Python中mask使用

1. Mask tensor can take 0 and 1 values only，mask中的内容只能是0或者是1
2. mask是一个 ByteTensor mask，作用是对原tensor中的内容进行遮罩，即要求出最后一层外其他的维度必须一样，例如：

 a=torch.tensor([[[5,5,5,5], [6,6,6,6], [7,7,7,7]], [[1,1,1,1],[2,2,2,2],[3,3,3,3]]])
 mask = torch.ByteTensor([[[1],[1],[0]],[[0],[1],[1]]])
 print('a.size()\n',a.size())
 print('mask.size()\n',mask.size())

输出：

a.size()
 torch.Size([2, 3, 4])
mask.size()
 torch.Size([2, 3, 1])

使用：

1. 指定为0的位置进行mask

代码

import torch
a=torch.tensor([[[5,5,5,5], [6,6,6,6], [7,7,7,7]], [[1,1,1,1],[2,2,2,2],[3,3,3,3]]])
print(a)
print(a.size())
print("#############################################3")
mask = torch.ByteTensor([[[1],[1],[0]],[[0],[1],[1]]])
print(mask.size())
b = a.masked_fill(mask==0, value=torch.tensor(-1e9))
print(b)
print(b.size())

输出：

tensor([[[5, 5, 5, 5],
         [6, 6, 6, 6],
         [7, 7, 7, 7]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]]])
torch.Size([2, 3, 4])
#############################################3
torch.Size([2, 3, 1])
tensor([[[          5,           5,           5,           5],
         [          6,           6,           6,           6],
         [-1000000000, -1000000000, -1000000000, -1000000000]],

        [[-1000000000, -1000000000, -1000000000, -1000000000],
         [          2,           2,           2,           2],
         [          3,           3,           3,           3]]])
torch.Size([2, 3, 4])

2. 默认为1的位置进行mask

代码：

import torch
a=torch.tensor([[[5,5,5,5], [6,6,6,6], [7,7,7,7]], [[1,1,1,1],[2,2,2,2],[3,3,3,3]]])
print(a)
print(a.size())
print("#############################################3")
mask = torch.ByteTensor([[[1],[1],[0]],[[0],[1],[1]]])
print(mask.size())
b = a.masked_fill(mask, value=torch.tensor(-1e9))
print(b)
print(b.size())

输出：

tensor([[[5, 5, 5, 5],
         [6, 6, 6, 6],
         [7, 7, 7, 7]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]]])
torch.Size([2, 3, 4])
#############################################3
torch.Size([2, 3, 1])
tensor([[[-1000000000, -1000000000, -1000000000, -1000000000],
         [-1000000000, -1000000000, -1000000000, -1000000000],
         [          7,           7,           7,           7]],

        [[          1,           1,           1,           1],
         [-1000000000, -1000000000, -1000000000, -1000000000],
         [-1000000000, -1000000000, -1000000000, -1000000000]]])
torch.Size([2, 3, 4])

参考连接：
https://zhuanlan.zhihu.com/p/151783950