Feature: Model Agnostic Explanability - Permutation Importance and Counterfactual

[rawanmahdi]

Feature Description

Not sure if this has been implemented in this repo before but I couldn't find anything on the docs. It would be nice to get some model agnostic feature importance for the autoML model. Counterfactual examples may also be useful to expose the model's behaviour and improve interpretability for the user.

Reason

Potential to improve model transperancy given these black box rapid prototyping autoML practices.

Solution

We can implement it using libraries like SHAP and DiCE, I have a good amount of experience working with them so I wouldn't mind contributing. Any suggestions in implementation would be much appreciated. Apologize if this has already been done.