Skip to content

sbyebss/Scalable-Wasserstein-Barycenter

Repository files navigation

Scalable Computations of Wasserstein Barycenter viaInput Convex Neural Networks

Jiaojiao Fan, Amirhossein Taghvaei, Yongxin Chen [arXiv]

Citation

@misc{jiao2021nwb,
title={Scalable Computations of Wasserstein Barycenter viaInput Convex Neural Networks},
author={Jiaojiao Fan},
year={2021}
}

Dependencies and Installation

  1. The following python packages are required:
  1. git large file storage has to be initialized to download the input data.

Code structure

  • The scripts in the root such as G2G_sameW_3loop.py are the core code for our NWB (Neural Wasserstein Barycenter) implementation. For example, if you are interested in Gaussian example, run in terminal python G2G_sameW_3loop.py --parameter_1 param1_value --parameter_2 param2_value.

  • generator_example/ contains the scripts to generate comparison results or visualization. For example, if you are interested in Gaussian example, run in terminal python ./generator_example/NWB1_gmm.py --parameter_1 param1_value --parameter_2 param2_value.

  • optimal_transport_modules/ contains the auxiliary utility modules.

config file

The configuration of an experiment is entirely described by a optimal_transport_modules/cfg.py config file. If you want to change the parameter, please change them there.

What if you find a bug?

This repository is still under construction. If you meet a bug when you run the code, please raise up an issue, thank you!