Numerical and visual tools to analyse the explainability of partial dependence functions.
Partial dependence plots are a popular tool to analyze black box machine learning models. The repo provides an R package that computes the explainability of a PDP, which is a measure to quantify how far a PDP is able to explain a model.
Supported functionalities:
- ...computation of explainability,
- ...matchplot of PDP vs. model predictions,
- ...computation of a forward variable selection based on explainability,
- ...visualization of 2D PDP vs. unexplained residual predictions,
- ...scatterplot matrix of 2D partial dependence plots.
Details are described in this paper. The examples are taken from the paper.