tensorflow - 如何修复'ValueError：输入0与层simple_rnn_1不兼容：预期形状=（无，无，20），找到形状=（无，无，2,20）'

Question

我有几个矩阵通过多个层，最后一个是密集层，为每个矩阵生成一个向量。现在我希望将这些矩阵提供给 keras 的 RNN，这就是我面临这个错误的地方。

我尝试将向量堆叠在一起，以便将它们传递给 RNN。这是该想法的一段代码：

input1 = Dense(20, activation = "relu")(input1)
input2 = Dense(20, activation = "relu")(input2)
out = Lambda(lambda x: tf.stack([x[0], x[1]], axis=1), output_shape=(None, 2, 20))([input1, input2])
out = SimpleRNN(50, activation="relu")(out)

我收到：

>Traceback (most recent call last):
  >>File "model.py", line 106, in <module>
    model = make_model()

  >>File "model.py", line 60, in make_model
    out = SimpleRNN(50, activation="relu")(out) 

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/layers/recurrent.py", line 532, in __call__
    return super(RNN, self).__call__(inputs, **kwargs)

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/engine/base_layer.py", line 440, in __call__
    self.assert_input_compatibility(inputs)

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/engine/base_layer.py", line 368, in assert_input_compatibility
    str(x_shape))

>>ValueError: Input 0 is incompatible with layer simple_rnn_1: expected shape=(None, None, 20), found shape=(None, None, 2, 20)

如果我更改output_shape=(None, None, 20)Lambda 层，我会得到：

Traceback (most recent call last):
 >> File "model.py", line 107, in <module>
    model.fit([input1, input2], y_train, epochs = 15, batch_size = 20, verbose = 2)

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/engine/training.py", line 952, in fit
    batch_size=batch_size)

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/engine/training.py", line 789, in _standardize_user_data
    exception_prefix='target')

  >>File "/home/yamini/.local/lib/python3.6/site-packages/keras/engine/training_utils.py", line 138, in standardize_input_data
    str(data_shape))

>>ValueError: Error when checking target: expected simple_rnn_1 to have shape (50,) but got array with shape (1,)

标签： tensorflowkerasrecurrent-neural-networkrnnkeras-layer

Answer 1

您可以更改output_shape，其中不应该包含batch_size.

from keras.layers import Dense,Lambda,SimpleRNN,Input
import tensorflow as tf

input1 = Input(shape=(20,))
input2 = Input(shape=(20,))
input1 = Dense(20, activation = "relu")(input1)
input2 = Dense(20, activation = "relu")(input2)
out = Lambda(lambda x: tf.stack([x[0], x[1]], axis=1), output_shape=(2, 20))([input1, input2])
out = SimpleRNN(50, activation="relu")(out)