Hi guys, about the feature importance for the NHITS model, is this on the roadmap(or maybe already there but I missed it), and how does the new model TimeXer compares to NHITS

You can do decomposition with the NHITS model to understand its components as detailed here. By feature importance, do you mean SHAP values? If that's the case, then including SHAP values for all models in on the roadmap. We haven't benchmarked TimeXer against NHITS yet, but both models are in neuralforecast, and you can fit both of them and compare their performance!

Thanks for the reply! Yes I meant SHAP values, okay that would be a very interesting tool for the modelling. I will try TimeXer out !

Hi @Marco. Do you have any update on including SHAP values for neural forecast?

Not yet! It's hard for me to give a timeline as we work on other priorities. I'll try to at least create a tutorial on that before implementing a full-fledged functionality in neuralforecast

It would be very helpful thanks!