PYTORCH 损失函数 (EXPECTED TO BE IN RANGE OF [-1, 0], BUT GOT 1)-检查batchsize

最新推荐文章于 2024-03-27 12:00:45 发布

FakeOccupational

最新推荐文章于 2024-03-27 12:00:45 发布

阅读量355

点赞数

分类专栏：语言学习笔记

本文链接：https://blog.csdn.net/ResumeProject/article/details/119633067

版权

语言学习笔记专栏收录该内容

183 篇文章 4 订阅

订阅专栏

一般情况下的分类问题计算交叉熵损失

import torch.nn as nn
import torch
func=nn.CrossEntropyLoss()
a=torch.Tensor([[ 0.0606,0.1610,0.2990,0.2101, 0.5104],
                [0.6388,0.4053, 0.4196, 0.7060, 0.2793],
                [ 0.3973,0.6114, 0.1127, 0.7732, 0.0592]])
b=[3,1,0]
b=torch.Tensor(b)
loss=func(a,b.long())
loss=func(a,b.long())
print("总loss:",loss)

以上是batchsize为3的情况，但是batchsize为1(如下)会报错：

a1=torch.Tensor([ 0.0606,0.1610,0.2990,0.2101, 0.5104])
a2=torch.Tensor([0.6388,0.4053, 0.4196, 0.7060, 0.2793])
a3=torch.Tensor([ 0.3973,0.6114, 0.1127, 0.7732, 0.0592])

b1=torch.Tensor([3])
b2=torch.Tensor([1])
b3=torch.Tensor([0])

loss_1=func(a1,b1.long())
loss_2=func(a2,b2.long())
loss_3=func(a3,b3.long())

原因是a.type: torch.Size([3, 5])的大小分出来是a1.type: torch.Size([5])，需要维度为[1，5]，解决方法：

a1=torch.unsqueeze(a1,0)
a2=torch.unsqueeze(a2,0)
a3=torch.unsqueeze(a3,0)

print("loss1:",loss_1)
print("loss2:",loss_2)
print("loss3",loss_3)
loss_sum=loss_1+loss_2+loss_3
print("loss_sum",loss_sum/3)