external/CBraMod/preprocessing/preprocessing_speech.py

import h5py
import scipy
from scipy import signal
import os
import lmdb
import pickle
import numpy as np
import pandas as pd


train_dir = '/data/datasets/BigDownstream/Imagined speech/mat/Training set'
val_dir = '/data/datasets/BigDownstream/Imagined speech/mat/Validation set'
test_dir = '/data/datasets/BigDownstream/Imagined speech/mat/Test set'



files_dict = {
    'train':sorted([file for file in os.listdir(train_dir)]),
    'val':sorted([file for file in os.listdir(val_dir)]),
    'test':sorted([file for file in os.listdir(test_dir)]),
}

print(files_dict)

dataset = {
    'train': list(),
    'val': list(),
    'test': list(),
}

db = lmdb.open('/data/datasets/BigDownstream/Imagined speech/processed', map_size=3000000000)

for file in files_dict['train']:
    data = scipy.io.loadmat(os.path.join(train_dir, file))
    print(data['epo_train'][0][0][0])
    eeg = data['epo_train'][0][0][4].transpose(2, 1, 0)
    labels = data['epo_train'][0][0][5].transpose(1, 0)
    eeg = eeg[:, :, -768:]
    labels = np.argmax(labels, axis=1)
    eeg = signal.resample(eeg, 600, axis=2).reshape(300, 64, 3, 200)
    print(eeg.shape, labels.shape)
    for i, (sample, label) in enumerate(zip(eeg, labels)):
        sample_key = f'train-{file[:-4]}-{i}'
        data_dict = {
            'sample': sample, 'label': label,
        }
        txn = db.begin(write=True)
        txn.put(key=sample_key.encode(), value=pickle.dumps(data_dict))
        txn.commit()
        print(sample_key)
        dataset['train'].append(sample_key)


for file in files_dict['val']:
    data = scipy.io.loadmat(os.path.join(val_dir, file))
    eeg = data['epo_validation'][0][0][4].transpose(2, 1, 0)
    labels = data['epo_validation'][0][0][5].transpose(1, 0)
    eeg = eeg[:, :, -768:]
    labels = np.argmax(labels, axis=1)
    eeg = signal.resample(eeg, 600, axis=2).reshape(50, 64, 3, 200)
    print(eeg.shape, labels.shape)
    for i, (sample, label) in enumerate(zip(eeg, labels)):
        sample_key = f'val-{file[:-4]}-{i}'
        data_dict = {
            'sample': sample, 'label': label,
        }
        txn = db.begin(write=True)
        txn.put(key=sample_key.encode(), value=pickle.dumps(data_dict))
        txn.commit()
        print(sample_key)
        dataset['val'].append(sample_key)


df = pd.read_excel("/data/datasets/BigDownstream/Imagined speech/mat/Track3_Answer Sheet_Test.xlsx")
df_=df.head(53)
all_labels=df_.values
print(all_labels.shape)
all_labels = all_labels[2:, 1:][:, 1:30:2].transpose(1, 0)
print(all_labels.shape)
print(all_labels)

for j, file in enumerate(files_dict['test']):
    data = h5py.File(os.path.join(test_dir, file))
    eeg = data['epo_test']['x'][:]
    labels = all_labels[j]
    eeg = eeg[:, :, -768:]
    eeg = signal.resample(eeg, 600, axis=2).reshape(50, 64, 3, 200)
    print(eeg.shape, labels.shape)
    for i, (sample, label) in enumerate(zip(eeg, labels)):
        sample_key = f'test-{file[:-4]}-{i}'
        data_dict = {
            'sample': sample, 'label': label-1,
        }
        txn = db.begin(write=True)
        txn.put(key=sample_key.encode(), value=pickle.dumps(data_dict))
        txn.commit()
        print(sample_key)
        dataset['test'].append(sample_key)


txn = db.begin(write=True)
txn.put(key='__keys__'.encode(), value=pickle.dumps(dataset))
txn.commit()
db.close()