有什么我做错了让我的Keras model.fit继续抛出错误？

Question

这是我的模型：

model = Sequential()

model.add(layers.Embedding(vocab_size, embedding_size, mask_zero=True, input_length = TO_BE_FOUND))

model.add(layers.LSTM(hidden_size, dropout=0.2, recurrent_dropout=0.2, return_sequences=True))

model.add(layers.TimeDistributed(layers.Dense(4, activation='softmax')))

model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['acc'])

这是我合适的型号：

model.fit(train_x_padded, train_y_padded,batch_size=32, epochs=10, verbose=2,shuffle=True, validation_data=(train_x_padded, train_y_padded)

这是我得到的错误：

AlreadyExistsError Traceback (most recent call last) <ipython-input-24-109d4dab5962> in <module>() ----> 1 model.fit(train_x_padded, train_y_padded,batch_size=32, epochs=10, verbose=2,shuffle=True, validation_data=(train_x_padded, train_y_padded))
 .
 .
 .
 AlreadyExistsError: Resource __per_step_26/training_6/RMSprop/gradients/lstm_5/while/ReadVariableOp_4/Enter_grad/ArithmeticOptimizer/AddOpsRewrite_Add/tmp_var/N10tensorflow19TemporaryVariableOp6TmpVarE
[[{{node training_6/RMSprop/gradients/lstm_5/while/ReadVariableOp_4/Enter_grad/ArithmeticOptimizer/AddOpsRewrite_Add/tmp_var}}]]

Answer 1

尝试来自keras backend的clear_session：

keras.backend.clear_session()

销毁当前的TF图并创建一个新图。有助于避免旧模型/图层混乱。