Linear Models

These optimization problems are solved exactly (or approximately, e.g., kernel) through solvers.

Each method below links to its conceptual guide and literature references, its estimator API documentation (where available), and a runnable notebook.

\(f\)-divergence DRO

See the $f$-divergence DRO guide and references for the standard and partial-shift formulations. Estimator references are available for KL-DRO, chi-square DRO, CVaR-DRO, TV-DRO, conditional CVaR-DRO, and marginal DRO. See the $f$-divergence DRO notebook for runnable standard and partial-shift linear examples.

Wasserstein DRO

See the Wasserstein DRO guide and references for the standard and robust-satisficing formulations, the Wasserstein DRO API reference for estimator details, and the Wasserstein DRO notebook for linear classification, regression, and robust-satisficing examples.

Group DRO

See the Group DRO guide and references for the formulation, supported losses, and implementation details. The Group DRO notebook provides an end-to-end linear and neural example.

MMD-DRO

See the MMD-DRO guide and references for the formulation and kernel guidance, the MMD-DRO API reference for estimator details, and the kernel DRO notebook for a runnable example and ambiguity-set kernel selection guidance.

Bayesian (Parametric) DRO

See the Bayesian DRO guide and references for the frequentist and Bayesian formulations, the Bayesian DRO API reference for estimator details, and the Bayesian DRO notebook for runnable frequentist and Bayesian parametric examples.

Mixed-distance DRO

See the mixed-distance DRO guide and references for the Sinkhorn, MOT, outlier-robust Wasserstein, and Holistic DRO formulations. Estimator references are available for Sinkhorn DRO, MOT DRO, outlier-robust Wasserstein DRO, and Holistic Robust DRO. See the mixed-distance DRO notebook for runnable examples of all four methods.