Robust Optimal Transport for Classification (ROT4C)

This repo implements different Robust Optimal Transport formulations as discussed in:

Pratik Jawanpuria, N T V Satya Dev, Bamdev Mishra. Efficient robust optimal transport: formulations and algorithms, arXiv preprint arXiv:2010.11852, 2020

Requirements

• Python (3.6+)

Install requirements from requirements.txt

Usage

Checkout --help of rot4c.py for detailed usage.

python rot4c.py --help

Sample

A helper bash script is provided that runs the multi-class learning setup on a sample of AwA data in ./data/ folder

Expected AUC: 0.903

NOTE

For GPU usage, checkout algo.py (line no. 1) and sinkhorn_gpu.py (line no. 12)