regime-shap

Quantify how stable a tree model’s SHAP feature importance is across structural breaks in time series data.

regime-shap takes a fitted tree model, a feature matrix, and a set of regimes (distinct time periods such as a crisis, a policy change, or an economic era), and measures whether the features the model relies on stay the same across those regimes or reorganise. A model can score well overall while resting on unstable foundations, and this package makes that visible.

Install it with pip install regime-shap. The package follows Semantic Versioning; during the 0.x series the public API may still change.

Where to start

• Getting started: the fastest ways to try it, a browser demo, pip install, or the practice datasets.

• Installation: install from PyPI or from source.

• Quickstart: the one-call RegimeSHAPAnalyzer workflow.

• Concepts: the method in full, regimes, per-regime SHAP importance, Spearman stability, the Akoglu bands, and bootstrap confidence intervals.

• Examples: worked notebooks on public finance and energy data.

• API reference: the public functions and classes.