pytorch rnn输入一个batch内的序列pad到同一长度

最新推荐文章于 2024-08-09 15:25:17 发布

u010867294

最新推荐文章于 2024-08-09 15:25:17 发布

阅读量7.6k

点赞数

分类专栏： pytorch

本文链接：https://blog.csdn.net/u010867294/article/details/79464027

版权

在PyTorch中训练RNN模型时，由于一个batch内的序列长度不一，需要使用pad操作统一长度。通过pack_padded_sequence和pad_packed_sequence确保序列正确输入和输出。RNN不会对填充元素进行计算，避免了无效运算。动态RNN或使用packed input可以更好地处理这种情况，避免对padding进行不必要的计算。

摘要由CSDN通过智能技术生成

import torch  
from torch.autograd import Variable  
from torch import nn  
  
batch_size = 4  
max_length = 3  
hidden_size = 2  
n_layers = 1  
feature_dim = 1  
  
# container  
batch_in = torch.zeros((batch_size, max_length, feature_dim))  
  
# data  
vec_1 = torch.FloatTensor([[1, 2, 3]])  
vec_2 = torch.FloatTensor([[1, 2, 0]])  
vec_3 = torch.FloatTensor([[1, 0, 0]])  
vec_4 = torch.FloatTensor([[2, 0, 0]])  
  
batch_in[0] = vec_1  
batch_in[1] = vec_2  
batch_in[2] = vec_3  
batch_in[3] = vec_4  
  
batch_in = Variable(batch_in)  
print(batch_in.size())  
seq_lengths = [3, 2, 1, 1]  # list of integers holding information about the batch size at each sequence step  
  
# pack it  
pack = torch.nn.utils.rnn.pack_padded_sequence(batch_in, seq_lengths, batch_first=True)  
print(pack)  
  
rnn = nn.RNN(feature_dim, hidden_size, n_layers, batch_first=True)  
h0 = V