data.py 文件源码

def get_dataset():
    f = gzip.open('mnist.pkl.gz', 'rb')
    u = pickle._Unpickler(f)
    u.encoding = 'latin1'
    train_set, valid_set, test_set = u.load()
    f.close()

    return train_set, valid_set, test_set