我如何正确使用Keras Add图层？

Posted 2021-03-28

我试图将两个LSTM层合并在一起，但没有成功。

answers_questions_lstm = LSTM(256,input_shape=(8,4,))
answers_contexts_lstm = LSTM(256,input_shape=(10,6,))
answers_combined_lstm = Add()([answers_questions_lstm,answers_contexts_lstm])
answers_hidden_1 = Dense(124)(answers_combined_lstm)
answers_output = Dense(outputTrain.shape[1])
answers_network_1.summary()

这给了我“应该在输入列表上调用合并层。”为什么？

答案

由于正在使用Keras Functional API，因此应从某些Input图层开始。然后在Inputs上调用您的LSTM层以获得张量输出，然后可以将它们传递到Add层：

answers_questions_input = Input(shape=(8,4)) answers_contexts_input = Input(shape=(10,6)) answers_questions_lstm = LSTM(256)(answers_questions_input) answers_contexts_lstm = LSTM(256)(answers_contexts_input) answers_combined_lstm = Add()([answers_questions_lstm, answers_contexts_lstm]) answers_hidden_1 = Dense(124)(answers_combined_lstm) answers_output = Dense(outputTrain.shape[1]) answers_network_1 = Model(inputs=[answers_questions_input, answers_contexts_input], outputs=answers_output) answers_network_1.summary()