pytorch中unsqueeze用法说明

在指定的位置插入一个维度，有两个参数，input是输入的tensor，dim是要插到的维度

需要注意的是dim的范围是[-input.dim()-1, input.dim()+1)，是一个左闭右开的区间，当dim为负值时，会自动转换为dim = dim+input.dim()+1，类似于使用负数对python列表进行切片。

import torcha = torch.randn(2,5)
print(a)print("")
b = a.unsqueeze(0)
print(b.shape)print("")
c = a.unsqueeze(a.dim())
print(c.shape)输出：
tensor([[-0.4734,  0.4115, -0.9415, -1.1280, -0.1065],[ 0.1613,  1.2594,  1.1261,  1.3881,  0.1112]])torch.Size([1, 2, 5])torch.Size([2, 5, 1])

以上是二维数据情况：

首先生成了一个二维矩阵，其大小为[2，5]

然后，在0维度上插入一个维度，可以看到现在新矩阵a的形状变为[1，2，5]，第0维度的大小默认是1

最后，在最后一个维度上插入一个维度，形状变为[2, 5, 1]

a=torch.rand(2,3,2)print("")
print("torch.unsqueeze(a,3) size: {}".format(torch.unsqueeze(a,3).size()))print("")
print("torch.unsqueeze(a,2) size: {}".format(torch.unsqueeze(a,2).size()))print("")
print("torch.unsqueeze(a,1) size: {}".format(torch.unsqueeze(a,1).size()))print("")
print("torch.unsqueeze(a,0) size: {}".format(torch.unsqueeze(a,0).size()))print("")
print("torch.unsqueeze(a,-1) size: {}".format(torch.unsqueeze(a,-1).size()))print("")
print("torch.unsqueeze(a,-2) size: {}".format(torch.unsqueeze(a,-2).size()))print("")
print("torch.unsqueeze(a,-3) size: {}".format(torch.unsqueeze(a,-3).size()))print("")
print("torch.unsqueeze(a,-4) size: {}".format(torch.unsqueeze(a,-4).size()))输出：
torch.unsqueeze(a,3) size: torch.Size([2, 3, 2, 1])torch.unsqueeze(a,2) size: torch.Size([2, 3, 1, 2])torch.unsqueeze(a,1) size: torch.Size([2, 1, 3, 2])torch.unsqueeze(a,0) size: torch.Size([1, 2, 3, 2])torch.unsqueeze(a,-1) size: torch.Size([2, 3, 2, 1])torch.unsqueeze(a,-2) size: torch.Size([2, 3, 1, 2])torch.unsqueeze(a,-3) size: torch.Size([2, 1, 3, 2])torch.unsqueeze(a,-4) size: torch.Size([1, 2, 3, 2])