2c5d24194dbe627dd78dbeabbe6811f3e274181a,generator.py,,data_generator,#,45

Before Change

def data_generator(data_file, index_list, batch_size=1, binary=True):
    nb_subjects = len(index_list)
    while True:
        shuffle(index_list)
        nb_batches = nb_subjects/batch_size
        // TODO: Edge case? Currently this is handled by flooring the number of training/testing samples
        for i in range(nb_batches):
            x = data_file.root.data[i*batch_size:(i+1)*batch_size]
            y = data_file.root.truth[i*batch_size:(i+1)*batch_size]
            if binary:
                y[y > 0] = 1
            else:
                raise NotImplementedError("Multi-class labels are not yet implemented")
            yield x, y

After Change

    while True:
        shuffle(index_list)
        x_list = list()
        y_list = list()
        // TODO: Edge case? Currently this is handled by flooring the number of training/testing samples
        for index in index_list:
            x_list.append(data_file.root.data[index])
            y_list.append(data_file.root.truth[index])
            if len(x_list) == batch_size:
                x = np.asarray(x_list)
                y = np.asarray(y_list)
                x_list = list()
                y_list = list()
                if binary:
                    y[y > 0] = 1
                else:
                    raise NotImplementedError("Multi-class labels are not yet implemented")
                yield x, y

Instances