Code for Kernel Distributionally Robust Optimization

Authors

Jia-Jie Zhu, Wittawat Jitkrittum

Citing this repository

@misc{zhu2020kernel,
      title={Kernel Distributionally Robust Optimization}, 
      author={Jia-Jie Zhu and Wittawat Jitkrittum and Moritz Diehl and Bernhard Schölkopf},
      year={2020},
      eprint={2006.06981},
      archivePrefix={arXiv},
      primaryClass={math.OC}
}

Instruction

kdro is a folder for the Python module kdro. The repository contains two experiments in the KDRO paper:

Robust least squares
Distributionally robust classification using SFG-DRO

The executable files for those two experiments are located in the examples/ folder. See the README files therein. Please first follow the instructions below to set up the environment first.

Dependency

dill
numpy
scipy
cvxpy (see https://www.cvxpy.org/install/)
Mosek solver
- optionally, install the solvers of your choice.
cvxopt
jupyter
matplotlib
sklearn

For testing SFG-DRO, you need

pytorch
torchvision

Development

To install the package for development purpose, follow the following steps:

Make a new Anaconda environment (if you use Anaconda. Recommended) for this project. Switch to this environment.
cd to the folder that contains this READMD.md file.
Issue the following command in a terminal to install the kdro Python package from this repository.
```
 pip install -e .
```
This will install the package to your environment selected in Step 1.
In a Python shell with the environment activated, make sure that you can import kdro without any error.

The -e flag offers an "edit mode", meaning that changes to any files in this repo will be reflected immediately in the imported package.

davidrmh / kdro