Sunnies 😎

Shapley values that uncover nonlinear dependencies

Herein lie code and results for the paper Explaining the data or explaining a model? Shapley values that uncover non-linear dependencies by Daniel Fryer, Inga Strumke and Hien Nguyen.

Guide

• For the R script that generates the majority of the results, see here. The other R files are dependencies of that script.
• For the Python script that produces SAGE and SHAP results for the dataset drift example, see here.
• The shapley function here can be used to calculate Shapley values given a data set and utility function (e.g., choosing a measure of non-linear dependence as the utility function will produce Sunnies values). Some utility functions can be found here.
• Run the Python script here to generate the violin plot in Figure 2.
• Other Python files in the Python directory are dependencies of the violin plot script, and can also be used to calculate Shapley and Sunnies values for any data generating process or data set.

Resources

• shapr https://norskregnesentral.github.io/shapr/articles/understanding_shapr.html#advanced
• SHAP Python explaining loss https://shap.readthedocs.io/en/latest/example_notebooks/tabular_examples/tree_based_models/Explaining%20the%20Loss%20of%20a%20Model.html
• SHAP Python tree based models https://shap.readthedocs.io/en/latest/tabular_examples.html#tree-based-models