37.809524
114
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-8.20.18.py
import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics from scipy.stats.mstats import gmean import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False filepath="weights.best.hdf5" output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = t.get_distance(index, entity2unique[i]) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = t.get_distance(index, entity2unique[j]) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = t.get_distance(index, entity2unique[p]) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 closest_positive_counts.append(closest_pos_count / min(len(expected_text), NNlen - 1)) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) print("Precision at 1: " + str(precise / len(entity2same))) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') parser.add_argument('--model', type=str, help='name for model file') args = parser.parse_args() filepath = args.model LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
21,235
36.061082
163
py
fuzzyJoiner
fuzzyJoiner-master/old/ANNBasedSampleSelection.py
import Named_Entity_Recognition_Modified from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from matcher_functions import connect import argparse import numpy as np from keras.layers import Embedding, Concatenate from keras.models import Model import names_cleanser from random import randint import keras.backend as K import sys MARGIN = 2 DEBUG = False def process_aliases(con, meta): aliases = Named_Entity_Recognition_Modified.get_aliases_with_ids(con, meta) entity2sames = {} namesToIds = {} def has_difft_id(name, entityid): if name in namesToIds: ids = namesToIds[name] if ids != entityid: return True namesToIds[name] = entityid return False i = 0 for row in aliases: i += 1 if DEBUG and i > 100: break entityid = row[2] # filter out names that are associated with multiple ids, this will confuse the model trying to learn the distance function if has_difft_id(row[0], entityid) or has_difft_id(row[1], entityid) or has_difft_id(row[2], entityid): continue if entityid not in entity2sames: entity2sames[entityid] = [row[0]] entity2sames[entityid].append(row[1]) else: entity2sames[entityid].append(row[1]) # print(entity2sames) entities = [] entity2names = {} for e,v in entity2sames.items(): entity2names[e] = [len(entities) + i for i in range(0, len(v))] entities.extend(v) print(entities) print(entity2names) return entities, entity2names if __name__ == '__main__': print('Processing text dataset') parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) parser.add_argument('-u', dest="user", help="username") parser.add_argument('-p', dest="password", help="password") parser.add_argument('-d', dest="db", help="dbname") parser.add_argument('-o', dest="output_file", help="output file name") parser.add_argument('-a', dest="num_pairs", help="number of same pairs in db", nargs='?', default=2, type=int) args = parser.parse_args() #change to get from sql and not read from file con, meta = connect(args.user, args.password, args.db) # get all names first entities, entity2names = process_aliases(con, meta) tokenizer = Tokenizer(num_words=Named_Entity_Recognition_Modified.MAX_NB_WORDS) tokenizer.fit_on_texts(entities) sequences = tokenizer.texts_to_sequences(entities) print(sequences) sequences = pad_sequences(sequences, maxlen=Named_Entity_Recognition_Modified.MAX_SEQUENCE_LENGTH) word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, Named_Entity_Recognition_Modified.EMBEDDING_DIM)) kz = KazumaCharEmbedding() for word, i in word_index.items(): if i >= Named_Entity_Recognition_Modified.MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector # note that we set trainable = False so as to keep the embeddings fixed Named_Entity_Recognition_Modified.check_for_zeroes(embedding_matrix, "here is the first pass") embedding_layer = Embedding(num_words, Named_Entity_Recognition_Modified.EMBEDDING_DIM, weights=[embedding_matrix], input_length=Named_Entity_Recognition_Modified.MAX_SEQUENCE_LENGTH, trainable=False) model = Named_Entity_Recognition_Modified.embedded_representation(embedding_layer) embedded_output = model.predict(sequences) print(np.shape(embedded_output)) sys.exit(0) t = AnnoyIndex(len(embedded_output[0]), metric='euclidean') for i in range(len(embedded_output)): v = embedded_output[i] t.add_item(i, v) t.build(100) # 100 trees with open(args.output_file, 'w') as f: for e, v in entity2names.items(): index_for_same = entity2names[e] anchor_index = index_for_same[0] nearest = t.get_nns_by_vector(embedded_output[anchor_index], 10) maximum_diff = -1 minimum_same = 100000 maximum_same = -1 same_pair_in_NN_set = False for i in range(1, len(index_for_same)): dist = t.get_distance(anchor_index, index_for_same[i]) print("same pair:" + entities[anchor_index] + "-" + entities[index_for_same[i]] + " distance:" + str(dist)) minimum_same = min(dist, minimum_same) maximum_same = max(dist, maximum_same) for i in nearest: if i == anchor_index: continue dist = t.get_distance(anchor_index, i) print(entities[anchor_index] + "-" + entities[i] + " distance:" + str(dist)) if i in index_for_same: same_pair_in_NN_set = True else: maximum_diff = max(dist, maximum_diff) if dist > minimum_same: f.write(entities[anchor_index] + "|" + entities[index_for_same[randint(1, len(index_for_same) - 1)]] + "|" + entities[i] + "\n") if (maximum_diff < minimum_same): print("hard entity because maximum different is less than minimum same") continue elif same_pair_in_NN_set: print("easy entity - same pair is in NN set") else: print("~hard entity" + entities[anchor_index]) # write a set of different completely items now print(maximum_same) j = 0 while j <= 30: k = randint(0, len(entities) - 1) if t.get_distance(anchor_index, k) > maximum_diff + MARGIN: f.write((entities[anchor_index] + "|" + entities[index_for_same[randint(1, len(index_for_same) - 1)]] + "|" + entities[k] + "\n")) k += 1 j += 1 print(len(entity2names))
6,038
32.181319
138
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-angular.py
import numpy as np import tensorflow as tf import random as random # import cntk as C # """ # The below is necessary in Python 3.2.3 onwards to # have reproducible behavior for certain hash-based operations. # See these references for further details: # https://docs.python.org/3.4/using/cmdline.html#envvar-PYTHONHASHSEED # https://github.com/keras-team/keras/issues/2280#issuecomment-306959926 import os os.environ['PYTHONHASHSEED'] = '0' # The below is necessary for starting Numpy generated random numbers # in a well-defined initial state. np.random.seed(42) # The below is necessary for starting core Python generated random numbers # in a well-defined state. random.seed(12345) # Force TensorFlow to use single thread. # Multiple threads are a potential source of # non-reproducible results. # For further details, see: https://stackoverflow.com/questions/42022950/which-seeds-have-to-be-set-where-to-realize-100-reproducibility-of-training-res session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1) # import theano as T from keras import backend as K # The below tf.set_random_seed() will make random number generation # in the TensorFlow backend have a well-defined initial state. # For further details, see: https://www.tensorflow.org/api_docs/python/tf/set_random_seed tf.set_random_seed(1234) sess = tf.Session(graph=tf.get_default_graph(), config=session_conf) K.set_session(sess) # """ # from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=1 ALPHA=30 USE_GRU=True DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=True LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = True def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(T.tensor.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): # margin = K.constant(MARGIN) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0]) + K.square(margin - y_pred[:,2,0])) # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(MARGIN) lambda_p = 0.02 threshold = 0.1 a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / 2) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def tanhNorm(x): square_sum = K.sum(K.square(x), axis=-1, keepdims=True) dist = K.sqrt(K.maximum(square_sum, K.epsilon())) tanh = K.tanh(dist) scale = tanh / dist return x * scale def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = expected_text negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(accuracy / total)) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, limit_pairs=False): num_names = 4 names_generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) entity2same = {} for entity in entities: ret = names_generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def get_hidden_layer(name, net, is_last): if USE_GRU: if is_last: return GRU(128, activation='relu', name=name)(net) else: return GRU(128, return_sequences=True, activation='relu', name=name)(net) else: return Dense(128, activation='relu', name=name)(net) def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) net = GRU(128, return_sequences=True, activation='relu', name='embed')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2a')(net) net = GRU(128, activation='relu', name='embed3')(net) """ for i in range(0, NUM_LAYERS): net = get_hidden_layer('embed' + str(i), net, False) net = get_hidden_layer('embed_last', net, True) """ # if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=modified_loss, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='loss function type: triplet-loss, improved-triplet-loss, modified-loss, or angular-loss') parser.add_argument('--use_l2_norm', type=bool, help='whether to add a l2 norm') parser.add_argument('--test_neighbor_len', type=int, help='size of the neighborhood for testing') parser.add_argument('--train_neighbor_len', type=int, help='size of the neighborhood for training') parser.add_argument('--embedding_type', type=str, help='encoding type to use for input: Kazuma (for Kazuma character embedding) or one-hot') parser.add_argument('--use_GRU', type=bool, help='use GRU or default to MLP') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') args = parser.parse_args() """ LOSS_FUNCTION = None if args.loss_function == 'triplet-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved_triplet_loss': LOSS_FUNCTION=improved_triplet_loss elif args.loss_function == 'modified_loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = true LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) MARGIN = args.margin print('Margin:' + str(MARGIN)) TRAIN_NEIGHBOR_LEN = args.train_neighbor_len TEST_NEIGHBOR_LEN = args.test_neighbor_len print('Train neighbor length: ' + str(TRAIN_NEIGHBOR_LEN)) print('Test neighbor length: ' + str(TEST_NEIGHBOR_LEN)) USE_L2_NORM = args.use_l2_norm print('Use L2Norm: ' + str(USE_L2_NORM)) EMBEDDING_TYPE = args.embedding_type print('Embedding type: ' + EMBEDDING_TYPE) USE_GRU = args.use_GRU print('Use GRU: ' + str(args.use_GRU)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) """ # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train) entity2same_test = generate_names(test, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) match_stats = 0 # num_iter = 100 num_iter = 1 counter = 0 current_model = embedder_model prev_match_stats = 0 while test_match_stats < .9 and counter < num_iter: counter += 1 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
21,847
37.329825
167
py
fuzzyJoiner
fuzzyJoiner-master/old/Triplet_Iteration.py
from sys import argv from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential import numpy as np from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex import Named_Entity_Recognition_Modified #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 DEBUG = False DEBUG_DATA_LENGTH = 100 def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_random_image(img_groups, group_names, gid): gname = group_names[gid] photos = img_groups[gname] pid = np.random.choice(np.arange(len(photos)), size=1)[0] pname = photos[pid] return gname + pname + ".jpg" def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) # i = 0 # while sum(embedding_vector) == 0 and i <= 1000: # embedding_vector = k.emb(word) # i++; # if i == 1000: # print("fail") if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_tokenizer(texts): tokenizer = Tokenizer(num_words=MAX_NB_WORDS) tokenizer.fit_on_texts(texts['anchor'] + texts['negative'] + texts['positive']) return tokenizer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) #removes the new line charecter at the end texts['negative'].append(line_array[2][:-1]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def get_test(texts, sequences, percent): indices = np.arange(sequences['anchor'].shape[0]) np.random.shuffle(indices) ret_sequence = {} ret_sequence['anchor'] = sequences['anchor'][indices] ret_sequence['positive'] = sequences['positive'][indices] ret_sequence['negative'] = sequences['negative'][indices] num_validation_samples = int(percent * sequences['anchor'].shape[0]) ret_train = {} ret_train['anchor'] = ret_sequence['anchor'][:-num_validation_samples] ret_train['positive'] = ret_sequence['positive'][:-num_validation_samples] ret_train['negative'] = ret_sequence['negative'][:-num_validation_samples] ret_test = {} ret_test['anchor']= ret_sequence['anchor'][-num_validation_samples:] ret_test['positive']= ret_sequence['positive'][-num_validation_samples:] ret_test['negative'] = ret_sequence['negative'][-num_validation_samples:] ret_texts = {} texts['anchor'] = np.array(texts['anchor']) texts['positive'] = np.array(texts['positive']) texts['negative'] = np.array(texts['negative']) ret_texts['anchor'] = texts['anchor'][indices] ret_texts['positive'] = texts['positive'][indices] ret_texts['negative'] = texts['negative'][indices] return ret_train, ret_test, ret_texts def triplet_loss(y_true, y_pred): margin = K.constant(1) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def assign_triplets(data, model): sequences = np.concatenate((data['positive'], data['negative'])) unique_sequence = [] anchor_place = {} place_to_item = {} unique_set = set([]) for item in data['anchor']: print(item) item2 = tuple(item) if item2 not in unique_set: anchor_place[item2] = len(unique_sequence) place_to_item[len(unique_sequence)] = item unique_sequence.append(item) unique_set.add(item2) for item in sequences: item2 = tuple(item) if item2 not in unique_set: place_to_item[len(unique_sequence)] = item unique_sequence.append(item) unique_set.add(item2) #make annoy index unique_sequence = np.array(unique_sequence) predictions = model.predict(unique_sequence) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed for i in range(len(predictions)): v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees #create nearest neighbors list anchor_to_nearest = {} new_data = {} new_data['anchor'] = [] new_data['positive'] = [] new_data['negative'] = [] index = 0 while index < len(data['anchor']): name = data['anchor'][index] hash_name = tuple(name) if hash_name in anchor_to_nearest: if anchor_to_nearest[hash_name]: new_data['anchor'].append(name) new_data['negative'].append(place_to_item[anchor_to_nearest[hash_name].pop()]) new_data['positive'].append(data['positive'][index]) index += 1 else: anchor_to_nearest[hash_name] = t.get_nns_by_item(anchor_place[hash_name], 5) new_data['anchor'] = np.array(new_data['anchor']) new_data['positive'] = np.array(new_data['positive']) new_data['negative'] = np.array(new_data['negative']) return new_data def do_annoy(model, texts, tokenizer, verbose): unique_text = [] entity_idx = [] entity2same = {} for i in range(len(texts['anchor'])): if not texts['anchor'][i] in entity2same: entity2same[texts['anchor'][i]] = [] entity_idx.append(len(unique_text)) unique_text.append(texts['anchor'][i]) l = entity2same[texts['anchor'][i]] if texts['positive'][i] not in l: entity2same[texts['anchor'][i]].append(texts['positive'][i]) unique_text.append(texts['positive'][i]) print(entity2same) print(unique_text) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed for i in range(len(predictions)): v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 for index in entity_idx: nearest = t.get_nns_by_vector(predictions[index], 5) print(nearest) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[unique_text[index]]) nearest_text.remove(unique_text[index]) print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) if verbose: print([t.get_distance(index, i) for i in nearest]) overlap = expected_text.intersection(nearest_text) print(overlap) m = len(overlap) match += m no_match += len(expected_text) - m print("match: {} no_match: {}".format(match, no_match)) def print_deb_data(debbuging_data): for i in range(debbuging_data['number']): print('anch: --{}-- pos:--{}-- neg:--{}--'.format(debbuging_data['texts']['anchor'][i], debbuging_data['texts']['positive'][i], debbuging_data['texts']['negative'][i])) print('sequences: anch: --{}-- pos:--{}-- neg:--{}--'.format(debbuging_data['sequences']['anchor'][i], debbuging_data['sequences']['positive'][i], debbuging_data['sequences']['negative'][i])) def debugging_text_and_sequences(reordered_text, training_data, number): debbuging_data = {} debbuging_data['number'] = number debbuging_data['sequences'] = {} debbuging_data['texts'] = {} debbuging_data['sequences']['anchor'] = [] debbuging_data['sequences']['positive'] = [] debbuging_data['sequences']['negative'] = [] debbuging_data['texts']['anchor'] = [] debbuging_data['texts']['positive'] = [] debbuging_data['texts']['negative'] = [] for i in range(number): debbuging_data['texts']['anchor'].append(reordered_text['anchor'][i]) debbuging_data['texts']['positive'].append(reordered_text['positive'][i]) debbuging_data['texts']['negative'].append(reordered_text['negative'][i]) debbuging_data['sequences']['anchor'].append(training_data['anchor'][i]) debbuging_data['sequences']['positive'].append(training_data['positive'][i]) debbuging_data['sequences']['negative'].append(training_data['negative'][i]) return debbuging_data # triples_data = create_triples(IMAGE_DIR) texts = read_file(argv[1]) print("anchor: {} positive: {} negative: {}".format(texts['anchor'][0], texts['positive'][0], texts['negative'][0])) tokenizer = get_tokenizer(texts) print('got tokenizer') sequences = get_sequences(texts, tokenizer) train_data, test_data, reordered_text = get_test(texts, sequences, 0.2) debbuging_data = debugging_text_and_sequences(reordered_text, train_data, 20) number_of_names = len(train_data['anchor']) print('sequenced words') Y_train = np.random.randint(2, size=(1,2,number_of_names)).T embedder = get_embedding_layer(tokenizer) print('got embeddings') main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) net = Flatten(name='flatten')(net) net = Dense(128, activation='relu', name='embed')(net) net = Dense(128, activation='relu', name='embed2')(net) net = Dense(128, activation='relu', name='embed3')(net) net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist')([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist')([net_anchor, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists' )([positive_dist, negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=triplet_loss, metrics=[accuracy]) for x in range(3): train_data = assign_triplets(train_data, base_model) print(train_data) number_of_names = len(train_data['anchor']) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T print('fitting round' + str(x)) model.fit([train_data['anchor'], train_data['positive'], train_data['negative']], Y_train, epochs=5, batch_size=15, validation_split=0.25) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) print("training data predictions") positives = test_positive_model.predict([train_data['anchor'], train_data['positive'], train_data['negative']]) negatives = test_negative_model.predict([train_data['anchor'], train_data['positive'], train_data['negative']]) print("f1score is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_data['anchor'], test_data['positive'], test_data['negative']]) negatives = test_negative_model.predict([test_data['anchor'], test_data['positive'], test_data['negative']]) print("f1score is: {}".format(f1score(positives, negatives))) # model.save('triplet_loss_resnet50.h5') inter_model = Model(input_anchor, net_anchor) do_annoy(inter_model, texts, tokenizer, False) print('annoy on embeddings for debbuging_data') do_annoy(Named_Entity_Recognition_Modified.embedded_representation(embedder), debbuging_data['texts'], tokenizer, True) print('annoy on full model for debbuging_data') do_annoy(inter_model, debbuging_data['texts'], tokenizer, True) print_deb_data(debbuging_data)
14,236
37.374663
199
py
fuzzyJoiner
fuzzyJoiner-master/old/ContrastiveLossLSTM-8.20.18.py
import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') args = parser.parse_args() LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
19,575
35.86629
163
py
fuzzyJoiner
fuzzyJoiner-master/old/seq2seq.py
'''Sequence to sequence example in Keras (character-level). This script demonstrates how to implement a basic character-level sequence-to-sequence model. We apply it to translating short English sentences into short French sentences, character-by-character. Note that it is fairly unusual to do character-level machine translation, as word-level models are more common in this domain. # Summary of the algorithm - We start with input sequences from a domain (e.g. English sentences) and correspding target sequences from another domain (e.g. French sentences). - An encoder LSTM turns input sequences to 2 state vectors (we keep the last LSTM state and discard the outputs). - A decoder LSTM is trained to turn the target sequences into the same sequence but offset by one timestep in the future, a training process called "teacher forcing" in this context. Is uses as initial state the state vectors from the encoder. Effectively, the decoder learns to generate `targets[t+1...]` given `targets[...t]`, conditioned on the input sequence. - In inference mode, when we want to decode unknown input sequences, we: - Encode the input sequence into state vectors - Start with a target sequence of size 1 (just the start-of-sequence character) - Feed the state vectors and 1-char target sequence to the decoder to produce predictions for the next character - Sample the next character using these predictions (we simply use argmax). - Append the sampled character to the target sequence - Repeat until we generate the end-of-sequence character or we hit the character limit. # Data download English to French sentence pairs. http://www.manythings.org/anki/fra-eng.zip Lots of neat sentence pairs datasets can be found at: http://www.manythings.org/anki/ # References - Sequence to Sequence Learning with Neural Networks https://arxiv.org/abs/1409.3215 - Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation https://arxiv.org/abs/1406.1078 ''' from __future__ import print_function from keras.models import Model from keras.layers import Input, LSTM, Dense import numpy as np batch_size = 64 # Batch size for training. epochs = 100 # Number of epochs to train for. latent_dim = 256 # Latent dimensionality of the encoding space. num_samples = 10000 # Number of samples to train on. # Path to the data txt file on disk. data_path = 'fra-eng/fra.txt' # Vectorize the data. input_texts = [] target_texts = [] input_characters = set() target_characters = set() with open(data_path, 'r', encoding='utf-8') as f: lines = f.read().split('\n') for line in lines[: min(num_samples, len(lines) - 1)]: input_text, target_text = line.split('\t') # We use "tab" as the "start sequence" character # for the targets, and "\n" as "end sequence" character. target_text = '\t' + target_text + '\n' input_texts.append(input_text) target_texts.append(target_text) for char in input_text: if char not in input_characters: input_characters.add(char) for char in target_text: if char not in target_characters: target_characters.add(char) input_characters = sorted(list(input_characters)) target_characters = sorted(list(target_characters)) num_encoder_tokens = len(input_characters) num_decoder_tokens = len(target_characters) max_encoder_seq_length = max([len(txt) for txt in input_texts]) max_decoder_seq_length = max([len(txt) for txt in target_texts]) print('Number of samples:', len(input_texts)) print('Number of unique input tokens:', num_encoder_tokens) print('Number of unique output tokens:', num_decoder_tokens) print('Max sequence length for inputs:', max_encoder_seq_length) print('Max sequence length for outputs:', max_decoder_seq_length) input_token_index = dict( [(char, i) for i, char in enumerate(input_characters)]) target_token_index = dict( [(char, i) for i, char in enumerate(target_characters)]) encoder_input_data = np.zeros( (len(input_texts), max_encoder_seq_length, num_encoder_tokens), dtype='float32') decoder_input_data = np.zeros( (len(input_texts), max_decoder_seq_length, num_decoder_tokens), dtype='float32') decoder_target_data = np.zeros( (len(input_texts), max_decoder_seq_length, num_decoder_tokens), dtype='float32') for i, (input_text, target_text) in enumerate(zip(input_texts, target_texts)): for t, char in enumerate(input_text): encoder_input_data[i, t, input_token_index[char]] = 1. for t, char in enumerate(target_text): # decoder_target_data is ahead of decoder_input_data by one timestep decoder_input_data[i, t, target_token_index[char]] = 1. if t > 0: # decoder_target_data will be ahead by one timestep # and will not include the start character. decoder_target_data[i, t - 1, target_token_index[char]] = 1. # Define an input sequence and process it. encoder_inputs = Input(shape=(None, num_encoder_tokens)) encoder = LSTM(latent_dim, return_state=True) encoder_outputs, state_h, state_c = encoder(encoder_inputs) # We discard `encoder_outputs` and only keep the states. encoder_states = [state_h, state_c] # Set up the decoder, using `encoder_states` as initial state. decoder_inputs = Input(shape=(None, num_decoder_tokens)) # We set up our decoder to return full output sequences, # and to return internal states as well. We don't use the # return states in the training model, but we will use them in inference. decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True) decoder_outputs, _, _ = decoder_lstm(decoder_inputs, initial_state=encoder_states) decoder_dense = Dense(num_decoder_tokens, activation='softmax') decoder_outputs = decoder_dense(decoder_outputs) # Define the model that will turn # `encoder_input_data` & `decoder_input_data` into `decoder_target_data` model = Model([encoder_inputs, decoder_inputs], decoder_outputs) # Run training model.compile(optimizer='rmsprop', loss='categorical_crossentropy') model.fit([encoder_input_data, decoder_input_data], decoder_target_data, batch_size=batch_size, epochs=epochs, validation_split=0.2) # Save model model.save('s2s.h5') # Next: inference mode (sampling). # Here's the drill: # 1) encode input and retrieve initial decoder state # 2) run one step of decoder with this initial state # and a "start of sequence" token as target. # Output will be the next target token # 3) Repeat with the current target token and current states # Define sampling models encoder_model = Model(encoder_inputs, encoder_states) decoder_state_input_h = Input(shape=(latent_dim,)) decoder_state_input_c = Input(shape=(latent_dim,)) decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c] decoder_outputs, state_h, state_c = decoder_lstm( decoder_inputs, initial_state=decoder_states_inputs) decoder_states = [state_h, state_c] decoder_outputs = decoder_dense(decoder_outputs) decoder_model = Model( [decoder_inputs] + decoder_states_inputs, [decoder_outputs] + decoder_states) # Reverse-lookup token index to decode sequences back to # something readable. reverse_input_char_index = dict( (i, char) for char, i in input_token_index.items()) reverse_target_char_index = dict( (i, char) for char, i in target_token_index.items()) def decode_sequence(input_seq): # Encode the input as state vectors. states_value = encoder_model.predict(input_seq) # Generate empty target sequence of length 1. target_seq = np.zeros((1, 1, num_decoder_tokens)) # Populate the first character of target sequence with the start character. target_seq[0, 0, target_token_index['\t']] = 1. # Sampling loop for a batch of sequences # (to simplify, here we assume a batch of size 1). stop_condition = False decoded_sentence = '' while not stop_condition: output_tokens, h, c = decoder_model.predict( [target_seq] + states_value) # Sample a token sampled_token_index = np.argmax(output_tokens[0, -1, :]) sampled_char = reverse_target_char_index[sampled_token_index] decoded_sentence += sampled_char # Exit condition: either hit max length # or find stop character. if (sampled_char == '\n' or len(decoded_sentence) > max_decoder_seq_length): stop_condition = True # Update the target sequence (of length 1). target_seq = np.zeros((1, 1, num_decoder_tokens)) target_seq[0, 0, sampled_token_index] = 1. # Update states states_value = [h, c] return decoded_sentence for seq_index in range(100): # Take one sequence (part of the training set) # for trying out decoding. input_seq = encoder_input_data[seq_index: seq_index + 1] decoded_sentence = decode_sequence(input_seq) print('-') print('Input sentence:', input_texts[seq_index]) print('Decoded sentence:', decoded_sentence)
9,104
40.013514
79
py
fuzzyJoiner
fuzzyJoiner-master/old/matcher.py
#using tutorial https://suhas.org/sqlalchemy-tutorial/ from sys import argv from matcher_functions import * #establish connection to database con, meta = connect(argv[1], argv[2], argv[3]) #load pairs from database aliases = get_aliases(con, meta) #create dictionaries assingning serial numbers to names and names from serial numbers num_to_word, word_to_num = create_double_num_dicts(aliases) #load the buckets from the database bucket_list is aranges as follows: #bucket_list[pair_of_buckets][bucket(this must be 0 or 1)][name (this represents a single name)][0 for number and 1 for pre-procced name] bucket_list, bucket_words = load_good_buckets('wordtable1', 'wordtable2', word_to_num, con, meta) #print out the number of names that are possible to get just based on bucketing: impossible = get_impossible(aliases, bucket_list, num_to_word) print("possible matches: " + str(len(aliases) - len(impossible))) #next make a list to store the outcomes of all our tests: matches_list = [] #then run our tests matches_list.append(run_test(lambda x : x.replace(" ", ""), lambda name1, name2 : name1 in name2 or name2 in name1, num_to_word, bucket_list)) matches_list.append(run_test(lambda x : set(x.split()), lambda name1, name2 : name1.issubset(name2) or name2.issubset(name1), num_to_word, bucket_list)) matches_list.append(run_special_test(bucket_list, num_to_word)) #next create a test dictionary relating each item in the first set to k items in other set test_dict = make_test_dict(set([]).union(*matches_list), 1000) #use this dictionary to calculate and print the f-score print("fscore: " + str(fscore(aliases, test_dict, 1))) #next export the items we missed export_missed(aliases, test_dict, con, meta) #lastly export the items we could not have gotten since they were not in the same bucket: export_unbucketed(impossible, con, meta)
1,840
62.482759
152
py
fuzzyJoiner
fuzzyJoiner-master/old/cleanser.py
from sys import argv from os import listdir from os.path import isfile, join if __name__ == '__main__': parser = argparse.ArgumentParser( description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) parser.add_argument('-f', dest="input_file or directory", help="file to cleanse file") parser.add_argument('-o', dest="output_file", help="cleansed data file name") parser.add_argument('-t', dest="entity_type", help="names or people") parser.add_argument('-u', dest="function", help="use test parsing function") parser.add_argument('-n', dest="number", help="process only this many from the file (useful for debugging)") args = parser.parse_args() if args.entity_type == 'names': cleaner = NameDataCleanser(number) else: cleaner = GenericDataCleanser(entity_type, function, number) cleaner.parse_file(args.filename, args.output, args.rejects) if len(args) >= 3: parsing_function = self.good_name_data if len(args) >= 4: parsing_function = self.function_dictionary[args[3]] if len(args) >= 6: self.number_of_names = int(args[5]) if args[4] == "test": parsing_function = self.function_dictionary[(args[3] + args[4])] elif args[4] == "get": self.get = True onlyfiles = [f for f in listdir(args[1]) if isfile(join(args[1], f))] output_file = open(args[2], "w", encoding='utf-8') output_rejects_file = open('./rejects.txt', 'w', encoding='utf-8') output_file = open(args[2], "w", encoding='utf-8') for file_path in onlyfiles: input_file = open(args[1] + "/" + file_path, encoding='utf-8') self.parse_file(input_file, output_file, parsing_function, output_rejects_file) output_rejects_file.close() input_file.close() #test_file.close(); output_file.close() else: print ("too few arguments please enter arguments in the following format: input_file output_file [function [get x OR test x]]") cleaner = GenericDataCleanser() cleaner.clean_file(argv)
1,994
38.9
130
py
fuzzyJoiner
fuzzyJoiner-master/old/image_join3.py
0
0
0
py
fuzzyJoiner
fuzzyJoiner-master/old/matcher_class.py
from matcher_functions import load_good_buckets, create_double_num_dicts, connect, get_aliases class matcher(object): def __init__(self, user, password, database, test_pairs, bucket_number): con, meta = connect(user, password, database) num_to_word, word_to_num = create_double_num_dicts(get_aliases(con, meta)) bucket_list, bucket_words = load_good_buckets('wordtable1', 'wordtable2', word_to_num, con, meta) self.rarity_match = {} for pair in bucket_list: if len(pair[0]) <= bucket_number and len(pair[1]) <= bucket_number: for i in range(len(pair[0])): for j in range(len(pair[1])): self.rarity_match[pair[0][i][1]] = pair[1][j][1] self.test_pairs = test_pairs def special_test(self, name1, name2): temp1 = False temp2 = False if name1 in self.rarity_match: temp1 = self.rarity_match[name1] == name2 if name2 in self.rarity_match: temp2 = self.rarity_match[name2] == name1 return temp1 or temp2 def match(self, name1, name2): for test_pair in self.test_pairs: temp1 = test_pair[0](name1) temp2 = test_pair[0](name2) if test_pair[1](temp1, temp2): return True return self.special_test(name1, name2)
1,366
46.137931
105
py
fuzzyJoiner
fuzzyJoiner-master/old/Named_Entity_Recognition_Modified.py
""" This code is modified from https://github.com/fchollet/keras/blob/master/examples/pretrained_word_embeddings.py and ttps://github.com/fchollet/keras/blob/master/examples/ for our own purposes """ from __future__ import absolute_import from __future__ import print_function import numpy as np from matcher_functions import connect, get_aliases, load_good_buckets, create_double_num_dicts from matcher_class import matcher import os from sqlalchemy import Table, Column, Integer, String, ForeignKey, Float import sys from sys import argv import numpy as np from embeddings import KazumaCharEmbedding import random from annoy import AnnoyIndex import random from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from keras.utils import to_categorical from keras.optimizers import RMSprop from keras import backend as K from keras import regularizers from keras.regularizers import L1L2 from sqlalchemy.sql import select import random import argparse BASE_DIR = './Machine_Learning/' GLOVE_DIR = BASE_DIR + 'glove/' TEXT_DATA_DIR = BASE_DIR + 'nerData' DO_ANN_ON_EMBEDDINGS = False # number of words an entity is allowed to have # distribution of number of words in peoples names can be found in peopleNamesDisbn # distribution of number of words in company names can be found in companyNamesDisbn # Note most of the names above that are fairly esoteric or just plain noise. Included is # python code to remove them MAX_SEQUENCE_LENGTH = 10 # Total number of unique tokens in peoples names is 90K, including a lot of non-English names. To remove those # we use an egregious hack to check if its UTF-8 # Total number of unique tokens in company names is 37K # Assuming no overlap between the two we get about 127K. We may need to tweak this parameter as we go # but according to the Keras documentation, this can even be left unset MAX_NB_WORDS = 150000 # Size of embeddings from Glove (we will try the 100 dimension encoding to start with) EMBEDDING_DIM = 100 VALIDATION_SPLIT = 0.05 def check_for_zeroes(to_check, intro_string): found = 0 for vector in to_check: if sum(vector) == 0: print(intro_string + str(vector)) found += 1 if not found: print(intro_string + " no problems found") else: print(intro_string + ' found this many: '+ str(found)) def get_diff_names_with_overlap(con, aliases): no_match_texts = [] texts = [] # load a mapping from entity id to names entityid2names = {} for row in aliases: if row[2] in entityid2names: names = entityid2names[row[2]]; else: names = [] entityid2names[row[2]] = names names.append(row[0]) names.append(row[1]) texts.append(row[0]) print(len(aliases)) print("getting word2entities") # load a mapping from words to entities word2entities = {} rows = con.execute("select word, entities from word2entities;") for row in rows: word2entities[row[0]] = row[1] for index in range(len(texts)): name_arr = texts[index].split() new_text = '' for n in name_arr: # a name part may have been filtered out of word2entities if n not in word2entities: continue if new_text: break for e in word2entities[n]: if e == texts[index] or e not in entityid2names: # if the entity is the same as this anchor's text skip it continue names = entityid2names[e] for x in names: if n in x: new_text = x break if new_text: no_match_texts.append(new_text) else: no_match_texts.append(texts[index + 1]) print("done processing matches with overlap") return no_match_texts def get_diff_names_with_no_overlap(aliases): entitylist = [] for row in aliases: entitylist.append(row[0]) s = [i for i in range(len(entitylist))] random.shuffle(s) ret = [] for i in range(len(entitylist)): if s[i] == i : s[i] = s[i+1] s[i+1] = i ret.append(entitylist[s[i]]) return ret #this returns a new set of texts to use as similar non-matches for texts1 def get_no_match_texts(user, password, db, texts1): def get_non_match(name1, bucket_words, matching_set): for word in name1.split(" "): if word in bucket_words: bucket = bucket_words[word] else: return None if len(bucket[1]) > 1: for name2 in bucket[1]: if (name1, name2[1]) not in matching_set: return name2[1] return None no_match_texts = [] #this should not be done here and needs to be fixed up before more work is done #it should instead be done by a singel function in matcher_functions #establish connection to database con, meta = connect(user, password, db) #load pairs from database aliases = get_aliases(con, meta) #create dictionaries assigning serial numbers to names and names from serial numbers num_to_word, word_to_num = create_double_num_dicts(aliases) #load the buckets from the database bucket_list is aranges as follows: #bucket_list[pair_of_buckets][bucket(this must be 0 or 1)][name (this represents a single name)][0 for number and 1 for pre-procced name] bucket_list, bucket_words = load_good_buckets('wordtable1', 'wordtable2', word_to_num, con, meta) for index in range(len(texts1)): new_text = get_non_match(texts1[index], bucket_words, aliases) if new_text == None: new_text = texts1[(index + 1) % len(texts1)] no_match_texts.append(new_text) return no_match_texts def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def eucl_dist_output_shape(shapes): shape1, shape2 = shapes return (shape1[0], 1) def contrastive_loss(y_true, y_pred): '''Contrastive loss from Hadsell-et-al.'06 http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf Equation 4 ''' margin = 1 return K.mean((1 - y_true) * K.square(y_pred) + y_true * K.square(K.maximum(margin - y_pred, 0))) #for now will take any bad pairs, will take only relivent ones later def create_pairs(x, y, z): '''Positive and negative pair creation. Alternates between positive and negative pairs. ''' pairs = [] labels = [] for index in range(len(x)): pairs += [[x[index], y[index]]] pairs += [[x[index], z[index]]] labels += [0, 1] # n = min([len(digit_indices[d]) for d in range(10)]) - 1 # for d in range(10): # for i in range(n): # z1, z2 = digit_indices[d][i], digit_indices[d][i + 1] # pairs += [[x[z1], x[z2]]] # inc = random.randrange(1, 10) # dn = (d + inc) % 10 # z1, z2 = digit_indices[d][i], digit_indices[dn][i] # pairs += [[x[z1], x[z2]]] # labels += [1, 0] return np.array(pairs), np.array(labels) def create_base_network(input_dim, embedding_layer, reg): '''Base network to be shared (eq. to feature extraction). ''' seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) seq.add(Dense(128, input_shape=(input_dim,), activation='relu')) seq.add(Dropout(0.1)) seq.add(Dense(128, activation='relu')) # kernel_regularizer=reg)) seq.add(Dropout(0.1)) final_layer = Dense(128, activation='relu') seq.add(final_layer) # kernel_regularizer=reg)) return seq, final_layer def embedded_representation(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def compute_accuracy(predictions, labels): '''Compute classification accuracy with a fixed threshold on distances. ''' same_correct = 0 diff_correct = 0 for i in range(len(labels)): if predictions[i] < 0.5 and labels[i] == 0: same_correct += 1 elif predictions[i] >= 0.5 and labels[i] == 1: diff_correct += 1 print("Precision computation: same - " + str(same_correct) + " different: " + str(diff_correct) + " from total: " + str(len(labels))) return (same_correct + diff_correct) / len(labels) def get_aliases_with_ids(con, meta): #load pairs from database aliases = con.execute("select alias1, alias2, entityid from aliases order by entityid;") entities = [] for row in aliases: entities.append((row[0], row[1], row[2])) return entities def f1score(predictions, labels): #labels[predictions.ravel() < 0.5].sum() predictions = predictions.ravel() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(labels)): if predictions[i] < 0.5: if labels[i] == 0: true_positive += 1 else: false_positive += 1 elif labels[i] == 0: false_negitive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) print (fscore) return fscore #compute accuracy using a rule based matcher def sequence_to_word(sequence, reverse_word_index): return " ".join([reverse_word_index[x] for x in sequence if x in reverse_word_index]) def sequence_pair_to_word_pair(sequence_pair, reverse_word_index): return [sequence_to_word(sequence_pair[0], reverse_word_index), sequence_to_word(sequence_pair[1], reverse_word_index)] if __name__ == '__main__': print('Processing text dataset') parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) parser.add_argument('-u', dest="user", help="username") parser.add_argument('-p', dest="password", help="password") parser.add_argument('-d', dest="db", help="dbname") parser.add_argument('-a', dest="num_pairs", help="number of same pairs in db", nargs='?', default=2, type=int) args = parser.parse_args() texts1 = [] # list of text samples in part 1 texts2 = [] # list of text samples in part 2 #change to get from sql and not read from file con, meta = connect(args.user, args.password, args.db) aliases = get_aliases_with_ids(con, meta) unique_aliases = [] # collect up all the anchors that are unique (anchors will get repeated if num_pairs > 2) prev = int(aliases[0][2]) unique_aliases.append(aliases[0]) for tuple in aliases: texts1.append(tuple[0]) texts2.append(tuple[1]) if int(tuple[2]) != prev: unique_aliases.append(tuple) prev = int(tuple[2]) print('Found %s texts.' % len(texts1)) texts3 = [] print(len(unique_aliases)) print(len(texts1)) print(len(texts2)) # get the different pairs if args.num_pairs == 2: print("args num pairs is 2") texts3.extend(get_diff_names_with_overlap(connect(args.user, args.password, args.db)[0], unique_aliases)) elif args.num_pairs == 3: print("args num pairs is 3") texts3.extend(get_diff_names_with_overlap(connect(args.user, args.password, args.db)[0], unique_aliases)) texts3.extend(get_diff_names_with_no_overlap(unique_aliases)) elif args.num_pairs == 4: print("args num pairs is 4") texts3.extend(get_diff_names_with_overlap(connect(args.user, args.password, args.db)[0], unique_aliases)) texts3.extend(get_diff_names_with_no_overlap(unique_aliases)) texts3.extend(get_diff_names_with_no_overlap(unique_aliases)) print(len(texts3)) assert len(texts1) == len(texts2) assert len(texts2) == len(texts3), str(len(texts3)) # vectorize the text samples into a 2D integer tensor tokenizer = Tokenizer(num_words=MAX_NB_WORDS) tokenizer.fit_on_texts(texts1 + texts2 + texts3) # this step should get similar but non-matching items to keep for later matching # this step creates a sequence of words ids for each word in each label sequences1 = tokenizer.texts_to_sequences(texts1) for sc in range(len(texts1)): if sum(sequences1[sc]) == 0: print('here is a problem word :' + texts1[sc] + '::') sequences2 = tokenizer.texts_to_sequences(texts2) no_match_sequences = tokenizer.texts_to_sequences(texts3) word_index = tokenizer.word_index check_for_zeroes(sequences1, " sequences") print('Found %s unique tokens.' % len(word_index)) annoy_data1 = pad_sequences(sequences1, maxlen=MAX_SEQUENCE_LENGTH) annoy_data2 = pad_sequences(sequences2, maxlen=MAX_SEQUENCE_LENGTH) no_match_data = pad_sequences(no_match_sequences, maxlen=MAX_SEQUENCE_LENGTH) print('Shape of data1 tensor:', annoy_data1.shape) print('Shape of data2 tensor:', annoy_data2.shape) # split the data into a training set and a validation set, shuffling items indices = np.arange(annoy_data1.shape[0]) np.random.shuffle(indices) texts1 = np.array(texts1) texts2 = np.array(texts2) texts3 = np.array(texts3) texts1 = texts1[indices] texts2 = texts2[indices] texts3 = texts3[indices] # for i in range(len(texts1)): # print(texts1[i] + " paired with: " + texts2[i]) # print(texts1[i] + " paired with: " + texts3[i]) data1 = annoy_data1[indices] data2 = annoy_data2[indices] no_match_data = no_match_data[indices] num_validation_samples = int(VALIDATION_SPLIT * data1.shape[0]) x_train = data1[:-num_validation_samples] y_train = data2[:-num_validation_samples] z_train = no_match_data[:-num_validation_samples] x_test = data1[-num_validation_samples:] y_test = data2[-num_validation_samples:] z_test = no_match_data[-num_validation_samples:] print('Preparing embedding matrix.') # prepare embedding matrix num_words = len(word_index) + 1 # word_index is indexed from 1-N embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) kz = KazumaCharEmbedding() for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector # note that we set trainable = False so as to keep the embeddings fixed check_for_zeroes(embedding_matrix, "here is the first pass") embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) print('Training model.') # the data, shuffled and split between train and test sets #need to change this not sure how input_dim = MAX_SEQUENCE_LENGTH epochs = 10 # create training+test positive and negative pairs # these next lines also need to change #digit_indices = [np.where(y_train == i)[0] for i in range(10)] print("x_train {} , y_train {} , z_train {} ".format(x_train, y_train, z_train)) tr_pairs, tr_y = create_pairs(x_train, y_train, z_train) #digit_indices = [np.where(y_test == i)[0] for i in range(10)] te_pairs, te_y = create_pairs(x_test, y_test, z_test) print (len(tr_y)) # network definition base_network, final_layer = create_base_network(input_dim, embedding_layer, L1L2(0.0,0.0)) input_a = Input(shape=(input_dim,)) input_b = Input(shape=(input_dim,)) # because we re-use the same instance `base_network`, # the weights of the network # will be shared across the two branches processed_a = base_network(input_a) processed_b = base_network(input_b) distance = Lambda(euclidean_distance, output_shape=eucl_dist_output_shape)([processed_a, processed_b]) model = Model([input_a, input_b], distance) print(base_network.summary()) # train rms = RMSprop() #change the optimizer (adam) model.compile(loss=contrastive_loss, optimizer=rms) model.fit([tr_pairs[:, 0], tr_pairs[:, 1]], tr_y, batch_size=128, epochs=epochs) # compute final accuracy on training and test sets #add an LSTM layer (later) # testpairs = [[lambda x : x.replace(" ", ""), lambda name1, name2 : name1 in name2 or name2 in name1], # [lambda x : set(x.split()), lambda name1, name2 : name1.issubset(name2) or name2.issubset(name1)]] # matcher = matcher(argv[1], argv[2], argv[3], test_pairs, 1) pred_learning = model.predict([tr_pairs[:, 0], tr_pairs[:, 1]]) # out = model.layers[2].get_output_at(0) # inp = model.input # func = K.function([inp], [out]) # evaluation functions # print("here should be a vector") # print(func([tr_pairs[:, 0][0], tr_pairs[:, 1][0]])) # Testing # print (layer_outs) tr_acc = compute_accuracy(pred_learning, tr_y) tr_f1 = f1score(pred_learning, tr_y) pred = model.predict([te_pairs[:, 0], te_pairs[:, 1]]) pred_learning = np.append(pred_learning, pred, axis=0) te_acc = compute_accuracy(pred, te_y) te_f1 = f1score(pred, te_y) x_test_text = texts1[-num_validation_samples:] y_test_text = texts2[-num_validation_samples:] z_test_text = texts3[-num_validation_samples:] text_pairs, text_y = create_pairs(x_test_text, y_test_text, z_test_text) # for i in range(len(text_pairs)): # print(str(text_pairs[i])) # print(pred[i]) # print(model.predict([np.array([te_pairs[i, 0]]), np.array([te_pairs[i, 1]])])) # from https://github.com/spotify/annoy f = 128 if DO_ANN_ON_EMBEDDINGS: inter_model = embedded_representation(embedding_layer) else: inter_model = Model(input_a, processed_a) intermediate_output1 = inter_model.predict(x_test) intermediate_output2 = inter_model.predict(y_test) intermediate_output3 = inter_model.predict(z_test) mid_predictions = np.concatenate((intermediate_output1, intermediate_output2, intermediate_output3)) # print(mid_predictions[0]) # print (len(mid_predictions[0])) if DO_ANN_ON_EMBEDDINGS: t = AnnoyIndex(len(mid_predictions[0]), metric='euclidean') # Length of item vector that will be indexed else: t = AnnoyIndex(f, metric='euclidean') # Length of item vector that will be indexed for i in range(len(mid_predictions)): v = mid_predictions[i] t.add_item(i, v) t.build(100) # 100 trees t.save('test.ann') # ... all_texts = np.concatenate((x_test_text, y_test_text, z_test_text)) match = 0 no_match = 0 for index in range(len(x_test_text)): nearest = t.get_nns_by_vector(mid_predictions[index], 5) # print(nearest) nearest_text = [all_texts[i] for i in nearest] # print("query={} names = {} true_match = {} reject= {}".format(x_test_text[index], nearest_text, y_test_text[index], z_test_text[index])) for i in nearest: # print(all_texts[i]) if i >= len(x_test_text) and (i < len(x_test_text) + len(y_test_text)): arr = np.array([y_test[i - len(x_test_text)]]) elif i >= len(x_test_text) + len(y_test_text): arr = np.array([z_test[i - len(x_test_text) - len(y_test_text)]]) else: arr = np.array([x_test[i]]) # print(model.predict([np.array([x_test[index]]), arr])) # print(t.get_distance(index, i)) # print("true match prediction:") # print(model.predict([np.array([x_test[index]]), np.array([y_test[index]])])) # print("true match distance:") # print(t.get_distance(index, index + len(x_test_text))) # print("true reject prediction:") # print(model.predict([np.array([x_test[index]]), np.array([z_test[index]])])) # print("true reject distance:") # print(t.get_distance(index, index + len(x_test_text) + len(y_test_text))) if y_test_text[index] in nearest_text: match += 1 # print("MATCH FOUND") else: no_match += 1 print("match: {} no_match: {}".format(match, no_match)) print("Machine Learning Accuracy") print(tr_acc) print('* Accuracy on training set: %0.2f%%' % (100 * tr_acc)) print('* Accuracy on test set: %0.2f%%' % (100 * te_acc)) print('* f1score on the training set: %0.4f' % (tr_f1)) print('* f1socre on test set: %0.4f' % (te_f1)) reverse_word_index = {v: k for k, v in tokenizer.word_index.items()} print(tr_pairs) print(sequence_to_word(tr_pairs[0][0], reverse_word_index)) print(sequence_to_word(tr_pairs[1][1], reverse_word_index)) print(tr_y[0]) test_pairs = [[lambda x : x.replace(" ", ""), lambda name1, name2 : name1 in name2 or name2 in name1], [lambda x : set(x.split()), lambda name1, name2 : name1.issubset(name2) or name2.issubset(name1)]] matcher = matcher(args.user, args.password, args.db, test_pairs, 1) pred_rules = np.asarray([int(not matcher.match(*sequence_pair_to_word_pair(name_pair, reverse_word_index))) for name_pair in tr_pairs]) tr_acc = compute_accuracy(pred_rules, tr_y) tr_f1 = f1score(pred_rules, tr_y) pred = np.asarray([int(not matcher.match(*sequence_pair_to_word_pair(name_pair, reverse_word_index))) for name_pair in te_pairs]) pred_rules = np.append(pred_rules, pred, axis=0) te_acc = compute_accuracy(pred, te_y) te_f1 = f1score(pred, te_y) print("Rule-Based Accuracy") print('* Accuracy on training set (rules): %0.2f%%' % (100 * tr_acc)) print('* Accuracy on test set (rules): %0.2f%%' % (100 * te_acc)) print('* f1score on the training set: %0.4f' % (tr_f1)) print('* f1score on test set: %0.4f' % (te_f1)) con, meta = connect(args.user, args.password, args.db) execute_pairs = [] if 'predictions' in meta.tables: meta.tables['predictions'].drop(con) predictions = Table('predictions', meta, Column('name1', String), Column('name2', String), Column('rule_predict', Integer), Column('learning_predict', Float), Column('true_pair', Integer), Column('te_or_tr', String), extend_existing=True) zipping_string = ('name1', 'name2', 'true_pair', 'rule_predict', 'learning_predict', 'te_or_tr') print(len(tr_y)) print(len(tr_pairs)) print(len(pred_rules)) print(len(pred_learning)) print(len(te_y)) print(len(te_pairs)) for i in range(len(tr_y)): execute_pairs.append(dict(zip(zipping_string, (sequence_to_word(tr_pairs[i][0], reverse_word_index), sequence_to_word(tr_pairs[i][1], reverse_word_index), int(tr_y[i]), int(pred_rules[i]), float(pred_learning[i][0].item()), 'tr')))) offset = len(tr_y) for i in range(len(te_y)): execute_pairs.append(dict(zip(zipping_string, (sequence_to_word(tr_pairs[i][0], reverse_word_index), sequence_to_word(te_pairs[i][1], reverse_word_index), int(te_y[i]), int(pred_rules[offset + i]), float(pred_learning[offset + i][0].item()), 'te')))) meta.create_all(con) con.execute(predictions.insert(), execute_pairs)
23,976
36.289269
258
py
fuzzyJoiner
fuzzyJoiner-master/old/Named_Entity_Recognition.py
""" This code is modified from https://github.com/fchollet/keras/blob/master/examples/pretrained_word_embeddings.py and ttps://github.com/fchollet/keras/blob/master/examples/ for our own purposes """ from __future__ import absolute_import from __future__ import print_function import numpy as np from matcher_functions import connect, get_aliases, load_good_buckets, create_double_num_dicts from matcher_class import matcher import os from sqlalchemy import Table, Column, Integer, String, ForeignKey, Float import sys from sys import argv import numpy as np from embeddings import KazumaCharEmbedding import random from annoy import AnnoyIndex import random from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from keras.utils import to_categorical from keras.optimizers import RMSprop from keras import backend as K from keras import regularizers from keras.regularizers import L1L2 BASE_DIR = './Machine_Learning/' # directory containing glove encodings from Wikipedia (we can swap this out for another encoding later) # Download glove.6B.zip from https://nlp.stanford.edu/projects/glove/ GLOVE_DIR = BASE_DIR + 'glove/' TEXT_DATA_DIR = BASE_DIR + 'nerData' # number of words an entity is allowed to have # distribution of number of words in peoples names can be found in peopleNamesDisbn # distribution of number of words in company names can be found in companyNamesDisbn # Note most of the names above that are fairly esoteric or just plain noise. Included is # python code to remove them MAX_SEQUENCE_LENGTH = 10 # Total number of unique tokens in peoples names is 90K, including a lot of non-English names. To remove those # we use an egregious hack to check if its UTF-8 # Total number of unique tokens in company names is 37K # Assuming no overlap between the two we get about 127K. We may need to tweak this parameter as we go # but according to the Keras documentation, this can even be left unset MAX_NB_WORDS = 140000 # Size of embeddings from Glove (we will try the 100 dimension encoding to start with) EMBEDDING_DIM = 100 VALIDATION_SPLIT = 0.05 def check_for_zeroes(to_check, intro_string): found = 0 for vector in to_check: if sum(vector) == 0: print(intro_string + str(vector)) found += 1 if not found: print(intro_string + " no problems found") else: print(intro_string + ' found this many: '+ str(found)) # first, build index mapping words in the glove embeddings set # to their embedding vector. This is a straightforward lookup of # words in Glove and then their embeddings which should be a 100 sized array of floats # print('Reading word embeddings: Indexing word vectors.') # embeddings_index = {} # f = open(os.path.join(GLOVE_DIR, 'glove.6B.100d.txt')) # for line in f: # values = line.split() # word = values[0] # coefs = np.asarray(values[1:], dtype='float32') # embeddings_index[word] = coefs # f.close() # print('Found %s word vectors.' % len(embeddings_index)) # second, prepare text samples and their labels print('Processing text dataset') texts1 = [] # list of text samples in part 1 texts2 = [] # list of text samples in part 2 labels_index = {} # dictionary mapping label name to numeric id - here the label name is just the name of the file in the data dir name_list = sorted(os.listdir(TEXT_DATA_DIR)) name = name_list[0] label_id = len(labels_index) labels_index[name] = label_id fpath = os.path.join(TEXT_DATA_DIR, name) if os.path.isdir(fpath): raise ValueError('bad data directory') if sys.version_info < (3,): f = open(fpath) else: f = open(fpath, encoding='latin-1') #change to get from sql and not read from file con, meta = connect(argv[1], argv[2], argv[3]) #load pairs from database aliases = get_aliases(con, meta) for pair in aliases: num_tokens = len(pair[0].strip().split(' ')) + len(pair[1].strip().split(' ')) if 0 < num_tokens < MAX_SEQUENCE_LENGTH: texts1.append(pair[0]) texts2.append(pair[1]) f.close() #this returns a new set of texts to use as similar non-matches for texts1 def get_no_match_texts(argv, texts1): def get_non_match(name1, bucket_words, matching_set): for word in name1.split(" "): if word in bucket_words: bucket = bucket_words[word] else: return None if len(bucket[1]) > 1: for name2 in bucket[1]: if (name1, name2[1]) not in matching_set: return name2[1] return None no_match_texts = [] #this should not be done here and needs to be fixed up before more work is done #it should instead be done by a singel function in matcher_functions #establish connection to database con, meta = connect(argv[1], argv[2], argv[3]) #load pairs from database aliases = get_aliases(con, meta) #create dictionaries assingning serial numbers to names and names from serial numbers num_to_word, word_to_num = create_double_num_dicts(aliases) #load the buckets from the database bucket_list is aranges as follows: #bucket_list[pair_of_buckets][bucket(this must be 0 or 1)][name (this represents a single name)][0 for number and 1 for pre-procced name] bucket_list, bucket_words = load_good_buckets('wordtable1', 'wordtable2', word_to_num, con, meta) for index in range(len(texts1)): new_text = get_non_match(texts1[index], bucket_words, aliases) if new_text == None: new_text = texts1[(index + 1) % len(texts1)] no_match_texts.append(new_text) return no_match_texts print('Found %s texts.' % len(texts1)) # finally, vectorize the text samples into a 2D integer tensor tokenizer = Tokenizer(num_words=MAX_NB_WORDS) #gets the special no match charectars texts3 = get_no_match_texts(argv, texts1) #this removes all non-ascii charectars from the 3 sets of strings texts1 = [str(item) for item in texts1] texts2 = [str(item) for item in texts2] texts3 = [str(item) for item in texts3] # for i in range(len(texts1)): print ("1 is {}".format(texts1[1])) print ("2 is {}".format(texts2[1])) print ("3 is {}".format(texts3[1])) tokenizer.fit_on_texts(texts1 + texts2 + texts3) #this step should get similar but non-matching items to keep for later matching # this step creates a sequence of words ids for each word in each label sequences1 = tokenizer.texts_to_sequences(texts1) for sc in range(len(texts1)): if sum(sequences1[sc]) == 0: print('here is a problem word :' + texts1[sc] + '::') sequences2 = tokenizer.texts_to_sequences(texts2) no_match_sequences = tokenizer.texts_to_sequences(texts3) word_index = tokenizer.word_index check_for_zeroes(sequences1, " sequences") print('Found %s unique tokens.' % len(word_index)) annoy_data1 = pad_sequences(sequences1, maxlen=MAX_SEQUENCE_LENGTH) annoy_data2 = pad_sequences(sequences2, maxlen=MAX_SEQUENCE_LENGTH) no_match_data = pad_sequences(no_match_sequences, maxlen=MAX_SEQUENCE_LENGTH) # print (data1[0]) # print (data2[0]) # print (texts1[0]) # print (texts2[0]) #labels = to_categorical(np.asarray(labels)) print('Shape of data1 tensor:', annoy_data1.shape) #print('Shape of label tensor:', labels.shape) print('Shape of data2 tensor:', annoy_data2.shape) # split the data into a training set and a validation set indices = np.arange(annoy_data1.shape[0]) np.random.shuffle(indices) data1 = annoy_data1[indices] data2 = annoy_data2[indices] no_match_data = no_match_data[indices] num_validation_samples = int(VALIDATION_SPLIT * data1.shape[0]) x_train = data1[:-num_validation_samples] y_train = data2[:-num_validation_samples] z_train = no_match_data[:-num_validation_samples] x_test = data1[-num_validation_samples:] y_test = data2[-num_validation_samples:] z_test = no_match_data[-num_validation_samples:] texts1 = np.array(texts1) texts2 = np.array(texts2) texts3 = np.array(texts3) texts1 = texts1[indices] texts2 = texts2[indices] texts3 = texts3[indices] print('Preparing embedding matrix.') # prepare embedding matrix # num_words = min(MAX_NB_WORDS, len(word_index)) num_words = len(word_index) + 1 # word_index is indexed from 1-N embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) kz = KazumaCharEmbedding() for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) # i = 0 # while sum(embedding_vector) == 0 and i <= 1000: # embedding_vector = k.emb(word) # i++; # if i == 1000: # print("fail") if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector #else: # print(word + ) # note that we set trainable = False so as to keep the embeddings fixed check_for_zeroes(embedding_matrix, "here is the first pass") embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) print('Training model.') # train a 1D convnet with global maxpooling #sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32') #need two of these #embedded_sequences = embedding_layer(sequence_input) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def eucl_dist_output_shape(shapes): shape1, shape2 = shapes return (shape1[0], 1) def contrastive_loss(y_true, y_pred): '''Contrastive loss from Hadsell-et-al.'06 http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf ''' margin = 1 return K.mean(y_true * K.square(y_pred) + (1 - y_true) * K.square(K.maximum(margin - y_pred, 0))) #for now will take any bad pairs, will take only relivent ones later def create_pairs(x, y, z): '''Positive and negative pair creation. Alternates between positive and negative pairs. ''' pairs = [] labels = [] for index in range(len(x)): pairs += [[x[index], y[index]]] pairs += [[x[index], z[index]]] labels += [1, 0] # n = min([len(digit_indices[d]) for d in range(10)]) - 1 # for d in range(10): # for i in range(n): # z1, z2 = digit_indices[d][i], digit_indices[d][i + 1] # pairs += [[x[z1], x[z2]]] # inc = random.randrange(1, 10) # dn = (d + inc) % 10 # z1, z2 = digit_indices[d][i], digit_indices[dn][i] # pairs += [[x[z1], x[z2]]] # labels += [1, 0] return np.array(pairs), np.array(labels) def create_base_network(input_dim, embedding_layer, reg): '''Base network to be shared (eq. to feature extraction). ''' seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) seq.add(Dense(128, input_shape=(input_dim,), activation='relu')) seq.add(Dropout(0.1)) seq.add(Dense(128, activation='relu')) # kernel_regularizer=reg)) seq.add(Dropout(0.1)) final_layer = Dense(128, activation='relu') seq.add(final_layer) # kernel_regularizer=reg)) return seq, final_layer def compute_accuracy(predictions, labels): '''Compute classification accuracy with a fixed threshold on distances. ''' return np.mean(np.equal(predictions.ravel() < 0.5, labels)) # return labels[predictions.ravel() < 0.5].mean() def f1score(predictions, labels): #labels[predictions.ravel() < 0.5].sum() predictions = predictions.ravel() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(labels)): if predictions[i] < 0.5: if labels[i] == 1: true_positive += 1 else: false_positive += 1 elif labels[i] == 1: false_negitive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) print (fscore) return fscore # the data, shuffled and split between train and test sets #need to change this not sure how input_dim = MAX_SEQUENCE_LENGTH epochs = 1 # create training+test positive and negative pairs # these next lines also need to change #digit_indices = [np.where(y_train == i)[0] for i in range(10)] print("x_train {} , y_train {} , z_train {} ".format(x_train, y_train, z_train)) tr_pairs, tr_y = create_pairs(x_train, y_train, z_train) #digit_indices = [np.where(y_test == i)[0] for i in range(10)] te_pairs, te_y = create_pairs(x_test, y_test, z_test) print (len(tr_y)) # network definition base_network, final_layer = create_base_network(input_dim, embedding_layer, L1L2(0.0,0.0)) input_a = Input(shape=(input_dim,)) input_b = Input(shape=(input_dim,)) # because we re-use the same instance `base_network`, # the weights of the network # will be shared across the two branches processed_a = base_network(input_a) processed_b = base_network(input_b) distance = Lambda(euclidean_distance, output_shape=eucl_dist_output_shape)([processed_a, processed_b]) model = Model([input_a, input_b], distance) print(base_network.summary()) # train rms = RMSprop() #change the optimizer (adam) model.compile(loss=contrastive_loss, optimizer=rms) model.fit([tr_pairs[:, 0], tr_pairs[:, 1]], tr_y, batch_size=128, epochs=epochs, validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y)) # compute final accuracy on training and test sets #add an LSTM layer (later) # testpairs = [[lambda x : x.replace(" ", ""), lambda name1, name2 : name1 in name2 or name2 in name1], # [lambda x : set(x.split()), lambda name1, name2 : name1.issubset(name2) or name2.issubset(name1)]] # matcher = matcher(argv[1], argv[2], argv[3], test_pairs, 1) pred_learning = model.predict([tr_pairs[:, 0], tr_pairs[:, 1]]) # out = model.layers[2].get_output_at(0) # inp = model.input # func = K.function([inp], [out]) # evaluation functions # print("here should be a vector") # print(func([tr_pairs[:, 0][0], tr_pairs[:, 1][0]])) # Testing # print (layer_outs) tr_acc = compute_accuracy(pred_learning, tr_y) tr_f1 = f1score(pred_learning, tr_y) pred = model.predict([te_pairs[:, 0], te_pairs[:, 1]]) pred_learning = np.append(pred_learning, pred, axis=0) te_acc = compute_accuracy(pred, te_y) te_f1 = f1score(pred, te_y) x_test_text = texts1[-num_validation_samples:] y_test_text = texts2[-num_validation_samples:] z_test_text = texts3[-num_validation_samples:] for i in range(len(x_test_text)): print(x_test_text[i] + "|" + y_test_text[i]) print(pred[i]) print(x_test_text[i] + "|" + z_test_text[i]) print(pred[i + 1]) inter_model = Model(input_a, processed_a) intermediate_output1 = inter_model.predict(annoy_data1) intermediate_output2 = inter_model.predict(annoy_data2) mid_predictions = np.concatenate((intermediate_output1, intermediate_output2)) # from https://github.com/spotify/annoy f = 128 # print(mid_predictions[0]) # print (len(mid_predictions[0])) t = AnnoyIndex(f, metric='euclidean') # Length of item vector that will be indexed for i in range(len(mid_predictions)): v = mid_predictions[i] t.add_item(i, v) t.build(100) # 100 trees t.save('test.ann') # ... all_texts = np.concatenate((texts1, texts2)) match = 0 no_match = 0 print("shape of annoy data1") print(annoy_data1[0].shape) print(tr_pairs[:, 0].shape) for index in range(len(texts1)): nearest = t.get_nns_by_vector(mid_predictions[index], 2) print("query={} names = {} true_match = {}".format(texts1[index], [all_texts[i] for i in nearest], texts2[index])) for i in nearest: print(t.get_distance(index, i)) print(model.predict([np.array([annoy_data1[index]]), np.array([annoy_data2[i - len(annoy_data1)]])])) print(t.get_distance(index, index + len(texts1))) print(model.predict([np.array([annoy_data1[index]]), np.array([annoy_data2[index]])])) if (index + len(texts1)) in nearest: match += 1 else: no_match += 1 print("match: {} no_match: {}".format(match, no_match)) print("Machine Learning Accuracy") print(tr_acc) print('* Accuracy on training set: %0.2f%%' % (100 * tr_acc)) print('* Accuracy on test set: %0.2f%%' % (100 * te_acc)) print('* f1score on the training set: %0.4f' % (tr_f1)) print('* f1socre on test set: %0.4f' % (te_f1)) #compute accuracy using a rule based matcher def sequence_to_word(sequence, reverse_word_index): return " ".join([reverse_word_index[x] for x in sequence if x in reverse_word_index]) def sequence_pair_to_word_pair(sequence_pair, reverse_word_index): return [sequence_to_word(sequence_pair[0], reverse_word_index), sequence_to_word(sequence_pair[1], reverse_word_index)] reverse_word_index = {v: k for k, v in tokenizer.word_index.items()} print(tr_pairs) print(sequence_to_word(tr_pairs[0][0], reverse_word_index)) print(sequence_to_word(tr_pairs[1][1], reverse_word_index)) print(tr_y[0]) test_pairs = [[lambda x : x.replace(" ", ""), lambda name1, name2 : name1 in name2 or name2 in name1], [lambda x : set(x.split()), lambda name1, name2 : name1.issubset(name2) or name2.issubset(name1)]] matcher = matcher(argv[1], argv[2], argv[3], test_pairs, 1) pred_rules = np.asarray([int(not matcher.match(*sequence_pair_to_word_pair(name_pair, reverse_word_index))) for name_pair in tr_pairs]) tr_acc = compute_accuracy(pred_rules, tr_y) tr_f1 = f1score(pred_rules, tr_y) pred = np.asarray([int(not matcher.match(*sequence_pair_to_word_pair(name_pair, reverse_word_index))) for name_pair in te_pairs]) pred_rules = np.append(pred_rules, pred, axis=0) te_acc = compute_accuracy(pred, te_y) te_f1 = f1score(pred, te_y) print("Rule-Based Accuracy") print('* Accuracy on training set (rules): %0.2f%%' % (100 * tr_acc)) print('* Accuracy on test set (rules): %0.2f%%' % (100 * te_acc)) print('* f1score on the training set: %0.4f' % (tr_f1)) print('* f1socre on test set: %0.4f' % (te_f1)) con, meta = connect(argv[1], argv[2], argv[3]) execute_pairs = [] if 'predictions' in meta.tables: meta.tables['predictions'].drop(con) predictions = Table('predictions', meta, Column('name1', String), Column('name2', String), Column('rule_predict', Integer), Column('learning_predict', Float), Column('true_pair', Integer), Column('te_or_tr', String), extend_existing=True) zipping_string = ('name1', 'name2', 'true_pair', 'rule_predict', 'learning_predict', 'te_or_tr') print(len(tr_y)) print(len(tr_pairs)) print(len(pred_rules)) print(len(pred_learning)) print(len(te_y)) print(len(te_pairs)) for i in range(len(tr_y)): execute_pairs.append(dict(zip(zipping_string, (sequence_to_word(tr_pairs[i][0], reverse_word_index), sequence_to_word(tr_pairs[i][1], reverse_word_index), int(tr_y[i]), int(pred_rules[i]), float(pred_learning[i][0].item()), 'tr')))) offset = len(tr_y) for i in range(len(te_y)): execute_pairs.append(dict(zip(zipping_string, (sequence_to_word(tr_pairs[i][0], reverse_word_index), sequence_to_word(te_pairs[i][1], reverse_word_index), int(te_y[i]), int(pred_rules[offset + i]), float(pred_learning[offset + i][0].item()), 'te')))) meta.create_all(con) con.execute(predictions.insert(), execute_pairs)
19,852
31.176661
254
py
fuzzyJoiner
fuzzyJoiner-master/old/face_vgg.py
0
0
0
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-modified.py
import numpy as np import tensorflow as tf import random as random # import cntk as C # """ # The below is necessary in Python 3.2.3 onwards to # have reproducible behavior for certain hash-based operations. # See these references for further details: # https://docs.python.org/3.4/using/cmdline.html#envvar-PYTHONHASHSEED # https://github.com/keras-team/keras/issues/2280#issuecomment-306959926 import os os.environ['PYTHONHASHSEED'] = '0' # The below is necessary for starting Numpy generated random numbers # in a well-defined initial state. np.random.seed(42) # The below is necessary for starting core Python generated random numbers # in a well-defined state. random.seed(12345) # Force TensorFlow to use single thread. # Multiple threads are a potential source of # non-reproducible results. # For further details, see: https://stackoverflow.com/questions/42022950/which-seeds-have-to-be-set-where-to-realize-100-reproducibility-of-training-res session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1) # import theano as T from keras import backend as K # The below tf.set_random_seed() will make random number generation # in the TensorFlow backend have a well-defined initial state. # For further details, see: https://www.tensorflow.org/api_docs/python/tf/set_random_seed tf.set_random_seed(1234) sess = tf.Session(graph=tf.get_default_graph(), config=session_conf) K.set_session(sess) # from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=1 ALPHA=30 USE_GRU=True DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = True def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(T.tensor.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): # margin = K.constant(MARGIN) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0]) + K.square(margin - y_pred[:,2,0])) # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(MARGIN) lambda_p = 0.02 threshold = 0.1 a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / 2) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def tanhNorm(x): square_sum = K.sum(K.square(x), axis=-1, keepdims=True) dist = K.sqrt(K.maximum(square_sum, K.epsilon())) tanh = K.tanh(dist) scale = tanh / dist return x * scale def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = expected_text negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(accuracy / total)) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, limit_pairs=False): num_names = 4 names_generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) entity2same = {} for entity in entities: ret = names_generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def get_hidden_layer(name, net, is_last): if USE_GRU: if is_last: return GRU(128, activation='relu', name=name)(net) else: return GRU(128, return_sequences=True, activation='relu', name=name)(net) else: return Dense(128, activation='relu', name=name)(net) def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) net = GRU(128, return_sequences=True, activation='relu', name='embed')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2a')(net) net = GRU(128, activation='relu', name='embed3')(net) """ for i in range(0, NUM_LAYERS): net = get_hidden_layer('embed' + str(i), net, False) net = get_hidden_layer('embed_last', net, True) """ if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=improved_loss, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='loss function type: triplet-loss, improved-triplet-loss, modified-loss, or angular-loss') parser.add_argument('--use_l2_norm', type=bool, help='whether to add a l2 norm') parser.add_argument('--test_neighbor_len', type=int, help='size of the neighborhood for testing') parser.add_argument('--train_neighbor_len', type=int, help='size of the neighborhood for training') parser.add_argument('--embedding_type', type=str, help='encoding type to use for input: Kazuma (for Kazuma character embedding) or one-hot') parser.add_argument('--use_GRU', type=bool, help='use GRU or default to MLP') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') args = parser.parse_args() """ LOSS_FUNCTION = None if args.loss_function == 'triplet-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved_triplet_loss': LOSS_FUNCTION=improved_triplet_loss elif args.loss_function == 'modified_loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = true LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) MARGIN = args.margin print('Margin:' + str(MARGIN)) TRAIN_NEIGHBOR_LEN = args.train_neighbor_len TEST_NEIGHBOR_LEN = args.test_neighbor_len print('Train neighbor length: ' + str(TRAIN_NEIGHBOR_LEN)) print('Test neighbor length: ' + str(TEST_NEIGHBOR_LEN)) USE_L2_NORM = args.use_l2_norm print('Use L2Norm: ' + str(USE_L2_NORM)) EMBEDDING_TYPE = args.embedding_type print('Embedding type: ' + EMBEDDING_TYPE) USE_GRU = args.use_GRU print('Use GRU: ' + str(args.use_GRU)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) """ # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train) entity2same_test = generate_names(test, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) match_stats = 0 # num_iter = 100 num_iter = 1 counter = 0 current_model = embedder_model prev_match_stats = 0 while test_match_stats < .9 and counter < num_iter: counter += 1 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
21,841
37.319298
167
py
fuzzyJoiner
fuzzyJoiner-master/old/ANNCharacteristics.py
import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def characterize_dataset(model, sequences, entity2unique, entity2same, unique_text, nnlens): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees for nnlen in nnlens: print("Characteristics at neighborhood length:" + str(nnlen)) pos_distances = [] neg_distances = [] match = 0 no_match = 0 for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], nnlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) overlap = expected_text.intersection(nearest_text) m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), nnlen - 1) - m # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' pos = expected_text neg = nearest_text - expected_text for i in pos: dist_pos = t.get_distance(index, entity2unique[i]) pos_distances.append(dist_pos) for i in neg: dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) recall = match / (match + no_match) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("recall:" + str(recall)) def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Characterize the dataset') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') args = parser.parse_args() LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) entity2same = generate_names(entities, people) print("number of entities") print(len(entity2same)) number_of_names = [] for i in entity2same: number_of_names.append(len(entity2same[i])) print("mean number of names:" + str(statistics.mean(number_of_names))) print("max number of names:" + str(max(number_of_names))) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same) tokenizer.fit_on_texts(unique_text) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) embedder_model = embedded_representation_model(embedder) characterize_dataset(embedder_model, sequences, entity2unique, entity2same, unique_text, [20, 100, 500, 1500])
19,558
35.355019
134
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenet.py
from sys import argv from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, LSTM from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential import numpy as np from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex import Named_Entity_Recognition_Modified #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 DEBUG = False DEBUG_DATA_LENGTH = 1000000 def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_random_image(img_groups, group_names, gid): gname = group_names[gid] photos = img_groups[gname] pid = np.random.choice(np.arange(len(photos)), size=1)[0] pname = photos[pid] return gname + pname + ".jpg" def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) # i = 0 # while sum(embedding_vector) == 0 and i <= 1000: # embedding_vector = k.emb(word) # i++; # if i == 1000: # print("fail") if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_tokenizer(texts): tokenizer = Tokenizer(num_words=MAX_NB_WORDS) tokenizer.fit_on_texts(texts['anchor'] + texts['negative'] + texts['positive']) return tokenizer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) #removes the new line charecter at the end texts['negative'].append(line_array[2][:-1]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def get_test(texts, sequences, percent): indices = np.arange(sequences['anchor'].shape[0]) np.random.shuffle(indices) ret_sequence = {} ret_sequence['anchor'] = sequences['anchor'][indices] ret_sequence['positive'] = sequences['positive'][indices] ret_sequence['negative'] = sequences['negative'][indices] num_validation_samples = int(percent * sequences['anchor'].shape[0]) ret_train = {} ret_train['anchor'] = ret_sequence['anchor'][:-num_validation_samples] ret_train['positive'] = ret_sequence['positive'][:-num_validation_samples] ret_train['negative'] = ret_sequence['negative'][:-num_validation_samples] ret_test = {} ret_test['anchor']= ret_sequence['anchor'][-num_validation_samples:] ret_test['positive']= ret_sequence['positive'][-num_validation_samples:] ret_test['negative'] = ret_sequence['negative'][-num_validation_samples:] ret_texts = {} texts['anchor'] = np.array(texts['anchor']) texts['positive'] = np.array(texts['positive']) texts['negative'] = np.array(texts['negative']) ret_texts['anchor'] = texts['anchor'][indices] ret_texts['positive'] = texts['positive'][indices] ret_texts['negative'] = texts['negative'][indices] return ret_train, ret_test, ret_texts def triplet_loss(y_true, y_pred): margin = K.constant(1) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def do_annoy(model, texts, tokenizer, verbose): unique_text = [] entity_idx = [] entity2same = {} for i in range(len(texts['anchor'])): if not texts['anchor'][i] in entity2same: entity2same[texts['anchor'][i]] = [] entity_idx.append(len(unique_text)) unique_text.append(texts['anchor'][i]) l = entity2same[texts['anchor'][i]] if texts['positive'][i] not in l: entity2same[texts['anchor'][i]].append(texts['positive'][i]) unique_text.append(texts['positive'][i]) print(entity2same) print(unique_text) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed for i in range(len(predictions)): v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 for index in entity_idx: nearest = t.get_nns_by_vector(predictions[index], 5) print(nearest) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[unique_text[index]]) if unique_text[index] in nearest_text: nearest_text.remove(unique_text[index]) print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) if verbose: print([t.get_distance(index, i) for i in nearest]) overlap = expected_text.intersection(nearest_text) print(overlap) m = len(overlap) match += m no_match += len(expected_text) - m print("match: {} no_match: {}".format(match, no_match)) def print_deb_data(debbuging_data): for i in range(debbuging_data['number']): print('anch: --{}-- pos:--{}-- neg:--{}--'.format(debbuging_data['texts']['anchor'][i], debbuging_data['texts']['positive'][i], debbuging_data['texts']['negative'][i])) print('sequences: anch: --{}-- pos:--{}-- neg:--{}--'.format(debbuging_data['sequences']['anchor'][i], debbuging_data['sequences']['positive'][i], debbuging_data['sequences']['negative'][i])) def debugging_text_and_sequences(reordered_text, training_data, number): debbuging_data = {} debbuging_data['number'] = number debbuging_data['sequences'] = {} debbuging_data['texts'] = {} debbuging_data['sequences']['anchor'] = [] debbuging_data['sequences']['positive'] = [] debbuging_data['sequences']['negative'] = [] debbuging_data['texts']['anchor'] = [] debbuging_data['texts']['positive'] = [] debbuging_data['texts']['negative'] = [] for i in range(number): debbuging_data['texts']['anchor'].append(reordered_text['anchor'][i]) debbuging_data['texts']['positive'].append(reordered_text['positive'][i]) debbuging_data['texts']['negative'].append(reordered_text['negative'][i]) debbuging_data['sequences']['anchor'].append(training_data['anchor'][i]) debbuging_data['sequences']['positive'].append(training_data['positive'][i]) debbuging_data['sequences']['negative'].append(training_data['negative'][i]) return debbuging_data # triples_data = create_triples(IMAGE_DIR) texts = read_file(argv[1]) print("anchor: {} positive: {} negative: {}".format(texts['anchor'][0], texts['positive'][0], texts['negative'][0])) tokenizer = get_tokenizer(texts) print('got tokenizer') sequences = get_sequences(texts, tokenizer) train_data, test_data, reordered_text = get_test(texts, sequences, 0.05) debbuging_data = debugging_text_and_sequences(reordered_text, train_data, 20) number_of_names = len(train_data['anchor']) print('sequenced words') Y_train = np.random.randint(2, size=(1,2,number_of_names)).T embedder = get_embedding_layer(tokenizer) print('got embeddings') main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) net = Flatten(name='flatten')(net) net = Dense(128, activation='relu', name='embed')(net) net = Dense(128, activation='relu', name='embed2')(net) net = Dense(128, activation='relu', name='embed3')(net) net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') print(base_model.summary()) input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist')([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist')([net_anchor, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists' )([positive_dist, negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=triplet_loss, metrics=[accuracy]) model.fit([train_data['anchor'], train_data['positive'], train_data['negative']], Y_train, epochs=10, batch_size=15, validation_split=0.2) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) print("training data predictions") positives = test_positive_model.predict([train_data['anchor'], train_data['positive'], train_data['negative']]) negatives = test_negative_model.predict([train_data['anchor'], train_data['positive'], train_data['negative']]) print("f1score is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_data['anchor'], test_data['positive'], test_data['negative']]) negatives = test_negative_model.predict([test_data['anchor'], test_data['positive'], test_data['negative']]) print("f1score is: {}".format(f1score(positives, negatives))) # model.save('triplet_loss_resnet50.h5') inter_model = Model(input_anchor, net_anchor) do_annoy(inter_model, texts, tokenizer, False) print('annoy on embeddings for debbuging_data') do_annoy(Named_Entity_Recognition_Modified.embedded_representation(embedder), debbuging_data['texts'], tokenizer, True) print('annoy on full model for debbuging_data') do_annoy(inter_model, debbuging_data['texts'], tokenizer, True) print_deb_data(debbuging_data)
12,095
37.893891
199
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-8.31.18.py
from random import shuffle import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics from scipy.stats.mstats import gmean import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False filepath="weights.best.hdf5" output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print(str("failed to find embedding for:" + word).encode('utf-8')) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath, 'r', encoding='utf8') as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r', encoding='utf8') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_semi_hard_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees triplets = {} triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] expected_text = set(entity2same[key]) expected_ids = [entity2unique[i] for i in expected_text] for positive in expected_text: k = entity2unique[positive] nearest = t.get_nns_by_vector(predictions[k], NNlen) dist_k = t.get_distance(index, k) semi_hards = [] for n in nearest: if n == index or n in expected_ids or n == k: continue n_dist = t.get_distance(index, n) if n_dist > dist_k: semi_hards.append(unique_text[n]) # shuffle(semi_hards) # semi_hards = semi_hards[0:20] for i in semi_hards: triplets['anchor'].append(key) triplets['positive'].append(unique_text[k]) triplets['negative'].append(i) return triplets def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = t.get_distance(index, entity2unique[i]) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = t.get_distance(index, entity2unique[j]) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = t.get_distance(index, entity2unique[p]) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 closest_positive_counts.append(closest_pos_count / min(len(expected_text), NNlen - 1)) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) print("Precision at 1: " + str(precise / len(entity2same))) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w', encoding='utf8') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') parser.add_argument('--model', type=str, help='name for model file') args = parser.parse_args() filepath = args.model LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(str(train).encode('utf-8')) print("TEST") print(str(test).encode('utf-8')) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) print(str(entity2same_train).encode('utf-8')) print(str(entity2same_test).encode('utf-8')) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: triplets = generate_triplets_from_ANN(embedder_model, sequences, entity2unique, entity2same_train, unique_text, True) print(len(triplets['anchor'])) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, train_match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
22,999
35.624204
163
py
fuzzyJoiner
fuzzyJoiner-master/old/image_join2.py
0
0
0
py
fuzzyJoiner
fuzzyJoiner-master/old/seq2seqTriplet.py
'''Sequence to sequence example in Keras (character-level). This script demonstrates how to implement a basic character-level sequence-to-sequence model. We apply it to translating short English sentences into short French sentences, character-by-character. Note that it is fairly unusual to do character-level machine translation, as word-level models are more common in this domain. # Summary of the algorithm - We start with input sequences from a domain (e.g. English sentences) and correspding target sequences from another domain (e.g. French sentences). - An encoder LSTM turns input sequences to 2 state vectors (we keep the last LSTM state and discard the outputs). - A decoder LSTM is trained to turn the target sequences into the same sequence but offset by one timestep in the future, a training process called "teacher forcing" in this context. Is uses as initial state the state vectors from the encoder. Effectively, the decoder learns to generate `targets[t+1...]` given `targets[...t]`, conditioned on the input sequence. - In inference mode, when we want to decode unknown input sequences, we: - Encode the input sequence into state vectors - Start with a target sequence of size 1 (just the start-of-sequence character) - Feed the state vectors and 1-char target sequence to the decoder to produce predictions for the next character - Sample the next character using these predictions (we simply use argmax). - Append the sampled character to the target sequence - Repeat until we generate the end-of-sequence character or we hit the character limit. # Data download English to French sentence pairs. http://www.manythings.org/anki/fra-eng.zip Lots of neat sentence pairs datasets can be found at: http://www.manythings.org/anki/ # References - Sequence to Sequence Learning with Neural Networks https://arxiv.org/abs/1409.3215 - Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation https://arxiv.org/abs/1406.1078 ''' from __future__ import print_function from keras.models import Model from keras.layers import Input, LSTM, Dense import numpy as np from keras.callbacks import ModelCheckpoint, EarlyStopping batch_size = 64 # Batch size for training. epochs = 100 # Number of epochs to train for. latent_dim = 256 # Latent dimensionality of the encoding space. num_samples = 100000 # Number of samples to train on. # Path to the data txt file on disk. data_path = 'tripletTranslate' # Vectorize the data. input_texts = [] target_texts = [] input_characters = set() target_characters = set() with open(data_path, 'r', encoding='utf-8') as f: lines = f.read().split('\n') for line in lines[: min(num_samples, len(lines) - 1)]: arr = line.split('|') input_text = arr[0] target_text = arr[1] arr = input_text.split() # data quality issues. In the canonical name, the first name had better not be just a letter or empty if arr[0].strip()=='' or len(arr[0].strip()) <= 1: continue # We use "tab" as the "start sequence" character # for the targets, and "\n" as "end sequence" character. target_text = '\t' + target_text + '\n' input_texts.append(input_text) target_texts.append(target_text) for char in input_text: if char not in input_characters: input_characters.add(char) for char in target_text: if char not in target_characters: target_characters.add(char) input_characters = sorted(list(input_characters)) target_characters = sorted(list(target_characters)) num_encoder_tokens = len(input_characters) num_decoder_tokens = len(target_characters) max_encoder_seq_length = max([len(txt) for txt in input_texts]) max_decoder_seq_length = max([len(txt) for txt in target_texts]) print('Number of samples:', len(input_texts)) print('Number of unique input tokens:', num_encoder_tokens) print('Number of unique output tokens:', num_decoder_tokens) print('Max sequence length for inputs:', max_encoder_seq_length) print('Max sequence length for outputs:', max_decoder_seq_length) input_token_index = dict( [(char, i) for i, char in enumerate(input_characters)]) target_token_index = dict( [(char, i) for i, char in enumerate(target_characters)]) encoder_input_data = np.zeros( (len(input_texts), max_encoder_seq_length, num_encoder_tokens), dtype='float32') decoder_input_data = np.zeros( (len(input_texts), max_decoder_seq_length, num_decoder_tokens), dtype='float32') decoder_target_data = np.zeros( (len(input_texts), max_decoder_seq_length, num_decoder_tokens), dtype='float32') for i, (input_text, target_text) in enumerate(zip(input_texts, target_texts)): for t, char in enumerate(input_text): encoder_input_data[i, t, input_token_index[char]] = 1. for t, char in enumerate(target_text): # decoder_target_data is ahead of decoder_input_data by one timestep decoder_input_data[i, t, target_token_index[char]] = 1. if t > 0: # decoder_target_data will be ahead by one timestep # and will not include the start character. decoder_target_data[i, t - 1, target_token_index[char]] = 1. # Define an input sequence and process it. encoder_inputs = Input(shape=(None, num_encoder_tokens)) encoder = LSTM(latent_dim, return_state=True) encoder_outputs, state_h, state_c = encoder(encoder_inputs) # We discard `encoder_outputs` and only keep the states. encoder_states = [state_h, state_c] # Set up the decoder, using `encoder_states` as initial state. decoder_inputs = Input(shape=(None, num_decoder_tokens)) # We set up our decoder to return full output sequences, # and to return internal states as well. We don't use the # return states in the training model, but we will use them in inference. decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True) decoder_outputs, _, _ = decoder_lstm(decoder_inputs, initial_state=encoder_states) decoder_dense = Dense(num_decoder_tokens, activation='softmax') decoder_outputs = decoder_dense(decoder_outputs) # Define the model that will turn # `encoder_input_data` & `decoder_input_data` into `decoder_target_data` model = Model([encoder_inputs, decoder_inputs], decoder_outputs) filepath = 's2s.h5' checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') # check 5 epochs early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] # Run training model.compile(optimizer='rmsprop', loss='categorical_crossentropy') model.fit([encoder_input_data, decoder_input_data], decoder_target_data, batch_size=batch_size, epochs=epochs, callbacks=callbacks_list, validation_split=0.2) # Next: inference mode (sampling). # Here's the drill: # 1) encode input and retrieve initial decoder state # 2) run one step of decoder with this initial state # and a "start of sequence" token as target. # Output will be the next target token # 3) Repeat with the current target token and current states # Define sampling models encoder_model = Model(encoder_inputs, encoder_states) decoder_state_input_h = Input(shape=(latent_dim,)) decoder_state_input_c = Input(shape=(latent_dim,)) decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c] decoder_outputs, state_h, state_c = decoder_lstm( decoder_inputs, initial_state=decoder_states_inputs) decoder_states = [state_h, state_c] decoder_outputs = decoder_dense(decoder_outputs) decoder_model = Model( [decoder_inputs] + decoder_states_inputs, [decoder_outputs] + decoder_states) # Reverse-lookup token index to decode sequences back to # something readable. reverse_input_char_index = dict( (i, char) for char, i in input_token_index.items()) reverse_target_char_index = dict( (i, char) for char, i in target_token_index.items()) def decode_sequence(input_seq): # Encode the input as state vectors. states_value = encoder_model.predict(input_seq) # Generate empty target sequence of length 1. target_seq = np.zeros((1, 1, num_decoder_tokens)) # Populate the first character of target sequence with the start character. target_seq[0, 0, target_token_index['\t']] = 1. # Sampling loop for a batch of sequences # (to simplify, here we assume a batch of size 1). stop_condition = False decoded_sentence = '' while not stop_condition: output_tokens, h, c = decoder_model.predict( [target_seq] + states_value) # Sample a token sampled_token_index = np.argmax(output_tokens[0, -1, :]) sampled_char = reverse_target_char_index[sampled_token_index] decoded_sentence += sampled_char # Exit condition: either hit max length # or find stop character. if (sampled_char == '\n' or len(decoded_sentence) > max_decoder_seq_length): stop_condition = True # Update the target sequence (of length 1). target_seq = np.zeros((1, 1, num_decoder_tokens)) target_seq[0, 0, sampled_token_index] = 1. # Update states states_value = [h, c] return decoded_sentence for seq_index in range(100): # Take one sequence (part of the training set) # for trying out decoding. input_seq = encoder_input_data[seq_index: seq_index + 1] decoded_sentence = decode_sequence(input_seq) print('-') print('Input sentence:', input_texts[seq_index]) print('Decoded sentence:', decoded_sentence)
9,670
39.634454
106
py
fuzzyJoiner
fuzzyJoiner-master/old/image_join.py
0
0
0
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM_hpo.py
import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = expected_text negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(accuracy / total)) obj = {} obj['accuracy'] = accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') args = parser.parse_args() LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
19,474
35.88447
163
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-schroffloss.py
import numpy as np import tensorflow as tf import random as random # import cntk as C # """ # The below is necessary in Python 3.2.3 onwards to # have reproducible behavior for certain hash-based operations. # See these references for further details: # https://docs.python.org/3.4/using/cmdline.html#envvar-PYTHONHASHSEED # https://github.com/keras-team/keras/issues/2280#issuecomment-306959926 import os os.environ['PYTHONHASHSEED'] = '0' # The below is necessary for starting Numpy generated random numbers # in a well-defined initial state. np.random.seed(42) # The below is necessary for starting core Python generated random numbers # in a well-defined state. random.seed(12345) # Force TensorFlow to use single thread. # Multiple threads are a potential source of # non-reproducible results. # For further details, see: https://stackoverflow.com/questions/42022950/which-seeds-have-to-be-set-where-to-realize-100-reproducibility-of-training-res session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1) # import theano as T from keras import backend as K # The below tf.set_random_seed() will make random number generation # in the TensorFlow backend have a well-defined initial state. # For further details, see: https://www.tensorflow.org/api_docs/python/tf/set_random_seed tf.set_random_seed(1234) sess = tf.Session(graph=tf.get_default_graph(), config=session_conf) K.set_session(sess) # from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser import sys import statistics import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=1 ALPHA=30 USE_GRU=True DEBUG = False DEBUG_DATA_LENGTH = 20000 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = True def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(T.tensor.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): # margin = K.constant(MARGIN) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0])) # return K.mean(K.square(y_pred[:,0,0]) + K.square(margin - y_pred[:,1,0]) + K.square(margin - y_pred[:,2,0])) # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(MARGIN) lambda_p = 0.02 threshold = 0.1 a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / 2) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def tanhNorm(x): square_sum = K.sum(K.square(x), axis=-1, keepdims=True) dist = K.sqrt(K.maximum(square_sum, K.epsilon())) tanh = K.tanh(dist) scale = tanh / dist return x * scale def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 accuracy = 0 total = 0 triplets = {} pos_distances = [] neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = expected_text negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(accuracy / total)) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, limit_pairs=False): num_names = 4 names_generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) entity2same = {} for entity in entities: ret = names_generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def get_hidden_layer(name, net, is_last): if USE_GRU: if is_last: return GRU(128, activation='relu', name=name)(net) else: return GRU(128, return_sequences=True, activation='relu', name=name)(net) else: return Dense(128, activation='relu', name=name)(net) def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) net = GRU(128, return_sequences=True, activation='relu', name='embed')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2')(net) net = GRU(128, return_sequences=True, activation='relu', name='embed2a')(net) net = GRU(128, activation='relu', name='embed3')(net) """ for i in range(0, NUM_LAYERS): net = get_hidden_layer('embed' + str(i), net, False) net = get_hidden_layer('embed_last', net, True) """ # if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=schroff_triplet_loss, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='loss function type: triplet-loss, improved-triplet-loss, modified-loss, or angular-loss') parser.add_argument('--use_l2_norm', type=bool, help='whether to add a l2 norm') parser.add_argument('--test_neighbor_len', type=int, help='size of the neighborhood for testing') parser.add_argument('--train_neighbor_len', type=int, help='size of the neighborhood for training') parser.add_argument('--embedding_type', type=str, help='encoding type to use for input: Kazuma (for Kazuma character embedding) or one-hot') parser.add_argument('--use_GRU', type=bool, help='use GRU or default to MLP') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') args = parser.parse_args() """ LOSS_FUNCTION = None if args.loss_function == 'triplet-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved_triplet_loss': LOSS_FUNCTION=improved_triplet_loss elif args.loss_function == 'modified_loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = true LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) MARGIN = args.margin print('Margin:' + str(MARGIN)) TRAIN_NEIGHBOR_LEN = args.train_neighbor_len TEST_NEIGHBOR_LEN = args.test_neighbor_len print('Train neighbor length: ' + str(TRAIN_NEIGHBOR_LEN)) print('Test neighbor length: ' + str(TEST_NEIGHBOR_LEN)) USE_L2_NORM = args.use_l2_norm print('Use L2Norm: ' + str(USE_L2_NORM)) EMBEDDING_TYPE = args.embedding_type print('Embedding type: ' + EMBEDDING_TYPE) USE_GRU = args.use_GRU print('Use GRU: ' + str(args.use_GRU)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) """ # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train) entity2same_test = generate_names(test, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) match_stats = 0 # num_iter = 100 num_iter = 1 counter = 0 current_model = embedder_model prev_match_stats = 0 while test_match_stats < .9 and counter < num_iter: counter += 1 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T filepath="weights.best.hdf5" checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
21,846
37.463028
167
py
fuzzyJoiner
fuzzyJoiner-master/old/file_parser.py
from sys import argv onlyfiles = [f for f in listdir(args[1]) if isfile(join(args[1], f))] for file_path in onlyfiles: input_file = open(args[1] + "/" + file_path, encoding='utf-8') output_file = open(argv[2] + "/" + file_path, "w", encoding='utf-8') write_line = False for line in input_file: if write_line: items = line.split(" ") output_file.write(items[2] + " | " + items[20] + "\n") else: write_line = True f.close() output_file.close()
460
31.928571
69
py
fuzzyJoiner
fuzzyJoiner-master/old/TripletLossFacenetLSTM-8.29.18.py
import numpy as np import pandas import tensorflow as tf import random as random import json from keras import backend as K from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences from keras.layers import Dense, Input, Flatten, Dropout, Lambda, GRU, Activation from keras.layers.wrappers import Bidirectional from keras.layers import Conv1D, MaxPooling1D, Embedding from keras.models import Model, model_from_json, Sequential from embeddings import KazumaCharEmbedding from annoy import AnnoyIndex from keras.callbacks import ModelCheckpoint, EarlyStopping from names_cleanser import NameDataCleanser, CompanyDataCleanser import sys import statistics from scipy.stats.mstats import gmean import argparse #must fix MAX_NB_WORDS = 140000 EMBEDDING_DIM = 100 MAX_SEQUENCE_LENGTH = 10 MARGIN=10 ALPHA=45 DEBUG = False DEBUG_DATA_LENGTH = 100 DEBUG_ANN = False USE_ANGULAR_LOSS=False LOSS_FUNCTION=None TRAIN_NEIGHBOR_LEN=20 TEST_NEIGHBOR_LEN=20 EMBEDDING_TYPE = 'Kazuma' NUM_LAYERS = 3 USE_L2_NORM = False filepath="weights.best.hdf5" output_file_name_for_hpo = "val_dict_list.json" def f1score(positive, negative): #labels[predictions.ravel() < 0.5].sum() fsocre = 0.0 true_positive = 0.0 false_positive = 0 false_negitive = 0 for i in range(len(positive)): if positive[i] <= negative[i]: true_positive += 1 else: false_negitive += 1 false_positive += 1 print('tp' + str(true_positive)) print('fp' + str(false_positive)) print('fn' + str(false_negitive)) fscore = (2 * true_positive) / ((2 * true_positive) + false_negitive + false_positive) return fscore def get_embedding_layer(tokenizer): word_index = tokenizer.word_index num_words = len(word_index) + 1 embedding_matrix = np.zeros((num_words, EMBEDDING_DIM)) print('about to get kz') kz = KazumaCharEmbedding() print('got kz') for word, i in word_index.items(): if i >= MAX_NB_WORDS: continue embedding_vector = kz.emb(word) if embedding_vector is not None: if sum(embedding_vector) == 0: print("failed to find embedding for:" + word) # words not found in embedding index will be all-zeros. embedding_matrix[i] = embedding_vector print("Number of words:" + str(num_words)) embedding_layer = Embedding(num_words, EMBEDDING_DIM, weights=[embedding_matrix], input_length=MAX_SEQUENCE_LENGTH, trainable=False) return embedding_layer def get_sequences(texts, tokenizer): sequences = {} sequences['anchor'] = tokenizer.texts_to_sequences(texts['anchor']) sequences['anchor'] = pad_sequences(sequences['anchor'], maxlen=MAX_SEQUENCE_LENGTH) sequences['negative'] = tokenizer.texts_to_sequences(texts['negative']) sequences['negative'] = pad_sequences(sequences['negative'], maxlen=MAX_SEQUENCE_LENGTH) sequences['positive'] = tokenizer.texts_to_sequences(texts['positive']) sequences['positive'] = pad_sequences(sequences['positive'], maxlen=MAX_SEQUENCE_LENGTH) return sequences def read_entities(filepath): entities = [] with open(filepath) as fl: for line in fl: entities.append(line) return entities def read_file(file_path): texts = {'anchor':[], 'negative':[], 'positive':[]} fl = open(file_path, 'r') i = 0 for line in fl: line_array = line.split("|") texts['anchor'].append(line_array[0]) texts['positive'].append(line_array[1]) texts['negative'].append(line_array[2]) i += 1 if i > DEBUG_DATA_LENGTH and DEBUG: break return texts def split(entities, test_split = 0.2): if DEBUG: ents = entities[0:DEBUG_DATA_LENGTH] else: random.shuffle(entities) ents = entities num_validation_samples = int(test_split * len(ents)) return ents[:-num_validation_samples], ents[-num_validation_samples:] """ define a single objective function based on angular loss instead of triplet loss """ def angular_loss(y_true, y_pred): alpha = K.constant(ALPHA) a_p = y_pred[:,0,0] n_c = y_pred[:,1,0] return K.mean(K.maximum(K.constant(0), K.square(a_p) - K.constant(4) * K.square(tf.tan(alpha)) * K.square(n_c))) """ Facenet triplet loss function: https://arxiv.org/pdf/1503.03832.pdf """ def schroff_triplet_loss(y_true, y_pred): margin = K.constant(0.2) return K.mean(K.maximum(K.constant(0), K.square(y_pred[:,0,0]) - K.square(y_pred[:,1,0]) + margin)) def triplet_loss(y_true, y_pred): margin = K.constant(MARGIN) return K.mean(K.square(y_pred[:,0,0]) + K.square(K.maximum(margin - y_pred[:,1,0], K.constant(0)))) def triplet_tanh_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + (K.constant(1) - K.tanh(y_pred[:,1,0]))) def triplet_tanh_pn_loss(y_true, y_pred): return K.mean(K.tanh(y_pred[:,0,0]) + ((K.constant(1) - K.tanh(y_pred[:,1,0])) + (K.constant(1) - K.tanh(y_pred[:,2,0]))) / K.constant(2)); # the following triplet loss function is from: Deep Metric Learning with Improved Triplet Loss for # Face clustering in Videos def improved_loss(y_true, y_pred): margin = K.constant(1) lambda_p = K.constant(0.02) threshold = K.constant(0.1) a_p_distance = y_pred[:,0,0] a_n_distance = y_pred[:,1,0] p_n_distance = y_pred[:,2,0] phi = a_p_distance - ((a_n_distance + p_n_distance) / K.constant(2)) + margin psi = a_p_distance - threshold return K.maximum(K.constant(0), phi) + lambda_p * K.maximum(K.constant(0), psi) def accuracy(y_true, y_pred): return K.mean(y_pred[:,0,0] < y_pred[:,1,0]) def l2Norm(x): return K.l2_normalize(x, axis=-1) def euclidean_distance(vects): x, y = vects return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon())) def n_c_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_n - ((x_a + x_p) / K.constant(2))), axis=1, keepdims=True), K.epsilon())) def a_p_angular_distance(vects): x_a, x_p, x_n = vects return K.sqrt(K.maximum(K.sum(K.square(x_a - x_p), axis=1, keepdims=True), K.epsilon())) def build_unique_entities(entity2same): unique_text = [] entity2index = {} for key in entity2same: entity2index[key] = len(unique_text) unique_text.append(key) vals = entity2same[key] for v in vals: entity2index[v] = len(unique_text) unique_text.append(v) return unique_text, entity2index def generate_triplets_from_ANN(model, sequences, entity2unique, entity2same, unique_text, test): predictions = model.predict(sequences) t = AnnoyIndex(len(predictions[0]), metric='euclidean') # Length of item vector that will be indexed t.set_seed(123) for i in range(len(predictions)): # print(predictions[i]) v = predictions[i] t.add_item(i, v) t.build(100) # 100 trees match = 0 no_match = 0 ann_accuracy = 0 total = 0 precise = 0 triplets = {} closest_positive_counts = [] pos_distances = [] neg_distances = [] all_pos_distances = [] all_neg_distances = [] triplets['anchor'] = [] triplets['positive'] = [] triplets['negative'] = [] if test: NNlen = TEST_NEIGHBOR_LEN else: NNlen = TRAIN_NEIGHBOR_LEN for key in entity2same: index = entity2unique[key] nearest = t.get_nns_by_vector(predictions[index], NNlen) nearest_text = set([unique_text[i] for i in nearest]) expected_text = set(entity2same[key]) # annoy has this annoying habit of returning the queried item back as a nearest neighbor. Remove it. if key in nearest_text: nearest_text.remove(key) # print("query={} names = {} true_match = {}".format(unique_text[index], nearest_text, expected_text)) overlap = expected_text.intersection(nearest_text) # collect up some statistics on how well we did on the match m = len(overlap) match += m # since we asked for only x nearest neighbors, and we get at most x-1 neighbors that are not the same as key (!) # make sure we adjust our estimate of no match appropriately no_match += min(len(expected_text), NNlen - 1) - m # sample only the negatives that are true negatives # that is, they are not in the expected set - sampling only 'semi-hard negatives is not defined here' # positives = expected_text - nearest_text positives = overlap negatives = nearest_text - expected_text # print(key + str(expected_text) + str(nearest_text)) for i in negatives: for j in positives: dist_pos = t.get_distance(index, entity2unique[j]) pos_distances.append(dist_pos) dist_neg = t.get_distance(index, entity2unique[i]) neg_distances.append(dist_neg) if dist_pos < dist_neg: ann_accuracy += 1 total += 1 # print(key + "|" + j + "|" + i) # print(dist_pos) # print(dist_neg) min_neg_distance = 1000000 for i in negatives: dist_neg = t.get_distance(index, entity2unique[i]) all_neg_distances.append(dist_neg) if dist_neg < min_neg_distance: min_neg_distance = dist_neg for j in expected_text: dist_pos = t.get_distance(index, entity2unique[j]) all_pos_distances.append(dist_pos) closest_pos_count = 0 for p in overlap: dist_pos = t.get_distance(index, entity2unique[p]) if dist_pos < min_neg_distance: closest_pos_count+=1 if closest_pos_count > 0: precise+=1 closest_positive_counts.append(closest_pos_count / min(len(expected_text), NNlen - 1)) for i in negatives: for j in expected_text: triplets['anchor'].append(key) triplets['positive'].append(j) triplets['negative'].append(i) print("mean closest positive count:" + str(statistics.mean(closest_positive_counts))) print("mean positive distance:" + str(statistics.mean(pos_distances))) print("stdev positive distance:" + str(statistics.stdev(pos_distances))) print("max positive distance:" + str(max(pos_distances))) print("mean neg distance:" + str(statistics.mean(neg_distances))) print("stdev neg distance:" + str(statistics.stdev(neg_distances))) print("max neg distance:" + str(max(neg_distances))) print("mean all positive distance:" + str(statistics.mean(all_pos_distances))) print("stdev all positive distance:" + str(statistics.stdev(all_pos_distances))) print("max all positive distance:" + str(max(all_pos_distances))) print("mean all neg distance:" + str(statistics.mean(all_neg_distances))) print("stdev all neg distance:" + str(statistics.stdev(all_neg_distances))) print("max all neg distance:" + str(max(all_neg_distances))) print("Accuracy in the ANN for triplets that obey the distance func:" + str(ann_accuracy / total)) print("Precision at 1: " + str(precise / len(entity2same))) obj = {} obj['accuracy'] = ann_accuracy / total obj['steps'] = 1 with open(output_file_name_for_hpo, 'w') as out: json.dump(obj, out) if test: return match/(match + no_match) else: return triplets, match/(match + no_match) def generate_names(entities, people, limit_pairs=False): if people: num_names = 4 generator = NameDataCleanser(0, num_names, limit_pairs=limit_pairs) else: generator = CompanyDataCleanser(limit_pairs) num_names = 2 entity2same = {} for entity in entities: ret = generator.cleanse_data(entity) if ret and len(ret) >= num_names: entity2same[ret[0]] = ret[1:] return entity2same def embedded_representation_model(embedding_layer): seq = Sequential() seq.add(embedding_layer) seq.add(Flatten()) return seq def build_model(embedder): main_input = Input(shape=(MAX_SEQUENCE_LENGTH,)) net = embedder(main_input) for i in range(0, NUM_LAYERS): net = GRU(128, return_sequences=True, activation='relu', name='embed' + str(i))(net) net = GRU(128, activation='relu', name='embed' + str(i+1))(net) if USE_L2_NORM: net = Lambda(l2Norm, output_shape=[128])(net) base_model = Model(embedder.input, net, name='triplet_model') base_model.summary() input_shape=(MAX_SEQUENCE_LENGTH,) input_anchor = Input(shape=input_shape, name='input_anchor') input_positive = Input(shape=input_shape, name='input_pos') input_negative = Input(shape=input_shape, name='input_neg') net_anchor = base_model(input_anchor) net_positive = base_model(input_positive) net_negative = base_model(input_negative) positive_dist = Lambda(euclidean_distance, name='pos_dist', output_shape=(1,))([net_anchor, net_positive]) negative_dist = Lambda(euclidean_distance, name='neg_dist', output_shape=(1,))([net_anchor, net_negative]) if USE_ANGULAR_LOSS: n_c = Lambda(n_c_angular_distance, name='nc_angular_dist')([net_anchor, net_positive, net_negative]) a_p = Lambda(a_p_angular_distance, name='ap_angular_dist')([net_anchor, net_positive, net_negative]) stacked_dists = Lambda( lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([a_p, n_c]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=angular_loss, metrics=[accuracy]) else: exemplar_negative_dist = Lambda(euclidean_distance, name='exemplar_neg_dist', output_shape=(1,))([net_positive, net_negative]) stacked_dists = Lambda( # lambda vects: C.splice(*vects, axis=C.Axis.new_leading_axis()).eval(vects), lambda vects: K.stack(vects, axis=1), name='stacked_dists', output_shape=(3, 1) )([positive_dist, negative_dist, exemplar_negative_dist]) model = Model([input_anchor, input_positive, input_negative], stacked_dists, name='triple_siamese') model.compile(optimizer="rmsprop", loss=LOSS_FUNCTION, metrics=[accuracy]) test_positive_model = Model([input_anchor, input_positive, input_negative], positive_dist) test_negative_model = Model([input_anchor, input_positive, input_negative], negative_dist) inter_model = Model(input_anchor, net_anchor) print("output_shapes") model.summary() # print(positive_dist.output_shape) # print(negative_dist.output_shape) # print(exemplar_negative_dist) # print(neg_dist.output_shape) return model, test_positive_model, test_negative_model, inter_model parser = argparse.ArgumentParser(description='Run fuzzy join algorithm') parser.add_argument('--debug_sample_size', type=int, help='sample size for debug run') parser.add_argument('--margin', type=int, help='margin') parser.add_argument('--loss_function', type=str, help='triplet loss function type: schroff-loss, improved-loss, angular-loss, tanh-loss, improved-tanh-loss') parser.add_argument('--use_l2_norm', type=str, help='whether to add a l2 norm') parser.add_argument('--num_layers', type=int, help='num_layers to use. Minimum is 2') parser.add_argument('--input', type=str, help='Input file') parser.add_argument('--entity_type', type=str, help='people or companies') parser.add_argument('--model', type=str, help='name for model file') args = parser.parse_args() filepath = args.model LOSS_FUNCTION = None if args.loss_function == 'schroff-loss': LOSS_FUNCTION=schroff_triplet_loss elif args.loss_function == 'improved-loss': LOSS_FUNCTION=improved_loss elif args.loss_function == 'our-loss': LOSS_FUNCTION=triplet_loss elif args.loss_function == 'tanh-loss': LOSS_FUNCTION=triplet_tanh_loss elif args.loss_function == 'improved-tanh-loss': LOSS_FUNCTION=triplet_tanh_pn_loss elif args.loss_function == 'angular-loss': USE_ANGULAR_LOSS = True LOSS_FUNCTION = angular_loss print('Loss function: ' + args.loss_function) if args.debug_sample_size: DEBUG=True DEBUG_DATA_LENGTH=args.debug_sample_size print('Debug data length:' + str(DEBUG_DATA_LENGTH)) print('Margin:' + str(MARGIN)) USE_L2_NORM = args.use_l2_norm.lower() in ("yes", "true", "t", "1") print('Use L2Norm: ' + str(USE_L2_NORM)) print('Use L2Norm: ' + str(args.use_l2_norm)) NUM_LAYERS = args.num_layers - 1 print('Num layers: ' + str(NUM_LAYERS)) people = 'people' in args.entity_type # read all entities and create positive parts of a triplet entities = read_entities(args.input) train, test = split(entities, test_split = .20) print("TRAIN") print(train) print("TEST") print(test) entity2same_train = generate_names(train, people) entity2same_test = generate_names(test, people, limit_pairs=True) print(entity2same_train) print(entity2same_test) # change the default behavior of the tokenizer to ignore all punctuation except , - and . which are important # clues for entity names tokenizer = Tokenizer(num_words=MAX_NB_WORDS, lower=True, filters='!"#$%&()*+/:;<=>?@[\]^_`{|}~', split=" ") # build a set of data structures useful for annoy, the set of unique entities (unique_text), # a mapping of entities in texts to an index in unique_text, a mapping of entities to other same entities, and the actual # vectorized representation of the text. These structures will be used iteratively as we build up the model # so we need to create them once for re-use unique_text, entity2unique = build_unique_entities(entity2same_train) unique_text_test, entity2unique_test = build_unique_entities(entity2same_test) print("train text len:" + str(len(unique_text))) print("test text len:" + str(len(unique_text_test))) tokenizer.fit_on_texts(unique_text + unique_text_test) sequences = tokenizer.texts_to_sequences(unique_text) sequences = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH) sequences_test = tokenizer.texts_to_sequences(unique_text_test) sequences_test = pad_sequences(sequences_test, maxlen=MAX_SEQUENCE_LENGTH) # build models embedder = get_embedding_layer(tokenizer) model, test_positive_model, test_negative_model, inter_model = build_model(embedder) embedder_model = embedded_representation_model(embedder) if DEBUG_ANN: generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) sys.exit() test_data, test_match_stats = generate_triplets_from_ANN(embedder_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, False) test_seq = get_sequences(test_data, tokenizer) print("Test stats:" + str(test_match_stats)) counter = 0 current_model = embedder_model prev_match_stats = 0 train_data, match_stats = generate_triplets_from_ANN(current_model, sequences, entity2unique, entity2same_train, unique_text, False) print("Match stats:" + str(match_stats)) number_of_names = len(train_data['anchor']) # print(train_data['anchor']) print("number of names" + str(number_of_names)) Y_train = np.random.randint(2, size=(1,2,number_of_names)).T checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max') early_stop = EarlyStopping(monitor='val_accuracy', patience=1, mode='max') callbacks_list = [checkpoint, early_stop] train_seq = get_sequences(train_data, tokenizer) # check just for 5 epochs because this gets called many times model.fit([train_seq['anchor'], train_seq['positive'], train_seq['negative']], Y_train, epochs=100, batch_size=40, callbacks=callbacks_list, validation_split=0.2) current_model = inter_model # print some statistics on this epoch print("training data predictions") positives = test_positive_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) negatives = test_negative_model.predict([train_seq['anchor'], train_seq['positive'], train_seq['negative']]) print("f1score for train is: {}".format(f1score(positives, negatives))) print("test data predictions") positives = test_positive_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) negatives = test_negative_model.predict([test_seq['anchor'], test_seq['positive'], test_seq['negative']]) print("f1score for test is: {}".format(f1score(positives, negatives))) test_match_stats = generate_triplets_from_ANN(current_model, sequences_test, entity2unique_test, entity2same_test, unique_text_test, True) print("Test stats:" + str(test_match_stats))
21,235
36.061082
163
py
smt
smt-master/setup.py
""" Author: Dr. John T. Hwang <hwangjt@umich.edu> Dr. Mohamed A. Bouhlel <mbouhlel@umich.edu> Remi Lafage <remi.lafage@onera.fr> Lucas Alber <lucasd.alber@gmail.com> This package is distributed under New BSD license. """ from setuptools import setup, Extension import sys import numpy as np from Cython.Build import cythonize from smt import __version__ CLASSIFIERS = """\ Development Status :: 5 - Production/Stable Intended Audience :: Science/Research Intended Audience :: Developers License :: OSI Approved :: BSD License Programming Language :: C++ Programming Language :: Python Programming Language :: Python :: 3 Programming Language :: Python :: 3.7 Programming Language :: Python :: Implementation :: CPython Topic :: Software Development Topic :: Scientific/Engineering Operating System :: Microsoft :: Windows Operating System :: Unix Operating System :: MacOS """ LONG_DESCRIPTION = """ The surrogate modeling toolbox (SMT) is a Python package that contains a collection of surrogate modeling methods, sampling techniques, and benchmarking functions. This package provides a library of surrogate models that is simple to use and facilitates the implementation of additional methods. SMT is different from existing surrogate modeling libraries because of its emphasis on derivatives, including training derivatives used for gradient-enhanced modeling, prediction derivatives, and derivatives with respect to the training data. It also includes new surrogate models that are not available elsewhere: kriging by partial-least squares reduction and energy-minimizing spline interpolation. """ extra_compile_args = [] if not sys.platform.startswith("win"): extra_compile_args.append("-std=c++11") ext = ( cythonize( Extension( "smt.surrogate_models.rbfclib", sources=["smt/src/rbf/rbf.cpp", "smt/src/rbf/rbfclib.pyx"], language="c++", extra_compile_args=extra_compile_args, include_dirs=[np.get_include()], ) ) + cythonize( Extension( "smt.surrogate_models.idwclib", sources=["smt/src/idw/idw.cpp", "smt/src/idw/idwclib.pyx"], language="c++", extra_compile_args=extra_compile_args, include_dirs=[np.get_include()], ) ) + cythonize( Extension( "smt.surrogate_models.rmtsclib", sources=[ "smt/src/rmts/rmtsclib.pyx", "smt/src/rmts/utils.cpp", "smt/src/rmts/rmts.cpp", "smt/src/rmts/rmtb.cpp", "smt/src/rmts/rmtc.cpp", ], language="c++", extra_compile_args=extra_compile_args, include_dirs=[np.get_include()], ) ) ) metadata = dict( name="smt", version=__version__, description="The Surrogate Modeling Toolbox (SMT)", long_description=LONG_DESCRIPTION, author="Mohamed Amine Bouhlel et al.", author_email="mbouhlel@umich.edu", license="BSD-3", classifiers=[_f for _f in CLASSIFIERS.split("\n") if _f], packages=[ "smt", "smt.surrogate_models", "smt.problems", "smt.sampling_methods", "smt.utils", "smt.utils.neural_net", "smt.applications", ], install_requires=[ "scikit-learn", "pyDOE2", "scipy", ], extras_require={ "numba": [ # pip install smt[numba] "numba~=0.56.4", ], }, python_requires=">=3.7", zip_safe=False, ext_modules=ext, url="https://github.com/SMTorg/smt", # use the URL to the github repo download_url="https://github.com/SMTorg/smt/releases", ) setup(**metadata)
3,762
29.346774
87
py
smt
smt-master/smt/__init__.py
__version__ = "2.0"
20
9.5
19
py
smt
smt-master/smt/examples/run_examples.py
""" Author: Dr. Mohamed A. Bouhlel <mbouhlel@umich> Dr. John T. Hwang <hwangjt@umich.edu> This package is distributed under New BSD license. """ import numpy as np from scipy import linalg from smt.utils import compute_rms_error from smt.problems import Sphere, NdimRobotArm from smt.sampling_methods import LHS from smt.surrogate_models import LS, QP, KPLS, KRG, KPLSK, GEKPLS, DesignSpace try: from smt.surrogate_models import IDW, RBF, RMTC, RMTB compiled_available = True except: compiled_available = False try: import matplotlib.pyplot as plt plot_status = True except: plot_status = False ########### Initialization of the problem, construction of the training and validation points ndim = 10 ndoe = int(10 * ndim) # Define the function fun = Sphere(ndim=ndim) # Construction of the DOE sampling = LHS(xlimits=fun.xlimits, criterion="m") xt = sampling(ndoe) # Compute the output yt = fun(xt) # Compute the gradient for i in range(ndim): yd = fun(xt, kx=i) yt = np.concatenate((yt, yd), axis=1) # Construction of the validation points ntest = 500 sampling = LHS(xlimits=fun.xlimits) xtest = sampling(ntest) ytest = fun(xtest) ydtest = np.zeros((ntest, ndim)) for i in range(ndim): ydtest[:, i] = fun(xtest, kx=i).T ########### The LS model # Initialization of the model t = LS(print_prediction=False) # Add the DOE t.set_training_values(xt, yt[:, 0]) # Train the model t.train() # Prediction of the validation points y = t.predict_values(xtest) print("LS, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the LS model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "LS, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The QP model t = QP(print_prediction=False) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("QP, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the QP model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "QP, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The Kriging model # The variable 'theta0' is a list of length ndim. t = KRG(theta0=[1e-2] * ndim, print_prediction=False) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("Kriging, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the Kriging model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "Kriging, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The KPLS model # The variables 'name' must be equal to 'KPLS'. 'n_comp' and 'theta0' must be # an integer in [1, ndim[ and a list of length n_comp, respectively. Here is an # an example using 2 principal components. t = KPLS(n_comp=2, theta0=[1e-2, 1e-2], print_prediction=False) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("KPLS, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the KPLS model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "KPLS, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() # KPLS + absolute exponential correlation kernel # The variables 'name' must be equal to 'KPLS'. 'n_comp' and 'theta0' must be # an integer in [1,ndim[ and a list of length n_comp, respectively. Here is an # an example using 2 principal components. t = KPLS(n_comp=2, theta0=[1e-2, 1e-2], print_prediction=False, corr="abs_exp") t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("KPLS + abs exp, err: " + str(compute_rms_error(t, xtest, ytest))) ########### The KPLSK model # 'n_comp' and 'theta0' must be an integer in [1, ndim[ and a list of length n_comp, respectively. t = KPLSK(n_comp=2, theta0=[1e-2, 1e-2], print_prediction=False) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("KPLSK, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the KPLSK model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "KPLSK, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The GEKPLS model using 1 approximating points # 'n_comp' and 'theta0' must be an integer in [1,ndim[ and a list of length n_comp, respectively. design_space = DesignSpace(fun.xlimits) t = GEKPLS( n_comp=1, theta0=[1e-2], design_space=design_space, delta_x=1e-2, extra_points=1, print_prediction=False, ) t.set_training_values(xt, yt[:, 0]) # Add the gradient information for i in range(ndim): t.set_training_derivatives(xt, yt[:, 1 + i].reshape((yt.shape[0], 1)), i) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("GEKPLS1, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the GEKPLS1 model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "GEKPLS1, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The GEKPLS model using 2 approximating points # 'n_comp' and 'theta0' must be an integer in [1,ndim[ and a list of length n_comp, respectively. t = GEKPLS( n_comp=1, theta0=[1e-2], xlimits=fun.xlimits, delta_x=1e-4, extra_points=2, print_prediction=False, ) t.set_training_values(xt, yt[:, 0]) # Add the gradient information for i in range(ndim): t.set_training_derivatives(xt, yt[:, 1 + i].reshape((yt.shape[0], 1)), i) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("GEKPLS2, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the GEKPLS2 model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "GEKPLS2, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() if compiled_available: ########### The IDW model t = IDW(print_prediction=False) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("IDW, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: plt.figure() plt.plot(ytest, ytest, "-.") plt.plot(ytest, y, ".") plt.xlabel(r"$y_{true}$") plt.ylabel(r"$\hat{y}$") plt.title("Validation of the IDW model") plt.show() ########### The RBF model t = RBF(print_prediction=False, poly_degree=0) t.set_training_values(xt, yt[:, 0]) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("RBF, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(4, 3) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[3, 2].arrow(0.3, 0.3, 0.2, 0) axarr[3, 2].arrow(0.3, 0.3, 0.0, 0.4) axarr[3, 2].text(0.25, 0.4, r"$\hat{y}$") axarr[3, 2].text(0.35, 0.15, r"$y_{true}$") axarr[3, 2].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[3, 2].get_xticklabels(), visible=False) plt.setp(axarr[3, 2].get_yticklabels(), visible=False) plt.suptitle( "Validation of the RBF model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:10" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "RBF, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 2: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The RMTB and RMTC models are suitable for low-dimensional problems # Initialization of the problem ndim = 3 ndoe = int(250 * ndim) # Define the function fun = NdimRobotArm(ndim=ndim) # Construction of the DOE sampling = LHS(xlimits=fun.xlimits) xt = sampling(ndoe) # Compute the output yt = fun(xt) # Compute the gradient for i in range(ndim): yd = fun(xt, kx=i) yt = np.concatenate((yt, yd), axis=1) # Construction of the validation points ntest = 500 sampling = LHS(xlimits=fun.xlimits) xtest = sampling(ntest) ytest = fun(xtest) ########### The RMTB model t = RMTB( xlimits=fun.xlimits, min_energy=True, nonlinear_maxiter=20, print_prediction=False, ) t.set_training_values(xt, yt[:, 0]) # Add the gradient information for i in range(ndim): t.set_training_derivatives(xt, yt[:, 1 + i].reshape((yt.shape[0], 1)), i) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("RMTB, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(3, 2) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[2, 0].arrow(0.3, 0.3, 0.2, 0) axarr[2, 0].arrow(0.3, 0.3, 0.0, 0.4) axarr[2, 0].text(0.25, 0.4, r"$\hat{y}$") axarr[2, 0].text(0.35, 0.15, r"$y_{true}$") axarr[2, 0].axis("off") axarr[2, 1].set_visible(False) axarr[2, 1].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[2, 0].get_xticklabels(), visible=False) plt.setp(axarr[2, 0].get_yticklabels(), visible=False) plt.suptitle( "Validation of the RMTB model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:3" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "RMTB, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 1: l = 0 k += 1 else: l += 1 if plot_status: plt.show() ########### The RMTC model t = RMTC( xlimits=fun.xlimits, min_energy=True, nonlinear_maxiter=20, print_prediction=False, ) t.set_training_values(xt, yt[:, 0]) # Add the gradient information for i in range(ndim): t.set_training_derivatives(xt, yt[:, 1 + i].reshape((yt.shape[0], 1)), i) t.train() # Prediction of the validation points y = t.predict_values(xtest) print("RMTC, err: " + str(compute_rms_error(t, xtest, ytest))) if plot_status: k, l = 0, 0 f, axarr = plt.subplots(3, 2) axarr[k, l].plot(ytest, ytest, "-.") axarr[k, l].plot(ytest, y, ".") l += 1 axarr[2, 0].arrow(0.3, 0.3, 0.2, 0) axarr[2, 0].arrow(0.3, 0.3, 0.0, 0.4) axarr[2, 0].text(0.25, 0.4, r"$\hat{y}$") axarr[2, 0].text(0.35, 0.15, r"$y_{true}$") axarr[2, 0].axis("off") axarr[2, 1].set_visible(False) axarr[2, 1].axis("off") # Fine-tune figure; hide x ticks for top plots and y ticks for right plots plt.setp(axarr[2, 0].get_xticklabels(), visible=False) plt.setp(axarr[2, 0].get_yticklabels(), visible=False) plt.suptitle( "Validation of the RMTC model (from left to right then from top to bottom): validation of the prediction model and the i-th prediction of the derivative---i=1:3" ) # Prediction of the derivatives with regards to each direction space yd_prediction = np.zeros((ntest, ndim)) for i in range(ndim): yd_prediction[:, i] = t.predict_derivatives(xtest, kx=i).T print( "RMTC, err of the " + str(i) + "-th derivative: " + str(compute_rms_error(t, xtest, ydtest[:, i], kx=i)) ) if plot_status: axarr[k, l].plot(ydtest[:, i], ydtest[:, i], "-.") axarr[k, l].plot(ydtest[:, i], yd_prediction[:, i], ".") if l == 1: l = 0 k += 1 else: l += 1 if plot_status: plt.show()
21,474
30.259098
173
py