midr - v0.5.0: Initial CRAN Release

This is the first stable release of midr, a new R package for interpreting black-box machine learning models (IML / XAI).

midr helps you understand and explain the behavior of complex black-box models, like GBDT, NN, etc., by creating globally interpretable surrogate models based on a method called Maximum Interpretation Decomposition (MID).

Key features include: - Decomposing model predictions into main effects and second-order interaction effects. - A comprehensive suite of graphics- or ggplot2-based functions to visualize the decomposition from multiple perspectives (effect plots, importance heatmaps, ceteris paribus curves, and prediction breakdowns). - A fast and exact method to calculate the model-specific Shapley values for the surrogate model, with seamless integration with the shapviz package.

CRAN: https://cran.r-project.org/web/packages/midr/index.html Paper: https://arxiv.org/abs/2506.08338 Online Documentation: https://ryo-asashi.github.io/midr/

- R
Published by ryo-asashi 8 months ago