Pytorch-区分nn.BCELoss()、nn.BCEWithLogitsLoss()和nn.CrossEntropyLoss() 的用法

详细理论部分可参考https://www.cnblogs.com/wanghui-garcia/p/10862733.html

BCELoss()和BCEWithLogitsLoss()的输出logits和目标labels的形状相同。

 1 import torch 
 2 import torch.nn as nn
 3 import torch.nn.functional as F
 4 
 5 m = nn.Sigmoid()
 6 
 7 loss_f1 = nn.BCELoss()  
 8 loss_f2 = nn.BCEWithLogitsLoss() 
 9 loss_f3 = nn.CrossEntropyLoss() 
10 
11 logits = torch.randn(3, 2)
12 labels = torch.FloatTensor([[0, 1], [1, 0], [1, 0]])
13 
14 print(loss_f1(m(logits), labels))      #tensor(0.9314)，注意logits先被激活函数作用 
15 print(loss_f2(logits, labels))         #tensor(0.9314)
16 
17 label2 = torch.LongTensor([1, 0, 0])
18 print(loss_f3(logits, label2))         #tensor(1.2842)

如果label2也想变成labels，然后通过BCELoss进行计算的话，可以先转变成独热编码的形式：

1 encode = F.one_hot(label2, num_classes = 2)               #encode的值和labels一样，但是类型是LongTensor
2 print(loss_f1(m(logits), encode.type(torch.float32)))     #tensor(0.9314)