shapiq
shapiq copied to clipboard
mmschlk
Shapley Interactions for Machine Learning
there are a few todos to be discussed f2f before the implementation notes: 1. `ConditionalImputer` requires `x` (as opposed to `MarginalImputer`), because `value_function` uses a distribution conditioned on `x` 2....
Next to interventional approaches also observational explanations should be created with shapiq. For this conditional modelling and sampling of and from the tabular data distributions is required. This [arfpy package](https://arxiv.org/abs/2311.07366)...
Add the structured sampling SV estimator. ## Tasks - [ ] Add the approximator - [ ] Add tests - [ ] Add the documentation and the correct reference to...
Add validator to convert CatBoost Trees to internal model.
Currently, all methods are dependent only on the number of players n. It would be useful to be able to work with individual player names. Applications could be to compute...
## Description The `InteractionValues` dataclass is the core data object containing the results and of approximators and explainers. For representing Shapley values (SVs) visually, the _shap_ package already contains a...
Currently, not the whole API is visible in the docs. Not all modules can be found. We need to add this before the initial release. For example there are no...
We need to be able to naturally support computing interactions for XGBoost Type of models. This needs to be done in the conversion mechanism where tree-models are transformed into their...
We need to support LightGBM naturally with TreeSHAP-IQ. For this we need to have a conversion mechanism. This happens in the conversion into the edge representation of the tree models....