Pytorch学习笔记(5)——交叉熵报错RuntimeError: 1D target tensor expected, multi-target not supported

当我使用交叉熵做损失函数时,发生了报错:

RuntimeError: 1D target tensor expected, multi-target not supported

我查了相关资料,里面的说法基本都是:

  1. 输入labels维度应该为1维,且精度不能是Double,必须换成long;
  2. 对输入标签进行降维。

但是却没法解决我的问题,因为我的标签数据在处理好后,用以下代码处理过:

torch.LongTensor(labels)

而且我也打印过我的标签数据的维度:

torch.Size([16, 11])

这里16指的是batch_size,所以也不是维度的问题。

但是我在看这篇博客(RuntimeError: multi-target not supported at)的时候给了我灵感。里面写的是:

pytorch 中计计算交叉熵损失函数时, 输入的正确 label 不能是 one-hot 格式。函数内部会自己处理成 one hot 格式。所以不需要输入 [ 0 0 0 0 1],只需要输入 4 就行。

而我的标签数据是多标签问题,如下:

tensor([0, 1, 0, 0, 1, 0, 0, 1, 1, 0, 0])

那么在经过loss的时候,CrossEntropyLoss会自动为其编码为one-hot编码,这样就会导致其升高一维,变为了:

tensor([[1., 0.],
        [0., 1.],
        [1., 0.],
        [1., 0.],
        [0., 1.],
        [1., 0.],
        [1., 0.],
        [0., 1.],
        [0., 1.],
        [1., 0.],
        [1., 0.]])

于是就导致了该错误。

所以解决方案就是采用多标签问题的损失函数。比如MultiLabelSoftMarginLoss,或者我采用的最原始的MSELoss

参考

[1] 菜鸡小王的技术之路. RuntimeError: multi-target not supported at[EB/OL]. (2019-12-10)[2021-10-27]. https://www.cnblogs.com/blogwangwang/p/12018897.html
[2] Python Free. 交叉熵损失函数中“一维目标张量期望,多目标不支持”的求解,算,lossFunction,报错,1Dtargettensorexpectedmultitargetnotsupported,解决办法[EB/OL]. (2020-07-04)[2021-10-27]. https://www.pythonf.cn/read/125399

你可能感兴趣的:(PyTorch,学习经验,pytorch,交叉熵,报错)