Datasets:
jblitzar
/
github-python-corpus

Dataset card Data Studio Files
xet
 Community
text
stringlengths
1
93.6k
try:
args.func(args)
except ValueError as e:
print(e)
exit(-1)
exit(0)
# <FILESEP>
import torch
import torchvision
import json
from PIL import Image
from resnet_utils import myResnet
import numpy as np
import torch.nn as nn
from Sublayers import Norm, 全连接层
import math
import torch.nn.functional as F
from 模型_策略梯度 import Transformer
from Batch import create_masks
device = torch.device("cuda:0" if (torch.cuda.is_available()) else "cpu")
resnet101=torchvision.models.resnet101(pretrained=True).eval()
resnet101=myResnet(resnet101).cuda(device).requires_grad_(False)
from random import shuffle
def gelu(x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
class 判断状态(nn.Module):
def __init__(self, 种类数, 隐藏层尺寸, 输入层尺寸=2048,输入尺寸A=36):
super().__init__()
self.隐藏层尺寸=隐藏层尺寸
self.输入层尺寸=输入层尺寸
self.输入尺寸A = 输入尺寸A
self.输入层 = 全连接层(输入层尺寸, 隐藏层尺寸)
self.隐藏层 = 全连接层(隐藏层尺寸, 隐藏层尺寸)
self.输出层 = 全连接层(隐藏层尺寸*输入尺寸A, 种类数)
def forward(self, 图向量):
图向量 = 图向量.reshape((图向量.shape[0], self.输入尺寸A,self.输入层尺寸))
中间量=gelu(self.输入层 (图向量))
中间量=self.隐藏层 (中间量)
中间量=中间量.reshape((中间量.shape[0],self.隐藏层尺寸*self.输入尺寸A))
结果=self.输出层 (中间量)
return 结果
def random_dic(dicts):
dict_key_ls = list(dicts.keys())
shuffle(dict_key_ls)
new_dic = {}
for key in dict_key_ls:
new_dic[key] = dicts.get(key)
return new_dic
#model_判断状态=判断状态(6,1024,2048).cuda(device)
model_判断状态=Transformer(6,768,2,12,0.0,6*6*2048).cuda(device)
#model_判断状态.load_state_dict(torch.load('weights/model_weights_判断状态C1'))
optimizer = torch.optim.Adam(model_判断状态.parameters(), lr=6.25e-5, betas=(0.9, 0.98), eps=1e-9)
路径json='../判断数据样本/判断新.json'
全部数据={}
状态辞典={'击杀小兵或野怪或推掉塔': 0, '击杀敌方英雄': 1, '被击塔攻击': 2, '被击杀': 3, '死亡': 4, '普通': 5}
状态列表=[]
for K in 状态辞典:
状态列表.append(K)
with open(路径json, encoding='ansi') as f:
while True:
df = f.readline()
df = df.replace('\'', '\"')
if df == "":
break
单元 = json.loads(df)
for key in 单元:
全部数据[key]=单元[key]
状态 = np.ones((1, ), dtype='int64')
for i in range(100):
打乱顺序=random_dic(全部数据)
for key in 打乱顺序:
状态编号=状态辞典[全部数据[key]]
状态[0]=状态编号
目标输出=torch.from_numpy(状态).cuda(device)
图片路径 = '../判断数据样本/' + key + '.jpg'
img = Image.open(图片路径)
img2 = np.array(img)
img2 = torch.from_numpy(img2).cuda(device).unsqueeze(0).permute(0, 3, 2, 1).float() / 255
_, out = resnet101(img2)
图片张量 = out.reshape(1, 6 * 6 * 2048)
操作序列=np.ones((1,1))
操作张量 = torch.from_numpy(操作序列.astype(np.int64)).cuda(device)
src_mask, trg_mask = create_masks(操作张量.unsqueeze(0), 操作张量.unsqueeze(0), device)