Markus Loecher

regular spikes in loss (cats-and-dogs notebook)

1

I am struggling with explaining the regular spikes in both training and validation losses observed when running the cats-and-dogs CNN from the book, see image below:  There is an...

partial_dependence() computes conditional partial dependence

12

### Describe the bug For the case of correlated predictors (clearly highly common) the `sklearn.inspection.partial_dependence()` function gives different answers for `method` = "recursion" and `method` = "brute", see my [post](https://markusloecher.github.io/Partial-Dependence-Trees/)...

Saabas Values

1

Dear authors, thanks for a great package. The original python version by Lundberg offers the Boolean `approximate=True` parameter which will compute _Saabas scores_ instead of SHAP values. Is that a...

In Fig. 4D the tuned Random Forests sometimes perform below the vanilla RF. Did you ever look into this strange behavior?

In Fig. 4A/B the CCP (cost complexity pruned) graphs are puzzling to me: CCP should protect against overfitting a lot more than shown? In particular, the "number of leaves" cannot...

And the same question w.r.t. Figure 3, the code does not seem to be part of the repository ? (I suppose, it would be nice to also be able to...

I am looking for the code which produced Figure 4 in the "Hierarchical Shrinkage" paper. I had hoped that [01_performance_curves.ipynb](https://github.com/Yu-Group/imodels-experiments/blob/master/notebooks/shrinkage/01_performance_curves.ipynb) in the _shrinkage_ folder was the right notebook for this,...

SHAP and MDI computation

1

How would you compute MDI and/or SHAP values on the shrunken forests ? Might you have example code available ? Thanks! Markus

I have a comment regarding the Y notation: while you precisely define the potential outcomes $Y_{0i}$ and $Y_{1i}$ the symbols $Y_{0}$ and $Y_{1}$ show up without exact definition. Later you...