Python的torch.tensor重构方法

涉及以下内容

• .softmax()
• .stack()，.cat()
• .view()，.permute()，.reshape()，.resize_()

实践

import torch
import torch.nn.functional as F


# 1，solfmax 计算
# 取决于dim,输入数据w×h,故0为列向,1为行向
x = torch.Tensor([[1, 2, 3, 4],
                  [1, 3, 4, 5],
                  [3, 4, 5, 6]]
                 )
x1 = F.softmax(x, dim=0)
x2 = F.softmax(x, dim=1)

print('\n', x, x.size(), x.shape, x.is_contiguous())
print('\n', x1, x2, sep='\n\n', end='\n\n')


# 2，列表中的多个张量转成一个张量
# .stack() 方法在列表中张量的基础上给列表顺序增加一个张量维度
tensor_list = [x1, x2]
x3 = torch.stack(tensor_list)
x4 = torch.cat(tensor_list)
print('\n', x3, x4, sep='\n\n', end='\n\n')

# 3，张量的形状改变
# tensor 调用过transpose,permute等方法的话会使该 tensor 在内存中变得不再连续
# 浅拷贝不为新变量开辟新的内存空间，仅拷贝id,深拷贝开辟新的内存空间
# .view() 方法只能改变内存连续的张量,否则需要先调用.contiguous()方法,改变形状的时候,总的数据个数不能变
# .reshape() 方法不受内存连续的张量限制,改变形状的时候,总的数据个数不能变
# .resize_() 不受以上两点限制,会直接修改原地址张量再返回,其他两种则是返回张量再修改它
# 以上方向变形过程都遵循，从左往右，从上往下，从外往里
x5 = x1.view(-1, 6)
x6 = x2.reshape(-1, 6)

print('\n', x5, x6, sep='\n\n', end='\n\n')

x7 = x.permute(1, 0)
x8 = x7
print(id(x), id(x7), id(x8))
print('\n', x, x7, x7.is_contiguous(), x8.resize_(2, 4), sep='\n\n', end='\n\n')


'''——————————————————————————————output——————————————————————————————'''
1：
tensor([[1., 2., 3., 4.],
        [1., 3., 4., 5.],
        [3., 4., 5., 6.]]) torch.Size([3, 4]) torch.Size([3, 4]) True

tensor([[0.1065, 0.0900, 0.0900, 0.0900],
        [0.1065, 0.2447, 0.2447, 0.2447],
        [0.7870, 0.6652, 0.6652, 0.6652]])

tensor([[0.0321, 0.0871, 0.2369, 0.6439],
        [0.0120, 0.0889, 0.2418, 0.6572],
        [0.0321, 0.0871, 0.2369, 0.6439]])


2：
tensor([[[0.1065, 0.0900, 0.0900, 0.0900],
         [0.1065, 0.2447, 0.2447, 0.2447],
         [0.7870, 0.6652, 0.6652, 0.6652]],

        [[0.0321, 0.0871, 0.2369, 0.6439],
         [0.0120, 0.0889, 0.2418, 0.6572],
         [0.0321, 0.0871, 0.2369, 0.6439]]])

tensor([[0.1065, 0.0900, 0.0900, 0.0900],
        [0.1065, 0.2447, 0.2447, 0.2447],
        [0.7870, 0.6652, 0.6652, 0.6652],
        [0.0321, 0.0871, 0.2369, 0.6439],
        [0.0120, 0.0889, 0.2418, 0.6572],
        [0.0321, 0.0871, 0.2369, 0.6439]])


3：
tensor([[0.1065, 0.0900, 0.0900, 0.0900, 0.1065, 0.2447],
        [0.2447, 0.2447, 0.7870, 0.6652, 0.6652, 0.6652]])

tensor([[0.0321, 0.0871, 0.2369, 0.6439, 0.0120, 0.0889],
        [0.2418, 0.6572, 0.0321, 0.0871, 0.2369, 0.6439]])

2427424505920 2427424493120 2427424493120

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

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

False

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