This message was deleted.

Hey. You need to define your model as a scikit-learn estimator (reference). One easy way of achieving this may be through skorch, I haven't used it but it seems to add the fit and predict methods, which are the ones needed by mlforecast. That could also help you saving the training loss with their callbacks. Let us know if you need any help setting this up

I see; thank you very much! Is there a way to extract train loss from the current models in MLforecast, or do I need to do something similar for those as well?

You can get the predictions on the training set after the training has finished with this

Got it, thank you!