grad can be implicitly created only for scalar outputs

grad can be implicitly created only for scalar outputs

自定义损失函数时报错：

      import torch
      import torch.nn as nn
      import numpy as np
      class CrossEntropyLoss(nn.Module):
      def __init__(self):
       super(CrossEntropyLoss, self).__init__()
      def forward(self, output, label):
       first = [-output[i][label[i]] for i in range(label.size()[0])]
       first_ = 0
      for i in range(len(first)):
       first_ += first[i]
       second = torch.exp(output)
       second = torch.sum(second, dim=1)
       second = torch.log(second + 1e-5)#**2
       second = torch.sum(second)
       loss = 1 / label.size()[0] * (first_ + second)
      return loss
      if __name__ == '__main__':
       output = torch.randn(3,3, 5, requires_grad=True)
       label = torch.empty((3,5), dtype=torch.long).random_(3)
      # x=0.9
      # output = to
  
 

文章来源: blog.csdn.net，作者：AI视觉网奇，版权归原作者所有，如需转载，请联系作者。

原文链接：blog.csdn.net/jacke121/article/details/118997850