torch.nn.LSTM()函数维度详解

import torch
import torch.nn as nn
lstm = nn.LSTM(10, 20, 2)
x = torch.randn(5, 3, 10)
h0 = torch.randn(2, 3, 20)
c0 = torch.randn(2, 3, 20)
output, (hn, cn)=lstm(x, (h0, c0))

>>
output.shape  torch.Size([5, 3, 20])
hn.shape  torch.Size([2, 3, 20])
cn.shape  torch.Size([2, 3, 20])

lstm=nn.LSTM(input_size,                     hidden_size,                      num_layers)
x                         seq_len,                          batch,                              input_size
h0            num_layers$\times$num_directions,   batch,                             hidden_size
c0            num_layers$\times$num_directions,   batch,                             hidden_size

output                 seq_len,                         batch,                num_directions$\times$hidden_size
hn            num_layers$\times$num_directions,   batch,                             hidden_size
cn            num_layers$\times$num_directions,    batch,                            hidden_size

举个例子：
对句子进行LSTM操作

假设有100个句子（sequence）,每个句子里有7个词，batch_size=64，embedding_size=300

此时，各个参数为：
input_size=embedding_size=300
batch=batch_size=64
seq_len=7

另外设置hidden_size=100, num_layers=1

import torch
import torch.nn as nn
lstm = nn.LSTM(300, 100, 1)
x = torch.randn(7, 64, 300)
h0 = torch.randn(1, 64, 100)
c0 = torch.randn(1, 64, 100)
output, (hn, cn)=lstm(x, (h0, c0))

>>
output.shape  torch.Size([7, 64, 100])
hn.shape  torch.Size([1, 64, 100])
cn.shape  torch.Size([1, 64, 100])