Tensorflow，如何访问一个RNN的所有中间状态，而不仅仅是最后一个状态

Question

我的理解是tf.nn.dynamic_rnn 在每个时间步返回一个 RNN 单元（例如 LSTM）的输出以及最终状态。我如何才能在所有时间步骤中访问单元状态，而不仅仅是最后一个？例如，我希望能够对所有隐藏状态进行平均，然后在后续层中使用它。

以下是我如何定义一个 LSTM 单元，然后使用 tf.nn.dynamic_rnn 展开它。但这仅给出了 LSTM 的最后一个单元状态。

import tensorflow as tf
import numpy as np

# [batch-size, sequence-length, dimensions] 
X = np.random.randn(2, 10, 8)
X[1,6:] = 0
X_lengths = [10, 6]

cell = tf.contrib.rnn.LSTMCell(num_units=64, state_is_tuple=True)

outputs, last_state = tf.nn.dynamic_rnn(
    cell=cell,
    dtype=tf.float64,
    sequence_length=X_lengths,
    inputs=X)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())                                 
out, last = sess.run([outputs, last_state], feed_dict=None)

Answer 1

这样的事情应该可以工作。

import tensorflow as tf
import numpy as np


class CustomRNN(tf.contrib.rnn.LSTMCell):
    def __init__(self, *args, **kwargs):
        kwargs['state_is_tuple'] = False # force the use of a concatenated state.
        returns = super(CustomRNN, self).__init__(*args, **kwargs) # create an lstm cell
        self._output_size = self._state_size # change the output size to the state size
        return returns
    def __call__(self, inputs, state):
        output, next_state = super(CustomRNN, self).__call__(inputs, state)
        return next_state, next_state # return two copies of the state, instead of the output and the state

X = np.random.randn(2, 10, 8)
X[1,6:] = 0
X_lengths = [10, 10]

cell = CustomRNN(num_units=64)

outputs, last_states = tf.nn.dynamic_rnn(
    cell=cell,
    dtype=tf.float64,
    sequence_length=X_lengths,
    inputs=X)

sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())                                 
states, last_state = sess.run([outputs, last_states], feed_dict=None)

这使用连接状态，因为我不知道您是否可以存储任意数量的元组状态。 states 变量的形状为 (batch_size, max_time_size, state_size)。

Answer 2

我会把你指向这个thread（我的亮点）：

如果每个时间步都需要 c 和 h 状态，您可以编写 LSTMCell 的变体，将两个状态张量作为输出的一部分返回。如果只需要h状态，那就是每个时间步的输出。

正如@jasekp 在其评论中所写，输出实际上是状态的h 部分。然后dynamic_rnn 方法将跨时间堆叠所有h 部分（参见this file 中_dynamic_rnn_loop 的字符串文档）：

def _dynamic_rnn_loop(cell,
                      inputs,
                      initial_state,
                      parallel_iterations,
                      swap_memory,
                      sequence_length=None,
                      dtype=None):
  """Internal implementation of Dynamic RNN.
    [...]
    Returns:
    Tuple `(final_outputs, final_state)`.
    final_outputs:
      A `Tensor` of shape `[time, batch_size, cell.output_size]`.  If
      `cell.output_size` is a (possibly nested) tuple of ints or `TensorShape`
      objects, then this returns a (possibly nsted) tuple of Tensors matching
      the corresponding shapes.