Description

We should support the python version of CatBoost in our TreeExplainer.

To do so, we need to write a conversion mechanism akin to our other converters, which can be found here and incorporate it in our validation step. The converters transform the tree structures inside of the original API's structure into our own TreeModel or list[TreeModel] which is then used inside the TreeExplainer.

For reference, one can take a look at our LightGBM transform or into shap's transformation using the CatBoostTreeModelLoader which is getting called here.

Important for this is that we do not run into import errors when users do not have CatBoost installed. This means we do not have import catboost lying around in files that get imported when they should not get imported. To fix this, either add the import statements inside function calls that only get run where it's supposed to.

Tasks

To close this issue we would need the following then:

• 1. Implement CatBoost conversion logic.
• 2. Add tests similar to our ExtraTrees/LightGBM/XGboost/DecisionTree tests, which compare our output for SV with SHAP's results.
• 3. Adjust existing documentation if necessary
• 4. Add CatBoost to pyproject.toml's optional dependencies list.
• 5. Optional: It would be nice to have a very small notebook showing that CatBoost for some ML data/model can be explained similar to our current notebooks and in particular the LightGBM example (can also be quite short).