python - 堆叠 RNN 的输入形状

Question

我正在尝试使用 Keras 和 TensorFlow 后端串联堆叠一些 RNN。我可以使用单个 SimpleRNN 层创建模型，但是当我尝试添加第二个 SimpleRNN 层时，我无法计算出适当的输入大小。

from keras import models
from keras.layers.recurrent import SimpleRNN
from keras.layers import Activation


model = models.Sequential()

hidden_units = 256
skeleton_dimensions = 3 * 16  # 3 dimensions for 16 joints
input_temporal_length = 7

in_shape = (input_temporal_length, skeleton_dimensions,)

# three hidden layers of 256 each
model.add(SimpleRNN(hidden_units, input_shape=in_shape,
                    activation='relu', use_bias=True,))
# what input shape is this supposed to have?
model.add(SimpleRNN(hidden_units, input_shape=(1, skeleton_dimensions,),
                    activation='relu', use_bias=True,))

我的第二个 SimpleRNN 应该使用什么作为输入形状？

Recurrent Layers 的文档似乎暗示:

Output shape

• if return_sequences: 3D tensor with shape (batch_size, timesteps, units).
• else, 2D tensor with shape (batch_size, units).

鉴于return_sequences自动设置为False我尝试适本地设置下一个维度的input_shape，但我收到错误:

Using TensorFlow backend.
Traceback (most recent call last):
  File "rnn_agony.py", line 19, in <module>
    activation='relu', use_bias=True,))
  File "/usr/local/lib/python3.5/dist-packages/keras/models.py", line 455, in add
    output_tensor = layer(self.outputs[0])
  File "/usr/local/lib/python3.5/dist-packages/keras/layers/recurrent.py", line 252, in __call__
    return super(Recurrent, self).__call__(inputs, **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/keras/engine/topology.py", line 511, in __call__
    self.assert_input_compatibility(inputs)
  File "/usr/local/lib/python3.5/dist-packages/keras/engine/topology.py", line 413, in assert_input_compatibility
    str(K.ndim(x)))
ValueError: Input 0 is incompatible with layer simple_rnn_2: expected ndim=3, found ndim=2

Answer 1

如果您正在堆叠 RNN，则需要设置 return_sequences=True 并且不再需要设置 input_shape。这很直观，因为 RNN 需要输入序列。