Show HN: KarateClub a Python library for unsupervised machine learning on graphs

[benitorosenberg]

https://github.com/benedekrozemberczki/karatecluB

[cl42]

This is a great name!! For those who aren't familiar, there's a famous graph data set about a karate club: https://en.wikipedia.org/wiki/Zachary%27s_karate_club

Thanks for bringing me back to undergrad... :) (...where I learned about this data set.)

[teruakohatu → cl42]

Thank you for sharing. The DOI is 10.2307/3629752 if anyone wants to look it up.

[lasershow → cl42]

+1 on the great name and thanks for the reference link

[lammalamma25]

Very cool. This kind of work makes python and ML better for everyone. I'm not sure I have a use case at the moment, but I'll definitely be keeping an eye on this one.

[fudged71]

I've recently been looking into community detection for NetworX graphs for my collection of notes. How does performance compare with other community detection packages for NetworkX like Python-Louvain?

[kugestu → fudged71]

Random side note: nowadays 'Leiden' is a recommended improvement on 'Louvain', fixing some connectivity issues and being faster: https://www.nature.com/articles/s41598-019-41695-z https://github.com/vtraag/leidenalg

[benitorosenberg → fudged71]

Performance measured by what? NMI? Modularity?

[fudged71 → benitorosenberg]

Not super familiar with the measures, I'm just curious how it compares to other available approaches

[benitorosenberg → fudged71]

Louvain greedily optimizes modularity - it is hard to beat it in terms of modularity.