process_data.py 文件源码

def process_training_data(num_clips):
    """
    Processes random training clips from the full training data. Saves to TRAIN_DIR_CLIPS by
    default.

    @param num_clips: The number of clips to process. Default = 5000000 (set in __main__).

    @warning: This can take a couple of hours to complete with large numbers of clips.
    """
    num_prev_clips = len(glob(c.TRAIN_DIR_CLIPS + '*'))

    for clip_num in xrange(num_prev_clips, num_clips + num_prev_clips):
        clip = process_clip()

        np.savez_compressed(c.TRAIN_DIR_CLIPS + str(clip_num), clip)

        if (clip_num + 1) % 100 == 0: print 'Processed %d clips' % (clip_num + 1)