torch.nn.utils.rnn.pad_sequence(sequences, batch_first=False, padding_value=0)
用padding_value 填充一系列可变长度的tensor,把它们填充到等长
Example
>>> from torch.nn.utils.rnn import pad_sequence
>>> a = torch.ones(25, 300)
>>> b = torch.ones(22, 300)
>>> c = torch.ones(15, 300)
>>> pad_sequence([a, b, c]).size()
torch.Size([25, 3, 300])
from torch.nn.utils.rnn import pad_sequence
import torch
a=torch.randn(3)
b=torch.randn(5)
c=torch.randn(7)
>>> a
tensor([ 0.7160, 1.2006, -1.8447])
>>> b
tensor([ 0.3941, 0.3839, 0.1166, -0.7221, 1.8661])
>>> c
tensor([-0.6521, 0.0681, 0.6626, -0.3679, -0.6042, 1.6951, 0.4937])
>>> pad_sequence([a,b,c])
tensor([[ 0.7160, 0.3941, -0.6521],
[ 1.2006, 0.3839, 0.0681],
[-1.8447, 0.1166, 0.6626],
[ 0.0000, -0.7221, -0.3679],
[ 0.0000, 1.8661, -0.6042],
[ 0.0000, 0.0000, 1.6951],
[ 0.0000, 0.0000, 0.4937]])
>>> pad_sequence([a,b,c],batch_first=True)
tensor([[ 0.7160, 1.2006, -1.8447, 0.0000, 0.0000, 0.0000, 0.0000],
[ 0.3941, 0.3839, 0.1166, -0.7221, 1.8661, 0.0000, 0.0000],
[-0.6521, 0.0681, 0.6626, -0.3679, -0.6042, 1.6951, 0.4937]])
>>> pad_sequence([a,b,c],batch_first=True,padding_value=1)
tensor([[ 0.7160, 1.2006, -1.8447, 1.0000, 1.0000, 1.0000, 1.0000],
[ 0.3941, 0.3839, 0.1166, -0.7221, 1.8661, 1.0000, 1.0000],
[-0.6521, 0.0681, 0.6626, -0.3679, -0.6042, 1.6951, 0.4937]])
参考:
pad_sequence