If you plan to use/evaluate tsnetwork, please use the code on https://github.com/HanKruiger/tsNET which is (just a little bit) less experimental.

This repository is still here because my MSc thesis referred to it.

tsnetwork

Graph layouts by t-distributed stochastic neighbour embedding.

This repository contains the implementation of a graph layout algorithm that makes use of the t-SNE dimensionality reduction technique.

The exploration and evaluation of using this technique for graph layouts was done as my MSc thesis project at Rijksuniversiteit Groningen, which I aim to finish in August 2016.

A large part of an essential module in this implementation is a heavily adjusted version of Paulo Rauber's thesne, which is an implementation of dynamic t-SNE.

Dependencies

This was developed and tested solely on ArchLinux.
For this implementation to work, you will need:

• python3
• numpy
• graph-tool
• theano
• graphviz
• matplotlib
• scikit-learn

For rendering fancy animations (even more heavily untested, probably only works on my system) you need:

• ffmpeg
• imagemagick

Benchmark layout animations

For a set of graphs that has been used as a benchmark, animations that show the state of the layout as a function of optimization time can be seen over here.

Warning

Usage of this software is at your own risk.
This utility writes and removes directories in a directory you specify, and (with me being not a professional software developer) you should not trust using this if you're afraid to lose data.