minimax_OT

Code for paper "A Swiss Army Knife for Minimax Optimal Transport", ICML'20

Dependencies:

• Numpy >= 1.18.1
• POT >= 0.6.0
• CVXPY >= 1.0.25
• MOSEK >= 9.1.9
• Umap >= 0.1.1
• Scikit-Image >= 0.16.2
• Scikit-Learn >= 0.22.1
• Subspace Robust Wasserstein code

Running the experiments:

Main paper

Figure 2

• Left: algoError.py
• Middle: algoErrorVsReg.py
• Right: algo_vs_LP.py

Figure 3

• Left: fragmentedCube.py
• Middle and right: Experiment.ipynb

Supplementary material:

Experiment.ipynb

For Mac users

execute Install Certificates.command in Python folder installation