torch.squeeze()的用法

用于去掉维数为1的维度

输入

import torch
a=torch.tensor([[[1,2,3],
                [4,5,6]]])
print(a)
print(a.shape)

结果

tensor([[[1, 2, 3],
         [4, 5, 6]]])
torch.Size([1, 2, 3])

输入 

b=a.squeeze()
print(b)
print(b.shape)

输出：将torch.Size([1, 2, 3])中的1维去掉了

tensor([[1, 2, 3],
        [4, 5, 6]])
torch.Size([2, 3])

下面试验一下shape里面有两个维数1，是否可以全部去掉

输入

import torch
a=torch.randn(1,2,1,3)
print(a)
print(a.shape)
b=a.squeeze()
print(b)
print(b.shape)

输出

tensor([[[[ 0.4250,  0.7191,  0.4334]],

         [[-1.4933, -0.5805, -0.4109]]]])
torch.Size([1, 2, 1, 3])
tensor([[ 0.4250,  0.7191,  0.4334],
        [-1.4933, -0.5805, -0.4109]])
torch.Size([2, 3])

是否可以制定去掉哪个维度

torch.squeeze(input, dim=None, *, out=None)

input是需要操作的tensor数据类型，dim指定需要去掉的维度

也可以是

input.squeeze(dim=None, *, out=None)

输入：

import torch
a=torch.randn(1,2,1,3)
print(a)
print(a.shape)
b=a.squeeze(0)
print(b)
print(b.shape)

输出

tensor([[[[ 0.0865,  1.7813, -0.2721]],

         [[ 0.2280,  0.7990,  0.5873]]]])
torch.Size([1, 2, 1, 3])
tensor([[[ 0.0865,  1.7813, -0.2721]],

        [[ 0.2280,  0.7990,  0.5873]]])
torch.Size([2, 1, 3])

如果指定去掉的维度不是1维，会怎样

输入

import torch
a=torch.randn(1,2,1,3)
print(a)
print(a.shape)
b=a.squeeze(1)
print(b)
print(b.shape)

输出 不会发生变化

tensor([[[[ 1.2664, -0.7684, -1.3079]],

         [[-0.5082,  2.2029,  0.1442]]]])
torch.Size([1, 2, 1, 3])
tensor([[[[ 1.2664, -0.7684, -1.3079]],

         [[-0.5082,  2.2029,  0.1442]]]])
torch.Size([1, 2, 1, 3])