Tensorflow 中 MultiRNNCell 的输出和状态 [英] Outputs and State of MultiRNNCell in Tensorflow

问题描述

我有一个堆叠的 MultiRNNCell 定义如下:

I have a stacked MultiRNNCell defined as below :

batch_size = 256
rnn_size = 512
keep_prob = 0.5

lstm_1 = tf.nn.rnn_cell.LSTMCell(rnn_size)
lstm_dropout_1 = tf.nn.rnn_cell.DropoutWrapper(lstm_1, output_keep_prob = keep_prob)

lstm_2 = tf.nn.rnn_cell.LSTMCell(rnn_size)
lstm_dropout_2 = tf.nn.rnn_cell.DropoutWrapper(lstm_2, output_keep_prob = keep_prob)

stacked_lstm = tf.nn.rnn_cell.MultiRNNCell([lstm_dropout_1, lstm_dropout_2])

rnn_inputs = tf.nn.embedding_lookup(embedding_matrix, ques_placeholder)

init_state = stacked_lstm.zero_state(batch_size, tf.float32)
rnn_outputs, final_state = tf.nn.dynamic_rnn(stacked_lstm, rnn_inputs, initial_state=init_state)

在这段代码中，有两个 RNN 层.我只想处理这个动态RNN的最终状态.我希望状态是形状 [batch_size, rnn_size*2] 的二维张量.

In this code, there are two RNN layers. I just want to process the final state of this dynamic RNN. I expected the state to be a 2D tensor of shape [batch_size, rnn_size*2].

final_state 的形状是 4D - [2,2,256,512]

The shape of the final_state is 4D - [2,2,256,512]

有人能解释一下为什么我会得到这个形状吗?另外，我如何处理这个张量，以便我可以通过一个完全连接的层?

Can someone please explain why am I getting this shape ? Also, how can I process this tensor so that I can pass it through a fully_connected layer ?

推荐答案

我无法重现 [2,2,256,512] 形状.但是用这段代码:

I can't reproduce the [2,2,256,512] shape. But with this piece of code:

rnn_size = 512
batch_size = 256
time_size = 5
input_size = 2
keep_prob = 0.5

lstm_1 = tf.nn.rnn_cell.LSTMCell(rnn_size)
lstm_dropout_1 = tf.nn.rnn_cell.DropoutWrapper(lstm_1, output_keep_prob=keep_prob)

lstm_2 = tf.nn.rnn_cell.LSTMCell(rnn_size)

stacked_lstm = tf.nn.rnn_cell.MultiRNNCell([lstm_dropout_1, lstm_2])

rnn_inputs = tf.placeholder(tf.float32, shape=[None, time_size, input_size])
# Shape of the rnn_inputs is (batch_size, time_size, input_size)

init_state = stacked_lstm.zero_state(batch_size, tf.float32)
rnn_outputs, final_state = tf.nn.dynamic_rnn(stacked_lstm, rnn_inputs, initial_state=init_state)
print(rnn_outputs)
print(final_state)

我得到了 run_outputs 的正确形状:(batch_size, time_size, rnn_size)

I get the right shape for run_outputs: (batch_size, time_size, rnn_size)

Tensor("rnn/transpose_1:0", shape=(256, 5, 512), dtype=float32)

final_state 确实是一对 LSTMStateTuple(对于 2 个单元格 lstm_dropout_1 和 lstm_2):

The final_state is indeed a pair of LSTMStateTuple (for the 2 cells lstm_dropout_1 and lstm_2):

(LSTMStateTuple(c=<tf.Tensor 'rnn/while/Exit_3:0' shape=(256, 512) dtype=float32>, h=<tf.Tensor 'rnn/while/Exit_4:0' shape=(256, 512) dtype=float32>),
 LSTMStateTuple(c=<tf.Tensor 'rnn/while/Exit_5:0' shape=(256, 512) dtype=float32>, h=<tf.Tensor 'rnn/while/Exit_6:0' shape=(256, 512) dtype=float32>))

如tf.nn.dynamic_run的字符串文档中所述:

as described in the string doc of tf.nn.dynamic_run:

  # 'outputs' is a tensor of shape [batch_size, max_time, 256]
  # 'state' is a N-tuple where N is the number of LSTMCells containing a
  # tf.contrib.rnn.LSTMStateTuple for each cell

这篇关于Tensorflow 中 MultiRNNCell 的输出和状态的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！