【笔记】nn.CrossEntropyLoss（）:可以看做先激活，再取对数

一维：

import torch
import torch.nn as nn
import math

criterion = nn.CrossEntropyLoss()
output = torch.randn(1, 5, requires_grad=True)
label = torch.empty(1, dtype=torch.long).random_(5)
loss = criterion(output, label)

print("网络输出为5类:")
print(output)
print("要计算label的类别:")
print(label)
print("计算loss的结果:")
print(loss)

first = 0
for i in range(1):
    first = -output[i][label[i]]
second = 0
for i in range(1):
    for j in range(5):
        second += math.exp(output[i][j])
res = 0
res = (first + math.log(second))
print("自己的计算结果：")
print(res)

多维:

import torch
import torch.nn as nn
import math
criterion = nn.CrossEntropyLoss()
output = torch.randn(3, 5, requires_grad=True)
label = torch.empty(3, dtype=torch.long).random_(5)
loss = criterion(output, label)

print("网络输出为3个5类:")
print(output)
print("要计算loss的类别:")
print(label)
print("计算loss的结果:")
print(loss)

first = [0, 0, 0]
for i in range(3):
    first[i] = -output[i][label[i]]
second = [0, 0, 0]
for i in range(3):
    for j in range(5):
        second[i] += math.exp(output[i][j])
res = 0
for i in range(3):
    res += (first[i] + math.log(second[i]))
print("自己的计算结果：")
print(res/3)

 

eg1:

import torch
import torch.nn as nn
import math
criterion = nn.CrossEntropyLoss()
# output = torch.randn(3, 5, requires_grad=True)
# label = torch.empty(3, dtype=torch.long).random_(5)

output=torch.tensor([[ 0.4368,  2.2467,  0.3081, -2.0271,  0.4074],
        [ 0.3731,  0.8674, -0.5239, -0.3642, -1.4294],
        [-1.9317, -0.4155,  0.1015, -0.7857,  0.4398]])
label=torch.tensor([ 2,  1,  4])

loss = criterion(output, label)
print(loss)


first = [0, 0, 0]
for i in range(3):
    first[i] = -output[i][label[i]]
print(first)

second = [0, 0, 0]
for i in range(3):
    for j in range(5):
        second[i] += math.exp(output[i][j])
res = 0
for i in range(3):
    res += (first[i] + math.log(second[i]))
print("自己的计算结果：")
print(res/3)

label = torch.empty(3, dtype=torch.long).random_(5)
print(label)

eg2:

import numpy as np
from torch import nn
import torch as t
import math
bbb=t.tensor([[ 0.1736,  0.5441,  0.2823],
        ])
ccc=t.tensor([0])
aaa=nn.CrossEntropyLoss()
print(bbb,ccc)
print(aaa(bbb,ccc))


print(-bbb[0][0]+np.log((np.exp(0.1736)+np.exp(0.5441)+np.exp(0.2823))))

# print(np.exp(0.1736))
# print(np.exp(0.5441))
# print(np.exp(0.2823))

eg3:

import numpy as np
from torch import nn
import torch as t
import math
bbb=t.tensor([[ 0.1736,  0.5441,  0.2823],
        ])
ccc=t.tensor([0])
aaa=nn.CrossEntropyLoss()
print(bbb,ccc)
print(aaa(bbb,ccc))


print(-np.log(np.exp(bbb[0][0])/((np.exp(0.1736)+np.exp(0.5441)+np.exp(0.2823)))))

# print(np.exp(0.1736))
# print(np.exp(0.5441))
# print(np.exp(0.2823))

eg4:

loss(predictions,labels)
tensor(1.1652)
labels
tensor([ 2,  1])
predictions
tensor([[ 0.2928,  0.1549,  0.5523],
        [ 0.7676,  0.0461,  0.1863]])
-np.log(np.exp(0.0461)/(np.exp(0.7676)+np.exp(0.0461)+np.exp(0.1863)))
1.4369924776751017
-np.log(np.exp(0.5523)/(np.exp(0.2928)+np.exp(0.1549)+np.exp(0.5523)))
0.8934324010163079
1.43+0.89
2.32
2.32/2
1.16