Python TensorFlow：预测工作不正常

我对我制作的模型有一些问题

1） 有时，预测返回0的数组（但这可能是正常的）

2） 如果我重复预测多次（使用循环），我总是得到相同的结果

3） 奇怪的是，如果我停止程序并重新运行它，预测就会改变（第2点仍然适用）

4） 结果是错误的。我插入了与训练相同的精确查询（训练时预测的响应是正确的），但得到了非常不同的响应

现在，奇怪的是，在训练的时候，我可以看到它在进步，而且运行得很好

以下是显示正确结果（培训）的代码：

这是我在预测方面遇到的问题：

就这样。我不知道为什么会这样。感谢您的帮助。我还包括了其他方法，以提供更多的上下文

---

这是下一个\u提要（）：

这是批处理法：

def batch_method(inputs, max_sequence_length=None):
    """
    Args:
        inputs:
            list of sentences (integer lists)
        max_sequence_length:
            integer specifying how large should `max_time` dimension be.
            If None, maximum sequence length would be used

    Outputs:
        inputs_time_major:
            input sentences transformed into time-major matrix
            (shape [max_time, batch_size]) padded with 0s
        sequence_lengths:
            batch-sized list of integers specifying amount of active
            time steps in each input sequence
    """

    sequence_lengths = [len(seq) for seq in inputs]
    batch_size_ = len(inputs)

    if max_sequence_length is None:
        max_sequence_length = max(sequence_lengths)

    inputs_batch_major = np.zeros(shape=[batch_size_, max_sequence_length], dtype=np.int32)  # == PAD

    for index, seq in enumerate(inputs):
        for j, element in enumerate(seq):
            inputs_batch_major[index, j] = element

    # [batch_size, max_time] -> [max_time, batch_size]
    inputs_time_major = inputs_batch_major.swapaxes(0, 1)

    return inputs_time_major, sequence_lengths

saver.restore(sess, MODEL_LOCATION)
batch__ = []
query = get_formatted_sentence(sentence)
batch__.append(words_to_batch(query))
batch, batch_len = batch_method(batch__)
for x in range(0,10): # This is the loop I was talking about
    prediction = sess.run(decoder_prediction, feed_dict={encoder_inputs: batch, encoder_inputs_length: batch_len}) # Notice that here I do not give decoder_targets ( If I understand them correctly they are the 'wanted' result. In this case I have no 'wanted' result, I just want to get a prediction
    print(prediction.T)
    print(batch_to_words(prediction.T[0]))

def next_feed():

    encoder_inputs_, encoder_input_lengths_ = batch_method(batches_query)
    decoder_targets_, _ = batch_method(
        [sequence + [EOS] + [PAD] for sequence in iter(batches_response)]
    )
    #print(encoder_inputs_)
    #print(encoder_input_lengths_)
    return {
        encoder_inputs: encoder_inputs_,
        encoder_inputs_length: encoder_input_lengths_,
        decoder_targets: decoder_targets_,
    }

def batch_method(inputs, max_sequence_length=None):
    """
    Args:
        inputs:
            list of sentences (integer lists)
        max_sequence_length:
            integer specifying how large should `max_time` dimension be.
            If None, maximum sequence length would be used

    Outputs:
        inputs_time_major:
            input sentences transformed into time-major matrix
            (shape [max_time, batch_size]) padded with 0s
        sequence_lengths:
            batch-sized list of integers specifying amount of active
            time steps in each input sequence
    """

    sequence_lengths = [len(seq) for seq in inputs]
    batch_size_ = len(inputs)

    if max_sequence_length is None:
        max_sequence_length = max(sequence_lengths)

    inputs_batch_major = np.zeros(shape=[batch_size_, max_sequence_length], dtype=np.int32)  # == PAD

    for index, seq in enumerate(inputs):
        for j, element in enumerate(seq):
            inputs_batch_major[index, j] = element

    # [batch_size, max_time] -> [max_time, batch_size]
    inputs_time_major = inputs_batch_major.swapaxes(0, 1)

    return inputs_time_major, sequence_lengths