tomasonjo
commented
3 years ago For community detection, there aren't a lot of hyper-parameters... just tolerance, maxIterations... But there could be a blog post and some mention in the book. On the other hand, similarity algorithms heavily depend on hyperparameters. I was planning a blog post on how to do a monopartite projection using various algorithms and just plain cypher.
There are so many people who run algorithms like Louvain and just go with the default values, not understanding when you would or would not want to use those values. But I don't think I have seen (and please correct me if I am wrong!) a good guide on how to tune the hyperparameters of each of these algorithms.
I don't know if a chapter can be added to the book at this stage, but it might be worth it in some of the major chapters (I am particularly thinking about community detection) to walk the reader through what those parameters really mean, how those parameters impact the results, and how to select the proper set to get the results you are looking for.