Interpretable ML chapter

[stineb]

Add Chapter 12: Interpretable machine learning

We already have material from ESDS here.

More contents can be found in Hands On Machine Learning in R.

@padasch could you create an additional chapter 12 based on the material from ESDS and evaluate whether there are contents from HOML that could/should be added to our AGDS chapter?

[padasch]

Since different models are introduced at different points in the book (ML in Chapter 8, KNN in Chapter 9, RF in Chapter 11), and since we mainly focus on regression tasks, I don't know how to align the new Chapter with the existing work.

Should the new chapter cover both model-agnostic and model-specific metrics? If so, should I move the already-covered metrics into this new chapter? Or should this new chapter be a separate model-agnostic evaluation chapter and then introduce model-specific metrics where the models are used?

So far, we have covered model evaluation for regression in 8.2.2.4 and classification in 8.3 (labelled as bonus material).

[stineb]

Good you're asking. I should probably be more specific in what I thought this could cover. This should not cover metrics that measure the goodness of fit. Rather, this should cover methods that extract information from fitted models, in particular the following methods:

• variable importance: vip::vip(), Boruta::Boruta() (take content from the DSM tutorial here you adopted)
• partial dependencies: pdp::partial() (and visreg::visreg() for lm()).

Covering these will be most important. We don't need to go much further. If we do

• Also, I've seen more and more Shapley values being calculated. See also in HOML here

These are all model-agnostic.

[padasch]

I added a short chapter with #146. Feedback is welcomed to improve. Based on what we discussed today, I kept it short to PDP and VIP. Did not look into {visreg} but can add this if needed.