expand的含义:为1的维度可以变大维度或者维度数增多
tensor_1.expand(size):把tensor_1扩展成size的形状
tensor_1.expand_as(tensor_2) :把tensor_1扩展成和tensor_2一样的形状
import torch
#1
x = torch.randn(2, 1, 1)#为1可以扩展为3和4
x = x.expand(2, 3, 4)
print('x :', x.size())
>>> x : torch.Size([2, 3, 4])
#2
#扩展一个新的维度必须在最前面,否则会报错
x = x.expand(2, 3, 4, 6)
>>> RuntimeError: The expanded size of the tensor (3) must match the existing size (2) at non-singleton dimension 1.
x = x.expand(6, 2, 3, 4)
>>> x : torch.Size([6, 2, 3, 4])
#3
#某一个维度为-1表示不改变该维度的大小
x = x.expand(6, -1, -1, -1)
>>> x : torch.Size([6, 2, 1, 1])
import torch
#1
x = torch.randn(2, 1, 1)#为1可以扩展为3和3
y = torch.randn(2, 3, 3)
x = x.expand_as(y)
print('x :', x.size())
>>> x : torch.Size([2, 3, 3])
#2
x = torch.randn(2, 2, 2)#为2不可以扩展为3和4
y = torch.randn(2, 3, 4)
x = x.expand_as(y)
print('x :', x.size())
>>> RuntimeError: The expanded size of the tensor (4) must match the existing size (2) at non-singleton dimension 2. Target sizes: [2, 3, 4].