Add community detection algorithms

[nwlandry]

No description provided.

[thomasrobiglio]

Just to get the ball rolling, I have been looking at what the other libraries have about this.

HGX

• Hy-MMSBM (hypergraph mixed-membership stochastic block model). They do the fitting part using EM.
• Hy-SC (spectral clustering). I'd have to read the paper for the details.
• Hypergraph-MT (mixed memberships, assortative structures), also here EM.
• Hyperlink communities. Hierarchical clustering on hyperedges distances/similarities.

HypernetX

• Modularity stuff. They have two methods that apply Louvain to maximize the extension of modularity to HGs (I think the one defined here--I have just skimmed the papers).

Which ones do we want to implement? Do we have a priority list or something of the kind?

[maximelucas]

I would go first for the most "basic" and standard one. And by standard I mean if not standard in hypergraphs yet, that its potential analogue in graphs is standard. Bonus points if it's not the slowest.

Any idea which one would fit? For example I like the mixed-membership ones, but maybe mixed-membership is already kind of more advanced?

[nwlandry]

I think Louvain-style algorithm would be simplest. In fact, I was hoping to implement the Kaminski et al. paper at one point: https://arxiv.org/abs/1810.04816. Maybe this is a good place to start?

[thomasrobiglio]

mmm yes... modularity and link communities (allows for mixed memberships on the nodes @maximelucas ) should be quite easy to implement and relatively fast... about the standard question, it depends on who you ask (aaaargh).

[thomasrobiglio]

but yeah, I guess we can have them... if no one calls dibs I can work on it.