python - 与张量流连接合并层keras的问题

Question

我想制作一个模型如下。

输入数据 输入数据 输入数据
| | | 转换输入 1 转换输入 2 转换输入 3 | | | 最大池 最大池 最大池

    - Dense layer -      - Dense layer -
       |                      |
    Dense layer            Dense layer

                   |
              Dense layer

---

我正在使用以下代码。

---

            kernel_stmt = []
            kernel_pos = []
            kernel_dep = []

            use_pos=False
            use_meta=True
            use_dep=True

            statement_input = Input(shape=(num_steps,), dtype='int32', name='main_input')
            x_stmt = Embedding(vocab_length+1,EMBED_DIM,weights=[embedding_matrix],input_length=num_steps,trainable=False)(statement_input) 

            # pos embed LSTM
            pos_input = Input(shape=(num_steps,), dtype='int32', name='pos_input')
            x_pos = Embedding(max(pos_dict.values()), max(pos_dict.values()), weights=[pos_embeddings], input_length=num_steps, trainable=False)(pos_input)

            # dep embed LSTM
            dep_input = Input(shape=(num_steps,), dtype='int32', name='dep_input')
            x_dep = Embedding(max(dep_dict.values()), max(dep_dict.values()), weights=[dep_embeddings], input_length=num_steps, trainable=False)(dep_input)


            for kernel in kernel_sizes:
                x_1 = Conv1D(filters=filter_size,kernel_size=kernel)(x_stmt)
                x_1 = GlobalMaxPool1D()(x_1)
                kernel_stmt.append(x_1)

                x_2 = Conv1D(filters=filter_size,kernel_size=kernel)(x_pos)
                x_2 = GlobalMaxPool1D()(x_2)
                kernel_pos.append(x_2)

                x_3 = Conv1D(filters=filter_size,kernel_size=kernel)(x_dep)
                x_3 = GlobalMaxPool1D()(x_3)
                kernel_dep.append(x_3)

            conv_in1 = keras.layers.concatenate(kernel_stmt)
            conv_in1 = Dropout(0.6)(conv_in1)
            conv_in1 = Dense(128, activation='relu')(conv_in1)

            conv_in2 = keras.layers.concatenate(kernel_pos)
            conv_in2 = Dropout(0.6)(conv_in2)
            conv_in2 = Dense(128, activation='relu')(conv_in2)

            conv_in3 = keras.layers.concatenate(kernel_dep)
            conv_in3 = Dropout(0.6)(conv_in3)
            conv_in3 = Dense(128, activation='relu')(conv_in3)

            # meta data
            meta_input = Input(shape=(X_train_meta.shape[1],), name='aux_input')
            x_meta = Dense(64, activation='relu')(meta_input)

            if use_pos and use_meta:
              if use_dep:
                x = keras.layers.concatenate([conv_in1, conv_in2, conv_in3, x_meta])
              else:
                x = keras.layers.concatenate([conv_in1, conv_in2, x_meta])
            elif use_meta:
              if use_dep:
                x = keras.layers.concatenate([conv_in1, conv_in3, x_meta])
              else:
                x = keras.layers.concatenate([conv_in1, x_meta])
            elif use_pos:
              if use_dep:
                x = keras.layers.concatenate([conv_in1, conv_in2, conv_in3])
              else:
                x = keras.layers.concatenate([conv_in1, conv_in2])
            else:
              if use_dep:
                x = keras.layers.concatenate([conv_in1, conv_in3])
              else:
                x = conv_in1



            main_output = Dense(6, activation='softmax', name='main_output')(x)

            if use_pos and use_meta:
              if use_dep:
                model_cnn = Model(inputs=[statement_input, pos_input, dep_input, meta_input], outputs=[main_output])
              else:
                model_cnn = Model(inputs=[statement_input, pos_input, meta_input], outputs=[main_output])
            elif use_meta:
              if use_dep:
                model_cnn = Model(inputs=[statement_input, dep_input, meta_input], outputs=[main_output])
              else:
                model_cnn = Model(inputs=[statement_input, meta_input], outputs=[main_output])
            elif use_pos:
              if use_dep:
                model_cnn = Model(inputs=[statement_input, pos_input, dep_input], outputs=[main_output])
              else:
                model_cnn = Model(inputs=[statement_input, pos_input], outputs=[main_output])
            else:
              if use_dep:
                model_cnn = Model(inputs=[statement_input, dep_input], outputs=[main_output])
              else:
                model_cnn = Model(inputs=[statement_input], outputs=[main_output])

---

但是，我得到了错误。

AttributeError：“连接”对象没有属性“outbound_nodes”

整个错误回溯如下。

            AttributeErrorTraceback (most recent call last)
            <ipython-input-121-c919448d2730> in <module>()
                 35 
                 36 conv_in1 = keras.layers.concatenate(kernel_stmt)
            ---> 37 conv_in1 = Dropout(0.6)(conv_in1)
                 38 conv_in1 = Dense(128, activation='relu')(conv_in1)
                 39 

            /usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/base_layer.pyc in __call__(self, inputs, *args, **kwargs)
                582           if base_layer_utils.have_all_keras_metadata(inputs):
                583             inputs, outputs = self._set_connectivity_metadata_(
            --> 584                 inputs, outputs, args, kwargs)
                585           if hasattr(self, '_set_inputs') and not self.inputs:
                586             # Subclassed network: explicitly set metadata normally set by

            /usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/base_layer.pyc in _set_connectivity_metadata_(self, inputs, outputs, args, kwargs)
               1414     kwargs.pop('mask', None)  # `mask` should not be serialized.
               1415     self._add_inbound_node(
            -> 1416         input_tensors=inputs, output_tensors=outputs, arguments=kwargs)
               1417     return inputs, outputs
               1418 

            /usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/base_layer.pyc in _add_inbound_node(self, input_tensors, output_tensors, arguments)
               1522         input_tensors=input_tensors,
               1523         output_tensors=output_tensors,
            -> 1524         arguments=arguments)
               1525 
               1526     # Update tensor history metadata.

            /usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/base_layer.pyc in __init__(self, outbound_layer, inbound_layers, node_indices, tensor_indices, input_tensors, output_tensors, arguments)
               1740         # For compatibility with external Keras, we use the deprecated
               1741         # accessor here.
            -> 1742         layer.outbound_nodes.append(self)
               1743     # For compatibility with external Keras, we use the deprecated
               1744     # accessor here.

标签： pythontensorflowkerasconv-neural-networklstm

Answer 1

我遇到了同样的问题，下面的修复对我有用。

代替

conv_in1 = keras.layers.concatenate(kernel_stmt)

尝试

concat = keras.layers.Concatenate(axis=1)  ## or whatever axis is relevant for you. In my case it was axis =1
conv_in1 = concat([kernel_stmt])