pytorch -- torch.nn下的常用损失函数

1.基础

loss function损失函数：预测输出与实际输出 差距 越小越好

- 计算实际输出和目标之间的差距
- 为我们更新输出提供依据（反向传播）

1. L1
torch.nn.L1Loss(size_average=None, reduce=None, reduction=‘mean’)
2. 平方差（L2）
torch.nn.MSELoss(size_average=None, reduce=None, reduction=‘mean’)
3. 交叉熵
torch.nn.CrossEntropyLoss(weight=None, size_average=None, ignore_index=-100, reduce=None, reduction=‘mean’, label_smoothing=0.0)

2. 例子

代码：

import torch
from torch import nn

input = torch.tensor([1,2,3],dtype=torch.float32)
input = torch.reshape(input,[1,1,1,3])
target = torch.tensor([1,2,5],dtype=torch.float32)
target = torch.reshape(target,[1,1,1,3])
# L1
l1 = nn.L1Loss(reduction='sum')
result1 = l1(input,target)
print(result1)
# L2
l2 = nn.MSELoss()
result2 = l2(input,target)
print(result2)

# 交叉熵损失
x = torch.tensor([0.1,0.2,0.3])
y = torch.tensor([1])
x = torch.reshape(x,[1,3])
loss_cross = nn.CrossEntropyLoss()
result = loss_cross(x,y)
print(result)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

输出