Can I use the soft-dtw loss of tslearn?

[weidongzhou1994]

As title, I want to use the soft-dtw loss, but when i use it, it give me this error: TypeError: forward() missing 2 required positional arguments: 'x' and 'y'

[leoniewgnr]

Hi @weidongzhou1994, sorry for the long wait. NeuralProphet supports all loss functions that are listed here: https://pytorch-enhance.readthedocs.io/en/latest/_modules/torch/nn/modules/loss.html
I think soft-dtw is not listed, so I'm afraid it is not possible or how did you try to use?

[weidongzhou1994]

I use the soft-dtw of tslearn. So except loss functions that are listed in pytorch, can i write the loss function by myself.

[leoniewgnr]

Yes this should be possible: https://towardsdatascience.com/implementing-custom-loss-functions-in-pytorch-50739f9e0ee1
Let me know how it goes!:)