Implement canonical vertex nomination / likelihood maximization?

[loftusa]

https://projecteuclid.org/journals/annals-of-applied-statistics/volume-9/issue-3/Vertex-nomination-schemes-for-membership-prediction/10.1214/15-AOAS834.full

Right now we have spectral nomination implemented for single graphs (good at large scales, according to the above paper)

There are two other main schemes the paper discusses, canonical vertex nomination and a likelihood-maximization approach.

Do we want either of them? canonical scales horribly but does much better than the others at ~n=10, likelihood-maximization is good at around n=500-2000 then falls off in time complexity+accuracy, spectral (the one we have implemented right now) is mainly good for large graphs (n=~10,000+)

EDIT: new paper finds an approximation to the one that scales horribly https://bit.ly/2YXzYp9, using a Monte Carlo approximation, maybe worth checking out