pytorch RNN层api的几个参数说明

classtorch.nn.RNN(*args, **kwargs)

input_size – The number of expected features in the input x

hidden_size – The number of features in the hidden state h

num_layers – Number of recurrent layers. E.g., setting num_layers=2 would mean stacking two RNNs together to form a stacked RNN, with the second RNN taking in outputs of the first RNN and computing the final results. Default: 1

nonlinearity – The non-linearity to use. Can be either ‘tanh’ or ‘relu’. Default: ‘tanh’

bias – If False, then the layer does not use bias weights b_ih and b_hh. Default: True

batch_first – If True, then the input and output tensors are provided as (batch, seq, feature)

dropout – If non-zero, introduces a Dropout layer on the outputs of each RNN layer except the last layer, with dropout probability equal to dropout. Default: 0

bidirectional – If True, becomes a bidirectional RNN. Default: False

 

有个参数一直理解错误，导致了认知困难

首先，RNN这里的序列长度，是动态的，不写在参数里的，具体会由输入的input参数而定

而num_layers并不是RNN的序列长度，而是堆叠层数，由上一层每个时间节点的输出作为下一层每个时间节点的输入

RNN的对象接受的参数，input维度是(seq_len, batch_size, input_dim)，h0维度是(num_layers * directions, batch_size, hidden_dim)

其中，input的seq_len决定了序列的长度，h0是提供给每层RNN的初始输入，所有num_layers要和RNN的num_layers对得上

返回两个值，一个output，一个hn

hn的维度是(num_layers * directions, batch_size, hidden_dim)，是RNN的右侧输出，如果是双向的话，就还有一个左侧输出

output的维度是(seq_len, batch_size, hidden_dim * directions)，是RNN的上侧输出