AutoSHAP is a library still in development by the Elo7 data science and search teams.
AutoSHAP is a simple framework built on top of SHAP to make it simple to engineers access explanations in a daily basis setting just a few parameters. It has two main classes, FullPipeline and SHAPViews. The first runs the entire pipeline depicted in image bellow, i.e., it instances a XGBoost model, makes hyperparameter optimization, computes the Shapley values with the best model, saves the XGBoost model and data and make plots as a result.
The second class, SHAPViews, allows one to see interactive plots of features and Shapley values. This is specially useful to analyze the distribution of some feature values while filtering them by some chosen Shapley Values.
Clone this repository,
git clone [email protected]:elo7/AutoSHAP.git
and inside it use pip to install:
pip install .
You can check application examples in the notebooks