search

HGX-Team / hypergraphx

HGX is a multi-purpose, open-source Python library for higher-order network analysis
https://hypergraphx.readthedocs.io/
Other
212 stars 26 forks source link

Initilization of community number K for Hy-MMSBM model #27

Closed mgao97 closed 10 months ago

mgao97 commented 10 months ago

Hi~

I have one question about how to initialize the number of communities K for the Hy-MMSBM method. What if one does not know the potential communities or ground truth of community information?

Thank you in advance and I look forward to your suggestions for such a situation.

nickruggeri commented 10 months ago

Hi @mgao97,

this is more of a "scientific" question than a "coding" one.

In general the number of communities K needs to be chosen via model selection techniques. We briefly touch upon the topic in the paper https://www.science.org/doi/10.1126/sciadv.adg9159 both in the main text and appendix H.

You can look at link prediction measures, such as the AUC or other measures of fit quality, on a test set consisting of some held-out hyperedges that are hidden from the model at training time. Other model selection techniques such as BIC or AIK are also feasible, but they are conjectured not to be very reliable on statical problems on networks.

Hope this helps!

mgao97 commented 10 months ago

Hi～

Thank you for your message and suggestions!

Yes, you are right. It is clear to me.