data_preprocessing.py

import numpy as np
from tqdm import tqdm
import scipy.io as scio


def intoBins(data, n_bins):
    k_labels = np.zeros((len(data), n_bins+1))
    bin_size = (np.max(data)+0.0001-np.min(data)) / n_bins
    for i in tqdm(range(len(data))):
        index = (data[i]-np.min(data)) // bin_size
        k_labels[i, int(index)] = 1
    return k_labels


def preprocessing(n_data, data_length, degrees_of_freedom, n_labels, n_bins, path='2500\\2500'):
    # save input
    all_data = np.zeros((n_data, degrees_of_freedom+n_labels+1+1, data_length))

    for i in tqdm(range(2500)):
        file_name = f'{path}\\Data{i+1}.mat'
        data = scio.loadmat(file_name)['Data'][:data_length, :]
        data = np.transpose(data)
        all_data[i] = data

    f_and_xs = all_data[:, 1:2+degrees_of_freedom, :]
    mean = np.mean(f_and_xs, (0, 2))
    std = np.std(f_and_xs, (0, 2))
    f_and_xs = f_and_xs - np.reshape(mean, (1, -1, 1))
    f_and_xs = f_and_xs / np.reshape(std, (1, -1, 1))
    dict = {'f_and_xs': f_and_xs}

    # save labels
    for i in range(n_labels):
        label = all_data[:, 2+degrees_of_freedom+i, 0]
        bins = intoBins(label, n_bins)[:, :-1]
        dict[f'l_{i}'] = bins

    np.save('dataset.npy', dict)


# preprocessing(2500, 10000, 6, 3, 10, path='2500\\2500')


def load_dataset(path='dataset.npy'):
    """
    :return:
        f_and_xs: numpy array of size [sample_number, channels, sample_length]
        label_0, label_1, label_2: one-hot encodes of size [sample_number, number_bins]
    """

    r = np.load(path, allow_pickle=True).item()
    f_and_xs = r['f_and_xs']
    label_0 = r['l_0']
    label_1 = r['l_1']
    label_2 = r['l_2']
    return f_and_xs, label_0, label_1, label_2


f_and_xs, label_0, label_1, label_2 = load_dataset()