Question

我必须输入要合并到一个网络中的输入，因此应该是这样的：

input1    input2  
   |        | 
   |        |
 hidden    hidden
     Merged

我尝试了以下代码：

b1_part = Sequential()
b1_part.add(Dense(units=b1_X_train.shape[1],activation='tanh', activity_regularizer=regularizers.l2(1e-2)))
b1_part.add(Dropout(0.5))
b1_part.add(Dense(units=b1_X_train.shape[1]/4,activation='tanh', activity_regularizer=regularizers.l2(1e-2)))


b2_part = Sequential()
b2_part.add(Dense(units=b2_X_train.shape[1],activation='tanh', activity_regularizer=regularizers.l2(1e-2)))
b2_part.add(Dropout(0.5))
b2_part.add(Dense(units=b2_X_train.shape[1]/4,activation='tanh', activity_regularizer=regularizers.l2(1e-2)))

result = Concatenate(axis=1)([b1_part, b2_part])

optimizer = Adagrad()
result.compile(optimizer=optimzier, loss=BinaryFocalLoss(gamma=2),
               metrics=['accuracy'])

但是得到了：

ValueError: Layer concatenate_10 was called with an input that isn't a symbolic tensor. Received type: <class 'tensorflow.python.keras.engine.sequential.Sequential'>. Full input: [<tensorflow.python.keras.engine.sequential.Sequential object at 0x7fb1cbf97048>, <tensorflow.python.keras.engine.sequential.Sequential object at 0x7fb1cbeb5358>]. All inputs to the layer should be tensors.

知道为什么吗？

Answer 1

首先，请记住指定顺序模型的输入形状。

然后，您可以通过以下方式组合顺序模型并定义完整模型：

b1_part = Sequential()
b1_part.add(Dense(units=100,activation='tanh', input_shape=(100,)))
b1_part.add(Dropout(0.5))
b1_part.add(Dense(units=25,activation='tanh'))

b2_part = Sequential()
b2_part.add(Dense(units=200,activation='tanh', input_shape=(200,)))
b2_part.add(Dropout(0.5))
b2_part.add(Dense(units=50,activation='tanh'))

result = Concatenate(axis=1)([b1_part.output, b2_part.output])
result = Dense(1, activation='sigmoid')(result)

model = Model([b1_part.input, b2_part.input], result)
model.compile('adam', 'binary_crossentropy', metrics='accuracy')

层的输入不是符号张量。接收类型为“顺序”

1 个答案: