Shapley

Shapley value utilities in Julia.

This repository is going to be confusing for anyone snooping around. Come back later and it might be cleaned up.

Methods

Shapley(team, game, removal)

Abstract types

• Team (for defining players) - supertype of (grouped) data set, ...
• Removal (for subsetting the team) - supertype of column select, marginalise, ...
• Game (for characteristic function) - supertype of model, dc, hsic, ...

Goals

Extensible and modular, e.g.:

• Ease of implementing similar Shapley value alternatives, e.g., asymmetric Shapley - but this suggests a more general package / structure
• Ease of implementing approximation techniques
• Ease of implementing alternative calculation methods, e.g., Harsanyi dividends

Long term goals

• Integration with popular model outputs including LightGBM and XGBoost
• Plot Shapley values over time, with e.g., ADL, ADP and ADR