错误：串联图层时，“ NoneType”对象没有属性“ _inbound_nodes”（多输入模型）

Question

我对深度学习和神经网络还很陌生。

我有一个具有文本和数字功能的数据集，我正在尝试使用此处提供的方法解决此问题

我将数据集分为文本（X_text）和数字（X_num）功能。 我将text（X_text）中的所有列添加到单个列中，其他列删除了。 然后，我在此列上运行TfidfVectorizer并将其转换为具有形状的数组（1905，20859）。 X_num的形状为（1905,34）

我在此之后使用的代码

from keras.models import Sequential
from keras.layers import Dense, Embedding, Flatten, LSTM, Input, Bidirectional, Concatenate
from keras.optimizers import adam
from keras import regularizers
from keras.backend import concatenate
from keras import Model

nlp_input = Input(shape=(20860,))
meta_input = Input(shape=(35,))
emb = Embedding(output_dim=32, input_dim=20859)(nlp_input)
nlp_output = Bidirectional(LSTM(128, dropout=0.3, recurrent_dropout=0.3, kernel_regularizer=regularizers.l2(0.01)))(emb)
x = concatenate([nlp_out, meta_input])
layer1 = Dense(32, activation='relu')(x)
layer2 = Dense(1, activation='sigmoid')(layer1)
model = Model(inputs=[nlp_input , meta_input], outputs=layer2)
optimizer=adam(lr=0.00001)
model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics = ['binary_accuracy'])

我得到的错误是：

    Traceback (most recent call last)
        <ipython-input-51-d98028f8916d> in <module>
             13 layer1 = Dense(32, activation='relu')(x)
             14 layer2 = Dense(1, activation='sigmoid')(layer1)
        ---> 15 model = Model(inputs=[nlp_input , meta_input], outputs=layer2)

        /anaconda3/lib/python3.6/site-packages/keras/legacy/interfaces.py in wrapper(*args, **kwargs)
             89                 warnings.warn('Update your `' + object_name + '` call to the ' +
             90                               'Keras 2 API: ' + signature, stacklevel=2)
        ---> 91             return func(*args, **kwargs)
             92         wrapper._original_function = func
             93         return wrapper

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in __init__(self, *args, **kwargs)
             91                 'inputs' in kwargs and 'outputs' in kwargs):
             92             # Graph network
        ---> 93             self._init_graph_network(*args, **kwargs)
             94         else:
             95             # Subclassed network

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in _init_graph_network(self, inputs, outputs, name)
            229         # Keep track of the network's nodes and layers.
            230         nodes, nodes_by_depth, layers, layers_by_depth = _map_graph_network(
        --> 231             self.inputs, self.outputs)
            232         self._network_nodes = nodes
            233         self._nodes_by_depth = nodes_by_depth

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in _map_graph_network(inputs, outputs)
           1364                   layer=layer,
           1365                   node_index=node_index,
        -> 1366                   tensor_index=tensor_index)
           1367 
           1368     for node in reversed(nodes_in_decreasing_depth):

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in build_map(tensor, finished_nodes, nodes_in_progress, layer, node_index, tensor_index)
           1351             tensor_index = node.tensor_indices[i]
           1352             build_map(x, finished_nodes, nodes_in_progress, layer,
        -> 1353                       node_index, tensor_index)
           1354 
           1355         finished_nodes.add(node)

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in build_map(tensor, finished_nodes, nodes_in_progress, layer, node_index, tensor_index)
           1351             tensor_index = node.tensor_indices[i]
           1352             build_map(x, finished_nodes, nodes_in_progress, layer,
        -> 1353                       node_index, tensor_index)
           1354 
           1355         finished_nodes.add(node)

        /anaconda3/lib/python3.6/site-packages/keras/engine/network.py in build_map(tensor, finished_nodes, nodes_in_progress, layer, node_index, tensor_index)
           1323             ValueError: if a cycle is detected.
           1324         """
        -> 1325         n

ode = layer._inbound_nodes[node_index]
       1326 
       1327         # Prevent cycles.

    AttributeError: 'NoneType' object has no attribute '_inbound_nodes'

我在其他地方读到可以使用Lambda层，该层将函数用作keras中的层，也许这就是问题所在。 但是据我所知，我没有函数可以调用。 任何想法如何解决这个问题？

Answer 1

当将nlp_output与meta_input串联时，您将使用keras.backend.concatenate ，而应在其中使用keras.layers.Concatenate 。 下面的代码应该工作：

nlp_input = Input(shape=(20860,))
meta_input = Input(shape=(35,))
emb = Embedding(output_dim=32, input_dim=20859)(nlp_input)
nlp_output = Bidirectional(LSTM(128, dropout=0.3, recurrent_dropout=0.3, kernel_regularizer=regularizers.l2(0.01)))(emb)
x = Concatenate()([nlp_output, meta_input])
layer1 = Dense(32, activation='relu')(x)
layer2 = Dense(1, activation='sigmoid')(layer1)
model = Model(inputs=[nlp_input , meta_input], outputs=layer2)
optimizer=adam(lr=0.00001)
model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics = ['binary_accuracy'])

注意： keras.backend的函数可以包装在Lambda层中，但是当已经有keras.layers层提供所需功能时，这没有多大意义。 对于您的情况，如果要在Lambda层中使用keras.backend.concatenate ，可以执行以下操作：

concatenated = keras.layers.Lambda(lambda x: keras.backend.concatenate(x))([input1, input2])