pytorch的HingeEmbeddingLoss 有先后順序

L1Loss  MSELoss 都沒問題，但是 HingeEmbeddingLoss 卻總報錯 說不能求梯度

 

#criterion = nn.MSELoss()
#criterion = nn.L1Loss()
criterion = nn.HingeEmbeddingLoss()

 

https://pytorch.org/docs/stable/generated/torch.nn.HingeEmbeddingLoss.html#torch.nn.HingeEmbeddingLoss

 

發現 Hinge損失函數還是區分了 

• Input: (*)(∗) where *∗ means, any number of dimensions. The sum operation operates over all the elements.

• Target: (*)(∗) , same shape as the input

• Output: scalar. If reduction is 'none', then same shape as the input

 

 保證模型輸出的變量在第一個就可以了，第二個是Label/Target

      loss =criterion(out, data.y)