PyTorch：Encoder-RNN|LSTM|GRU

男娘i 2022-12-09 15:55 238阅读 0赞

[\-柚子皮-][-_-]

## RNN ##

### 参数 ###

**Parameters**  
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 堆叠层数

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). Default: False

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 是否使用双向rnn。

Note: 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。

**Inputs: input, h\_0**  
input维度 input of shape (seq\_len, batch, input\_size): tensor containing the features of the input sequence. The input can also be a packed variable length sequence. See torch.nn.utils.rnn.pack\_padded\_sequence() or torch.nn.utils.rnn.pack\_sequence() for details.

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3BpcGlzb3JyeQ_size_16_color_FFFFFF_t_70][]

\[[https://blog.csdn.net/zwqjoy/article/details/86490098][https_blog.csdn.net_zwqjoy_article_details_86490098]\]

h0维度 h\_0 of shape (num\_layers \* num\_directions, batch, hidden\_size): tensor containing the initial hidden state for each element in the batch. Defaults to zero if not provided. If the RNN is bidirectional, num\_directions should be 2, else it should be 1.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。

**Outputs: output, h\_n**  
output of shape (seq\_len, batch, num\_directions \* hidden\_size): tensor containing the output features (h\_t) from the last layer of the RNN, for each t. If a torch.nn.utils.rnn.PackedSequence has been given as the input, the output will also be a packed sequence.For the unpacked case, the directions can be separated using output.view(seq\_len, batch, num\_directions, hidden\_size), with forward and backward being direction 0 and 1 respectively. Similarly, the directions can be separated in the packed case.RNN的上侧输出。

h\_n of shape (num\_layers \* num\_directions, batch, hidden\_size): tensor containing the hidden state for t = seq\_len.Like output, the layers can be separated using h\_n.view(num\_layers, num\_directions, batch, hidden\_size).RNN的右侧输出，如果是双向的话，就还有一个左侧输出。

具体参数和返回结果参考\[[https://pytorch.org/docs/stable/generated/torch.nn.RNN.html\#torch.nn.RNN][https_pytorch.org_docs_stable_generated_torch.nn.RNN.html_torch.nn.RNN]\]

### 示例 ###

rnn=nn.RNN(10,20,2) \#(each\_input\_size, hidden\_state, num\_layers)  
input=torch.randn(5,3,10) \# (seq\_len, batch, input\_size)  
h0=torch.randn(2,3,20) \#(num\_layers \* num\_directions, batch, hidden\_size)  
output,hn=rnn(input,h0)  
print(output.size(),hn.size())

## LSTM ##

具体参数和返回结果参考\[[https://pytorch.org/docs/stable/generated/torch.nn.LSTM.html\#torch.nn.LSTM][https_pytorch.org_docs_stable_generated_torch.nn.LSTM.html_torch.nn.LSTM]\]

### 示例 ###

rnn=nn.LSTM(10,20,2) \#(each\_input\_size, hidden\_state, num\_layers)  
input=torch.randn(5,3,10) \# (seq\_len, batch, input\_size)  
h0=torch.randn(2,3,20) \#(num\_layers \* num\_directions, batch, hidden\_size)  
c0=torch.randn(2,3,20) \#(num\_layers \* num\_directions, batch, hidden\_size)  
output,(hn,cn)=rnn(input,(h0,c0))   \#seq\_len x batch x hidden\*bi\_directional  
print(output.size(),hn.size(),cn.size())

##    GRU ##

gru = nn.GRU(embed\_size, hidden\_size, n\_layers, dropout=dropout, bidirectional=True)

具体参数参考：\[[https://pytorch.org/docs/stable/generated/torch.nn.GRU.html\#gru][https_pytorch.org_docs_stable_generated_torch.nn.GRU.html_gru]\]

### 示例 ###

import torch  
import torch.nn as nn  
rnn = nn.GRU(2, 4, 2,bidirectional=True)  
input = torch.randn(2, 2, 2)  
h0 = torch.randn(4, 2, 4)  
output, hn = rnn(input, h0)  
print(output)  
print(hn)  
print(output.size(),hn.size())

from: [\-柚子皮-][-_-]

ref:\[[LSTM和GRU原理及pytorch代码，输入输出大小说明][LSTM_GRU_pytorch]\]

[-_-]: https://me.csdn.net/pipisorry
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3BpcGlzb3JyeQ_size_16_color_FFFFFF_t_70]: /images/20221123/ddb3d0c5d7664f1aa1daae56b0aa765d.png
[https_blog.csdn.net_zwqjoy_article_details_86490098]: https://blog.csdn.net/zwqjoy/article/details/86490098
[https_pytorch.org_docs_stable_generated_torch.nn.RNN.html_torch.nn.RNN]: https://pytorch.org/docs/stable/generated/torch.nn.RNN.html#torch.nn.RNN
[https_pytorch.org_docs_stable_generated_torch.nn.LSTM.html_torch.nn.LSTM]: https://pytorch.org/docs/stable/generated/torch.nn.LSTM.html#torch.nn.LSTM
[https_pytorch.org_docs_stable_generated_torch.nn.GRU.html_gru]: https://pytorch.org/docs/stable/generated/torch.nn.GRU.html#gru
[LSTM_GRU_pytorch]: https://blog.csdn.net/qq_30468133/article/details/85244957

PyTorch：Encoder-RNN|LSTM|GRU

发表评论取消回复

还没有评论，来说两句吧...

相关阅读