博客摘录「 关于pytorch中的CrossEntropyLoss（）的理解」2023年8月23日

target = torch.empty(3, dtype = torch.long).random_(5)
在0~5范围内随机生成3个数，作为分类

二分类交叉熵

多分类交叉熵

 

 first[]和second[]用来计算softmax函数

import torch
import torch.nn as nn
import numpy as np
import math
#我们这次计算有三个待预测像素的点的loss值
loss = nn.CrossEntropyLoss()
input = torch.randn(3, 5, requires_grad= True)
target = torch.empty(3, dtype = torch.long).random_(5)
output = loss(input, target)
print('input is ', input)
print('target is ', target)
print('loss=', output)
#自己跟据公式计算loss
first = []
for i in range(3):
    val = input[i][target[i]]
    first.append(val)
# print(first)
second = []
sum = 0
for i in range(3):
    for j in range(5):
        sum += math.exp(input[i][j])
    second.append(sum)
    sum = 0
# print(second)
L = []
for i in range(3):
    l = -first[i] + math.log(second[i])
    L.append(l)
# print(L)

Loss = 0
for i in range(3):
    Loss += L[i]
Loss = Loss / 3 #需要除以3
print('自己计算的loss', Loss)