如何查看 tf.train.batch() 的输出

Question

我一直在wildml上关注"RNNs in TensorFlow, a Practical Guide and Undocumented Features" post，但无法查看tf.train.batch()函数的输出。存储、加载和处理输入的代码如下：

sequences = [[1, 2, 3], [4, 5, 1], [1, 2]]
label_sequences = [[0, 1, 0], [1, 0, 0], [1, 1]]

def make_example(sequence, labels):
    # The object we return
    ex = tf.train.SequenceExample()
    # A non-sequential feature of our example
    sequence_length = len(sequence)
    ex.context.feature["length"].int64_list.value.append(sequence_length)
    # Feature lists for the two sequential features of our example
    fl_tokens = ex.feature_lists.feature_list["tokens"]
    fl_labels = ex.feature_lists.feature_list["labels"]
    for token, label in zip(sequence, labels):
        fl_tokens.feature.add().int64_list.value.append(token)
        fl_labels.feature.add().int64_list.value.append(label)
    return ex
fname = "/home/someUser/PycharmProjects/someTensors"
writer = tf.python_io.TFRecordWriter(fname)
for sequence, label_sequence in zip(sequences, label_sequences):
    ex = make_example(sequence, label_sequence)
    print ex
    writer.write(ex.SerializeToString())
writer.close()
print("Wrote to {}".format(fname))
reader = tf.TFRecordReader()
filename_queue = tf.train.string_input_producer([fname])
_, serialized_example = reader.read(filename_queue)
context_parsed, sequence_parsed = tf.parse_single_sequence_example(
serialized=serialized_example, context_features=context_features,
sequence_features=sequence_features)

init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
tf.train.start_queue_runners(sess=sess)

batched_data = tf.train.batch(tensors=
[context_parsed['length'], sequence_parsed['tokens'],     
sequence_parsed['labels']], batch_size=5, dynamic_pad= True)

batched_context_data = tf.train.batch(tensors= [context_parsed['length']],
batch_size=5, dynamic_pad= True)

batched_tokens_data = tf.train.batch(tensors=
[sequence_parsed['tokens']], batch_size=5, dynamic_pad= True)

batched_labels_data = tf.train.batch(tensors=
[sequence_parsed['labels']], batch_size=5, dynamic_pad= True)

基于post，应该可以查看批次的输出如下：

res = tf.contrib.learn.run_n({"y": batched_data}, n=1, feed_dict=None)
print("Batch shape: {}".format(res[0]["y"].shape))
print(res[0]["y"])

或者更具体的情况如下：

res = tf.contrib.learn.run_n({"y": batched_context_data}, n=1, feed_dict=None)
print("Batch shape: {}".format(res[0]["y"].shape))
print(res[0]["y"])

不幸的是，TensorFlow 需要很长时间才能计算这两种情况，所以我最终终止了进程。谁能告诉我我做错了什么？

非常感谢！

Answer 1

我怀疑问题出在这一行，调用 tf.train.start_queue_runners():

tf.train.start_queue_runners(sess=sess)

...出现在这些行之前，其中包含对 tf.train.batch():

的调用

batched_data = tf.train.batch(...)

batched_context_data = tf.train.batch(...)

batched_tokens_data = tf.train.batch(...)

batched_labels_data = tf.train.batch(...)

如果在调用 tf.train.batch() 之后将调用移动到 tf.train.start_queue_runners() ，那么您的程序应该不会再死锁。

为什么会这样？ tf.train.batch() 函数在内部创建队列来缓冲正在批处理的数据，在 TensorFlow 中填充这些队列的常用方法是创建一个 "queue runner"，它（通常）是一个移动的后台线程元素放入队列。 tf.train.start_queue_runners() 函数在调用时为所有已注册队列运行器启动后台线程，但如果在创建队列运行器之前调用它，则不会启动这些线程。

如何查看 tf.train.batch() 的输出

How to view the output of tf.train.batch()

lstm

tensorflow

recurrent-neural-network