Wasserstein t-SNE

Fynn Bachmann, Philipp Hennig & Dmitry Kobak

This repository reproduces the figures in the Wasserstein t-SNE paper in the proceedings of ECML/PKDD 2022.

For the python package WassersteinTSNE see the repository WassersteinTSNE.

Dependencies

This repository uses the WasserteinTSNE package version 1.1.0 and its dependencies

• numpy
• matplotlib
• pandas
• scipy
• openTSNE
• scikit-learn
• igraph
• leidenalg

Usage

Running python main.py will reproduce all Figures in Figures/Figure{i}.pdf. This takes about two minutes when using the cached files (default). To recompute these files please see the respective lines in main.py or Experiments/Figure{i}.py.

Data

The dataset used in the paper is publically available at the website of the Bundeswahlleiter. It is also included in the repository at Datasets/GER2017/....

A second dataset, the European Values Study (which wasn't used in the paper) is also available at GESIS and in Datasets/EVS2020/....

Figures

All figures are already included in this repository at Figures/Figures{i}.pdf. See params.py for customization.