RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #2 ‘target‘

1.问题

在运行PyTorch模型脚本训练分类网络时，出现如下报错：

RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #2 'target'

出错代码为：

loss_function = torch.nn.CrossEntropyLoss()
loss = loss_function (out, label)

 备注：CrossEntropyLoss的源码里“label”参数使用的dtype是“long”类型。

2.解决办法

修改的时候，直接在label后面加上.long()，如下所示：

loss_function = torch.nn.CrossEntropyLoss()
loss = loss_function (out, label.long())