python - 属性错误 : ‘LSTMStateTuple’ object has no attribute ‘get_shape’ while building a Seq2Seq Model using Tensorflow

Question

我正在尝试使用 Amazon Reviews 数据集进行文本摘要。我在构建模型时遇到错误。

AttributeError: ‘LSTMStateTuple’ object has no attribute ‘get_shape’

我知道我错过了什么。但无法弄清楚它是什么。我是 tensorflow 的新手。

我想问题出在我的编码层，即我尝试连接输出的方式。

def encoding_layer(embeded_rnn_input,rnn_size,keep_prob,num_layers,batch_size,source_sequence_length):
    cell_fw = tf.contrib.rnn.MultiRNNCell([get_lstm(rnn_size,keep_prob) for _ in range(num_layers)])
    cell_bw = tf.contrib.rnn.MultiRNNCell([get_lstm(rnn_size,keep_prob) for _ in range(num_layers)])
    encoder_outputs,encoder_states = tf.nn.bidirectional_dynamic_rnn(cell_fw=cell_fw,cell_bw=cell_bw,inputs=embeded_rnn_input,
                                    sequence_length=source_sequence_length,dtype=tf.float32)
    encoder_outputs = tf.concat(encoder_outputs, 2)
    return encoder_outputs,encoder_states

编辑:

删除了指向笔记本的链接，因为它将在未来发生变化。为错误添加堆栈跟踪。

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-71-85ee67bc88e5> in <module>()
      9     # Create the training and inference logits
     10     training_logits, inference_logits = seq2seq_model(input_,target,embeding_matrix,vocab_to_int,source_seq_length,target_seq_length,
---> 11                   max_target_seq_length,rnn_size,keep_probability,num_layers,batch_size)
     12 
     13     # Create tensors for the training logits and inference logits

<ipython-input-70-5ad1bf459bd7> in seq2seq_model(source_input, target_input, embeding_matrix, vocab_to_int, source_sequence_length, target_sequence_length, max_target_length, rnn_size, keep_prob, num_layers, batch_size)
     15     training_logits, inference_logits = decoding_layer(target_input,encoder_states,embedings,
     16                                                                 vocab_to_int,rnn_size,target_sequence_length,
---> 17                                                                 max_target_length,batch_size,num_layers)
     18 
     19     return training_logits, inference_logits

<ipython-input-69-c2b4542605d2> in decoding_layer(target_inputs, encoder_state, embedding, vocab_to_int, rnn_size, target_sequence_length, max_target_length, batch_size, num_layers)
     12 
     13         training_logits = training_decoder(embed,decoder_cell,encoder_state,output_layer,
---> 14                                          target_sequence_length,max_target_length)
     15 
     16 

<ipython-input-67-91fcb3f89090> in training_decoder(dec_embed_input, decoder_cell, encoder_state, output_layer, target_sequence_length, max_target_length)
      8 
      9     final_outputs, final_state,_ = tf.contrib.seq2seq.dynamic_decode(decoder=decoder,impute_finished=True,
---> 10                                                      maximum_iterations=max_target_length)
     11 
     12     return final_outputs

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/contrib/seq2seq/python/ops/decoder.py in dynamic_decode(decoder, output_time_major, impute_finished, maximum_iterations, parallel_iterations, swap_memory, scope)
    284         ],
    285         parallel_iterations=parallel_iterations,
--> 286         swap_memory=swap_memory)
    287 
    288     final_outputs_ta = res[1]

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/control_flow_ops.py in while_loop(cond, body, loop_vars, shape_invariants, parallel_iterations, back_prop, swap_memory, name)
   2773     context = WhileContext(parallel_iterations, back_prop, swap_memory, name)
   2774     ops.add_to_collection(ops.GraphKeys.WHILE_CONTEXT, context)
-> 2775     result = context.BuildLoop(cond, body, loop_vars, shape_invariants)
   2776     return result
   2777 

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/control_flow_ops.py in BuildLoop(self, pred, body, loop_vars, shape_invariants)
   2602       self.Enter()
   2603       original_body_result, exit_vars = self._BuildLoop(
-> 2604           pred, body, original_loop_vars, loop_vars, shape_invariants)
   2605     finally:
   2606       self.Exit()

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/control_flow_ops.py in _BuildLoop(self, pred, body, original_loop_vars, loop_vars, shape_invariants)
   2552         structure=original_loop_vars,
   2553         flat_sequence=vars_for_body_with_tensor_arrays)
-> 2554     body_result = body(*packed_vars_for_body)
   2555     if not nest.is_sequence(body_result):
   2556       body_result = [body_result]

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/contrib/seq2seq/python/ops/decoder.py in body(time, outputs_ta, state, inputs, finished, sequence_lengths)
    232       """
    233       (next_outputs, decoder_state, next_inputs,
--> 234        decoder_finished) = decoder.step(time, inputs, state)
    235       next_finished = math_ops.logical_or(decoder_finished, finished)
    236       if maximum_iterations is not None:

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/contrib/seq2seq/python/ops/basic_decoder.py in step(self, time, inputs, state, name)
    137     """
    138     with ops.name_scope(name, "BasicDecoderStep", (time, inputs, state)):
--> 139       cell_outputs, cell_state = self._cell(inputs, state)
    140       if self._output_layer is not None:
    141         cell_outputs = self._output_layer(cell_outputs)

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in __call__(self, inputs, state, scope)
    178       with vs.variable_scope(vs.get_variable_scope(),
    179                              custom_getter=self._rnn_get_variable):
--> 180         return super(RNNCell, self).__call__(inputs, state)
    181 
    182   def _rnn_get_variable(self, getter, *args, **kwargs):

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/layers/base.py in __call__(self, inputs, *args, **kwargs)
    448         # Check input assumptions set after layer building, e.g. input shape.
    449         self._assert_input_compatibility(inputs)
--> 450         outputs = self.call(inputs, *args, **kwargs)
    451 
    452         # Apply activity regularization.

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in call(self, inputs, state)
    936                                       [-1, cell.state_size])
    937           cur_state_pos += cell.state_size
--> 938         cur_inp, new_state = cell(cur_inp, cur_state)
    939         new_states.append(new_state)
    940 

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in __call__(self, inputs, state, scope)
    772                              self._recurrent_input_noise,
    773                              self._input_keep_prob)
--> 774     output, new_state = self._cell(inputs, state, scope)
    775     if _should_dropout(self._state_keep_prob):
    776       new_state = self._dropout(new_state, "state",

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in __call__(self, inputs, state, scope)
    178       with vs.variable_scope(vs.get_variable_scope(),
    179                              custom_getter=self._rnn_get_variable):
--> 180         return super(RNNCell, self).__call__(inputs, state)
    181 
    182   def _rnn_get_variable(self, getter, *args, **kwargs):

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/layers/base.py in __call__(self, inputs, *args, **kwargs)
    448         # Check input assumptions set after layer building, e.g. input shape.
    449         self._assert_input_compatibility(inputs)
--> 450         outputs = self.call(inputs, *args, **kwargs)
    451 
    452         # Apply activity regularization.

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in call(self, inputs, state)
    399       c, h = array_ops.split(value=state, num_or_size_splits=2, axis=1)
    400 
--> 401     concat = _linear([inputs, h], 4 * self._num_units, True)
    402 
    403     # i = input_gate, j = new_input, f = forget_gate, o = output_gate

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in _linear(args, output_size, bias, bias_initializer, kernel_initializer)
   1019   # Calculate the total size of arguments on dimension 1.
   1020   total_arg_size = 0
-> 1021   shapes = [a.get_shape() for a in args]
   1022   for shape in shapes:
   1023     if shape.ndims != 2:

~/anaconda2/envs/tensorflow/lib/python3.5/site-packages/tensorflow/python/ops/rnn_cell_impl.py in <listcomp>(.0)
   1019   # Calculate the total size of arguments on dimension 1.
   1020   total_arg_size = 0
-> 1021   shapes = [a.get_shape() for a in args]
   1022   for shape in shapes:
   1023     if shape.ndims != 2:

AttributeError: 'LSTMStateTuple' object has no attribute 'get_shape'

Answer 1

我找到了答案。经过一些研究，我发现 encoder_states 的格式不正确。在解码层使用它们之前，需要将它们连接到每一层。

步骤:

1. 合并 encoder_fw_state.c 和 encoder_bw_state.c 以创建总的 encoder_state_c

2. 合并 encoder_fw_state.h 和 encoder_bw_state.h 以创建总的 encoder_state_h

3. 使用 encoder_state_c 和 encoder_state_h 创建另一个 LSTMStateTuple

例子:

encoder_state_c = tf.concat(values=(encoder_fw_state[i].c,encoder_bw_state[i].c),axis=1,name="encoder_fw_state_c")
encoder_state_h = tf.concat(values=(encoder_fw_state[i].h,encoder_bw_state[i].h),axis=1,name="encoder_fw_state_h")
encoder_state = tf.contrib.rnn.LSTMStateTuple(c=encoder_state_c, h=encoder_state_h)

下面是对我有用的完整编码器。

def encoding_layer(embeded_rnn_input,rnn_size,keep_prob,num_layers,batch_size,source_sequence_length):

    cell_fw = tf.contrib.rnn.MultiRNNCell([get_lstm(rnn_size,keep_prob) for _ in range(num_layers)])


    cell_bw = tf.contrib.rnn.MultiRNNCell([get_lstm(rnn_size,keep_prob) for _ in range(num_layers)])

    ((encoder_fw_outputs,
              encoder_bw_outputs),
             (encoder_fw_state,
              encoder_bw_state)) = tf.nn.bidirectional_dynamic_rnn(cell_fw=cell_fw,cell_bw=cell_bw,inputs=embeded_rnn_input,
                                    sequence_length=source_sequence_length,dtype=tf.float32)

    encoder_outputs = tf.concat((encoder_fw_outputs, encoder_bw_outputs), 2)

    encoder_states = []

    for i in range(num_layers):
        if isinstance(encoder_fw_state[i],tf.contrib.rnn.LSTMStateTuple):
            encoder_state_c = tf.concat(values=(encoder_fw_state[i].c,encoder_bw_state[i].c),axis=1,name="encoder_fw_state_c")
            encoder_state_h = tf.concat(values=(encoder_fw_state[i].h,encoder_bw_state[i].h),axis=1,name="encoder_fw_state_h")
            encoder_state = tf.contrib.rnn.LSTMStateTuple(c=encoder_state_c, h=encoder_state_h)
        elif isinstance(encoder_fw_state[i], tf.Tensor):
            encoder_state = tf.concat(values=(encoder_fw_state[i], encoder_bw_state[i]), axis=1, name='bidirectional_concat')

        encoder_states.append(encoder_state)

    encoder_states = tuple(encoder_states)

    return encoder_outputs,encoder_states