tf_load_and_enqueue.py

def load_and_enqueue(input_dir, sess, coord, enqueue_op, queue_inputs, queue_targets, num_examples, examples_per_file=100, rewrite_targets=True):

    # Check if we have a sufficient number of HDF5 files to load all the samples
    filenames_queue = glob.glob(os.path.join(input_dir, "train/*.h5"))
    filenames_queue.sort()
    assert len(filenames_queue) > 0

    examples_available = len(filenames_queue)*examples_per_file
    num_examples = min(examples_available, num_examples)

    num_files = int(math.ceil(num_examples / examples_per_file))
    filenames_queue = filenames_queue[0:num_files]

    while True:
        for filename in filenames_queue:

            # Read 100 examples from HDF5 file, shuffle files within file
            inputs, targets, motion_types = read_examples_from_file(filename, shuffle=True, rewrite_targets=rewrite_targets)

            feed_dict = {
                queue_inputs:  inputs,
                queue_targets: targets
            }

            # Feed examples to the FIFO queue
            sess.run(enqueue_op, feed_dict=feed_dict)

            if coord.should_stop(): return

        random.shuffle(filenames_queue)