dro 0.0.1

A package of distributionally robust optimization (DRO) methods. Implemented via cvxpy and PyTorch

DRO Package

Jiashuo Liu*, Tianyu Wang*, Peng Cui, Hongseok Namkoong

Tsinghua University, Columbia University

DRO is a python package that implements 12 typical DRO methods on linear models (SVM, logistic regression, and linear regression). It is built based on cvxpy. Implemented DRO methods include:

• $f$-DRO
  • CVaR-DRO
  • KL-DRO
  • TV-DRO
  • Marginal DRO (CVaR)
• Wasserstein DRO
  • Wasserstein DRO
  • Augmented Wasserstein DRO
  • Regularized Wasserstein DRO
• MMD-DRO
• Sinkhorn-DRO
• Holistic DRO
• Unified-DRO
  • $L_2$ cost
  • $L_{inf}$ cost

Current version only contains linear models. And further version will incorporate neural network implementations via some approximations.