ValueError:没有为任何变量提供渐变：['embedding/embeddings:0'，']

Question

我是Tensorflow 2的新手，我想在keras/Tensorflow中训练一个多输入神经网络。这是我的示例代码：

First_inputs = Input(shape=(2000, ),name="first")
Second_inputs = Input(shape=(4, ),name="second")
embedding_layer = Embedding(3,3,  input_length=2000,)(First_inputs)
flatten = Flatten()(embedding_layer)
first_dense = Dense(neuronCount,kernel_initializer=initializer, )(flatten)
merge = concatenate([first_dense, Second_inputs])
drop = Dropout(dropout)(merge)
output = Dense(1, )(drop)
model = Model(inputs=[First_inputs, Second_inputs], outputs=output)
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.1,shuffle=True,  random_state=42)
First_inputs =x_train[:,0:2000]
Second_inputs =x_train[:,2000:2004]
model.fit(([First_inputs, Second_inputs], y_train),validation_data=([First_inputs, Second_inputs], y_train),verbose=1,epochs=100,steps_per_epoch=209)

但是，我得到了这个错误：

ValueError: No gradients provided for any variable: ['embedding/embeddings:0', 'dense/kernel:0', 'dense/bias:0', 'dense_1/kernel:0'].

有人知道问题出在哪里吗？谢谢

Answer 1

您的数据是numpy数组，必须为fit（）方法提供两个单独的参数，即作为输入的np.array列表和作为标签的np.array列表。（删除作为输入的元组）：

First_inputs = Input(shape=(2000, ),name="first")
Second_inputs = Input(shape=(4, ),name="second")
embedding_layer = Embedding(3,3,  input_length=2000,)(First_inputs)
flatten = Flatten()(embedding_layer)
first_dense = Dense(neuronCount,kernel_initializer=initializer, )(flatten)
merge = concatenate([first_dense, Second_inputs])
drop = Dropout(dropout)(merge)
output = Dense(1, )(drop)
model = Model(inputs=[First_inputs, Second_inputs], outputs=output)
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.1,shuffle=True,  
random_state=42)
First_inputs =x_train[:,0:2000]
Second_inputs =x_train[:,2000:2004]
model.fit([First_inputs, Second_inputs], y_train,validation_data=([First_inputs, 
Second_inputs], y_train),verbose=1,epochs=100,steps_per_epoch=209)