loss_class = nn.NLLLoss()
loss_class_test = loss_class(class_prd,labels)
报错:RuntimeError: 0D or 1D target tensor expected, multi-target not supported
print class_prd.shape, labels.shape 发现分别为
torch.Size([64, 3]) #三分类问题
torch.Size([64, 1])
labels.shape维度应为[64]
解决:loss_class_test = loss_class(class_prd,labels.squeeze(dim=1))