Robbuffet

Conformal prediction + robust decision making with PyTorch predictors and CVXPY optimizers.

What it does - Split conformal calibration with geometry-aware scores (L2, L1, Linf, Mahalanobis). - Prediction regions (balls, ellipsoids, unions) that can be sampled, visualized, and used in downstream optimization. - Deterministic robust counterparts for affine uncertainty, scenario-based robust optimization, and gradient-based (Danskin) solvers for unions.

Install - pip install robbuffet (Python 3.10+). For development: pip install -e .[dev].

Getting started - Read the Quickstart. - Browse runnable Examples. - See API for core classes.