This repository contains sample code on a few explainability approaches.
The data used is from the House Prices: Advanced Regressions Techniques from Kaggle (available here: https://www.kaggle.com/c/house-prices-advanced-regression-techniques )
The Jypyter notebook loads and does basic pre-processing of the data. Then I implement/test the following approaches, stating their advantages and pitfalls:
- Partial dependency plots (PDPs)
- Individual Conditonal Expectations(ICE)
- Feature importances
- Global Surrogate model
- LIME
- Shapley
- Anchors
To be implemented:
-
Data Shapley
This is still work in progress, and new approaches are invented quite frequently, so I will try to keep up with them.