Pytorch CrossEntropyLoss调错记录

1、dimension out of range (expected to be in range of [-1, 0], but got 1)

参考https://github.com/pytorch/pytorch/issues/5554

二分类问题，target.size         [batch_size,]

修改，output_class.size()   为[batch_size,2]

 

2、RuntimeError: multi-target not supported at  ...

target.size =[batch_size,1]出错

参考：https://blog.csdn.net/ccbrid/article/details/79844005

https://blog.csdn.net/york1996/article/details/81875508

修改，target=target.squeeze()

 

3、element 0 of tensors does not require grad and does not have a grad_fn

参考：https://blog.csdn.net/jacke121/article/details/82733197

element 0 of tensors does not require grad and does not have a grad_fn

这个是因为requires_grad=False，应该为true
x = Variable(torch.ones(2,2),requires_grad=False)

4、CrossEntropyLoss（output，target）

output type : torch.FloatTorch

target type : torch.LongTorch

output_class = Variable(output_class.type(torch.FloatTensor)

label = Variable(label.type(torch.LongTensor))