Tensorflow rnn api 阅读笔记

Tensorflow rnn api 阅读笔记

概述

tensorflow rnn api
rnn api 是用于将rnncell按照时间步骤进行自环的

常用rnn

1. 带有static前缀的api要求输入的序列具有固定长度。　　
2. 带有dynamic前缀的api可以选择输入一个sequence_length（可以是一个list）参数,该参数对应的是输入sequence的序列长度，用来动态处理sequence的长度（代码中是设置了一个专门记录序列长度的tensor,控制rnn自环的轮数）。
3. tf.nn.raw_rnn是原始的rnn api，能够使用该api实现各种定制化的操作。很好用。
   
   • tf.nn.static_rnn
   • tf.nn.static_state_saving_rnn
   • tf.nn.static_bidirectional_rnn
   • tf.nn.stack_bidirectional_dynamic_rnn
   • tf.nn.dynamic_rnn
   • tf.nn.dynamic_bidirectional_rnn
   • tf.nn.raw_rnn

dynamic_rnn

    def dynamic_rnn(cell, inputs, sequence_length=None, initial_state=None,
                    dtype=None, parallel_iterations=None, swap_memory=False,
                    time_major=False, scope=None):
    """Creates a recurrent neural network specified by RNNCell `cell`.

    Performs fully dynamic unrolling of `inputs`.

    Example:

    ```python
    # create a BasicRNNCell
    rnn_cell = tf.nn.rnn_cell.BasicRNNCell(hidden_size)

    # 'outputs' is a tensor of shape [batch_size, max_time, cell_state_size]

    # defining initial state
    initial_state = rnn_cell.zero_state(batch_size, dtype=tf.float32)

    # 'state' is a tensor of shape [batch_size, cell_state_size]
    outputs, state = tf.nn.dynamic_rnn(rnn_cell, input_data,
                                        initial_state=initial_state,
                                        dtype=tf.float32)
    ```

上下两个为官方给的使用样例

  # create 2 LSTMCells
  rnn_layers = [tf.nn.rnn_cell.LSTMCell(size) for size in [128, 256]]

  # create a RNN cell composed sequentially of a number of RNNCells
  multi_rnn_cell = tf.nn.rnn_cell.MultiRNNCell(rnn_layers)

  # '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
  outputs, state = tf.nn.dynamic_rnn(cell=multi_rnn_cell,
                                     inputs=data,
                                     dtype=tf.float32)

  Args:
    cell: An instance of RNNCell.
    inputs: The RNN inputs.
      If `time_major == False` (default), this must be a `Tensor` of shape:
        `[batch_size, max_time, ...]`, or a nested tuple of such
        elements.
      If `time_major == True`, this must be a `Tensor` of shape:
        `[max_time, batch_size, ...]`, or a nested tuple of such
        elements.
      This may also be a (possibly nested) tuple of Tensors satisfying
      this property.  The first two dimensions must match across all the inputs,
      but otherwise the ranks