Pytorch MSELoss

CLASS `torch.nn.MSELoss`(size_average=None, reduce=None, reduction='mean')

Creates a criterion that measures the mean squared error (squared L2 norm) between each element in the input

创建一个准则测量输入x和目标y每个元素间的平方二范数。

x和y可以是任意形状的张量，每个张量总的元素个数为n。

Note: `size_average` and `reduce` are in the process of being deprecated, and in the meantime, specifying either of those two args will override `reduction`. Default: `'mean'`

size_average和reduce正在被弃用，因此我们只看reduction参数。

reduction='none'

The unreduced (i.e. with reduction set to 'none') loss can be described as:

返回每个元素的平方二范数，shape与输入相同。

reduction='sum'和reduction='mean'

对所有对应元素的平方二范数求和或求平均（sum / n）。

 1 a = torch.tensor([[[1., 2.]], [[1, 2]]])  # [2, 1, 2]
 2 b = torch.tensor([[[2., 4.]], [[2, 4.]]]) # [2, 1, 2]
 3 
 4 C =  torch.nn.MSELoss(reduction='none')
 5 
 6 loss = C(a, b)
 7 
 8 print(a)
 9 print(b)
10 print(loss)
11 
12 >>>
13 
14 tensor([[[1., 2.]],
15         [[1., 2.]]])
16 
17 tensor([[[2., 4.]],
18         [[2., 4.]]])
19 
20 tensor([[[1., 4.]],
21         [[1., 4.]]])  # 分别计算对应元素的平方二范数
22 
23 C =  torch.nn.MSELoss(reduction='sum')
24 
25 loss = C(a, b)
26 print(loss)
27 
28 >>> tensor(10.) # 所有元素平方二范数之和
29 
30 C =  torch.nn.MSELoss(reduction='mean')
31 
32 loss = C(a, b)
33 print(loss)
34 >>> tensor(2.5000) # 所有元素平方二范数的均值