NLLLOSS(weight=)中weight解释

https://blog.csdn.net/weixin_37724529/article/details/107021786?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522162144406116780271573141%2522%252C%2522scm%2522%253A%252220140713.130102334.pc%255Fall.%2522%257D&request_id=162144406116780271573141&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~first_rank_v2~rank_v29-15-107021786.first_rank_v2_pc_rank_v29&utm_term=+weight+nllloss&spm=1018.2226.3001.4187

 

• weight：可选的，应该是一个tensor，里面的值对应类别的权重，如果样本不均衡的话，这个参数非常有用，长度是类别数目
• szie_average：默认是True，会将mini-batch的loss求平均值；否则就是把loss累加起来

import torch
import torch.nn as nn
 
a = torch.Tensor([[1,2,3]])
target = torch.Tensor([2]).long()
logsoftmax = nn.LogSoftmax()
ce = nn.CrossEntropyLoss()
nll = nn.NLLLoss()
 
#测试CrossEntropyLoss
cel = ce(a,target)
print(cel)
#输出：tensor(0.4076)
 
#测试LogSoftmax+NLLLoss
lsm_a = logsoftmax(a)
nll_lsm_a = nll(lsm_a,target)
#输出tensor(0.4076)

看来直接用nn.CrossEntropy和nn.LogSoftmax+nn.NLLLoss是一样的结果