torch.nn.BCELoss用法

1. 定義
　　數學公式為 Loss = -w * [p * log(q) + (1-p) * log(1-q)] ，其中p、q分別為理論標簽、實際預測值，w為權重。這里的log對應數學上的ln。

　　PyTorch對應函數為：
　　　　torch.nn.BCELoss(weight=None, size_average=None, reduce=None, reduction=‘mean’)
　　計算目標值和預測值之間的二進制交叉熵損失函數。

　　有四個可選參數：weight、size_average、reduce、reduction

1. weight必須和target的shape一致，默認為none。定義BCELoss的時候指定即可。
2. 默認情況下 nn.BCELoss()，reduce = True，size_average = True。
3. 如果reduce為False，size_average不起作用，返回向量形式的loss。
4. 如果reduce為True，size_average為True，返回loss的均值，即loss.mean()。
5. 如果reduce為True，size_average為False，返回loss的和，即loss.sum()。
6. 如果reduction = ‘none’，直接返回向量形式的 loss。
7. 如果reduction = ‘sum’，返回loss之和。
8. 如果reduction = ''elementwise_mean，返回loss的平均值。
9. 如果reduction = ''mean，返回loss的平均值

2. 驗證代碼

1>

import torch import torch.nn as nn m = nn.Sigmoid() loss = nn.BCELoss(size_average=False, reduce=False) input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("計算loss的結果:") print(output)

 

 

 2>

import torch import torch.nn as nn m = nn.Sigmoid() loss = nn.BCELoss(size_average=True, reduce=False) input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("計算loss的結果:") print(output)

 

 

 3>

import torch import torch.nn as nn m = nn.Sigmoid() loss = nn.BCELoss(size_average=True, reduce=True) input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("計算loss的結果:") print(output)

 

 

 4>

import torch import torch.nn as nn m = nn.Sigmoid() loss = nn.BCELoss(size_average=False, reduce=True) input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("計算loss的結果:") print(output)

 

 

 5>

import torch import torch.nn as nn m = nn.Sigmoid() loss = nn.BCELoss(reduction = 'none') input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("計算loss的結果:") print(output)

 

 6>

import torch import torch.nn as nn m = nn.Sigmoid() weights=torch.randn(3) loss = nn.BCELoss(weight=weights,size_average=False, reduce=False) input = torch.randn(3, requires_grad=True) target = torch.empty(3).random_(2) lossinput = m(input) output = loss(lossinput, target) print("輸入值:") print(lossinput) print("輸出的目標值:") print(target) print("權重值") print(weights) print("計算loss的結果:") print(output)

 

 

 

 

2. 驗證代碼
1>
import torchimport torch.nn as nn
m = nn.Sigmoid()
loss = nn.BCELoss(size_average=False, reduce=False)input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("計算loss的結果:")print(output)1234567891011121314151617
2>
import torchimport torch.nn as nn
m = nn.Sigmoid()
loss = nn.BCELoss(size_average=True, reduce=False)input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("計算loss的結果:")print(output)1234567891011121314151617
3>
import torchimport torch.nn as nn
m = nn.Sigmoid()
loss = nn.BCELoss(size_average=True, reduce=True)input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("計算loss的結果:")print(output)1234567891011121314151617
4>
import torchimport torch.nn as nn
m = nn.Sigmoid()
loss = nn.BCELoss(size_average=False, reduce=True)input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("計算loss的結果:")print(output)1234567891011121314151617
5>
import torchimport torch.nn as nn
m = nn.Sigmoid()
loss = nn.BCELoss(reduction = 'none')input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("計算loss的結果:")print(output)1234567891011121314151617
6>
import torchimport torch.nn as nn
m = nn.Sigmoid()weights=torch.randn(3)
loss = nn.BCELoss(weight=weights,size_average=False, reduce=False)input = torch.randn(3, requires_grad=True)target = torch.empty(3).random_(2)lossinput = m(input)output = loss(lossinput, target)
print("輸入值:")print(lossinput)print("輸出的目標值:")print(target)print("權重值")print(weights)print("計算loss的結果:")print(output)1234567891011121314151617181920
————————————————版權聲明：本文為CSDN博主「qq_29631521」的原創文章，遵循CC 4.0 BY-SA版權協議，轉載請附上原文出處鏈接及本聲明。原文鏈接：https://blog.csdn.net/qq_29631521/article/details/104907401