Pytorch学习-GRU使用

import torch.nn as nn
import torch

# gru = nn.GRU(input_size=50, hidden_size=50, batch_first=True)
# embed = nn.Embedding(3, 50)
# x = torch.LongTensor([[0, 1, 2]])
# x_embed = embed(x)
# out, hidden = gru(x_embed)


gru = nn.GRU(input_size=5, hidden_size=6,
             num_layers=2,  # gru层数
             batch_first=False,  # 默认参数 True:(batch, seq, feature) False：True:( seq,batch, feature),
             bidirectional=False,  # 默认参数
             )

# N=batch size
# L=sequence length
# D=2 if bidirectional=True else 1
# Hin=input size
# Hout=outout size


input_ = torch.randn(1, 3, 5)  # (L,N,hin)(序列长度，batch size大小，输入维度大小)
h0 = torch.randn(2 * 1, 3, 6)  # (D∗num_layers,N,Hout)（是否双向乘以层数，batch size大小，输出维度大小）

output, hn = gru(input_, h0)
# output:[1, 3, 6] (L,N,D*Hout)=(1,3,1*6)
# hn:[2, 3, 6] (D*num_layers,N,Hout)(1*2,3,6)

print(output.shape, hn.shape)
# torch.Size([1, 3, 6]) torch.Size([2, 3, 6])