Fix explain_forecast and implement adaptive within that framework.

NorskRegnesentral / shapr

Explaining the output of machine learning models with more accurately estimated Shapley values

https://norskregnesentral.github.io/shapr/

Other

147 stars 34 forks source link

Fix explain_forecast and implement adaptive within that framework. #405

Closed jonlachmann closed 1 month ago

jonlachmann commented 1 month ago

Enable explain_forecast to comply with the adaptive/iterative estimation procedure.

martinju commented 1 month ago

I ran devtools::document() + added tests to be ran on GHA and pushed that. Not sure why it still does not run, though. Maybe due to a few merge conflicts?

Anyway, I go quite a few failing tests locally. Would be great if you could take a look at those. I think they should not be too hard to fix. Also, maybe some of them gets resolved when rebasing to shapr-1.0.0. Once that is done I'll look more closely at your code. Overall, it looks great though :-)

martinju commented 1 month ago

We agreed on the following

Deal with the slighly off printout for forecast later.
Jon fixes the boostrap code
Martin checks if there is an easy fix for shapley_reweighting [I just checked and the issues arrise in the boostrapping procedure, so I think this should be OK when we fix that using X_list-loops instead]
Add tests with adaptive = TRUE