Calibrated Explanations comparisons (CE-first)

All code examples in this repo use WrapCalibratedExplainer.

vs SHAP

• ✅ Endpoint envelopes describing how features shift predictive boundaries

• ✅ Calibrated predictions included

• ⚠️ Requires held-out calibration set for guarantees (SHAP does not)

vs LIME

• ✅ Built-in uncertainty quantification and more stable rules

• ⚠️ Needs calibration set

vs Conformal Prediction packages

• ✅ Provides both calibrated predictions and feature-level uncertainty / explanations

• ✅ Builds on conformal methods (Venn-Abers, CPS)

• ⚠️ Conformal focuses on prediction sets or intervals