On 23.03.2017 03:14, Peter Straka wrote: > Hi Yerali, > for assortative communities you would maybe want to find a partition with > maximum modularity, which is a bit different to minimizing blockmodel > description length.
I don't think this is a good idea for the reason that I have mentioned: Maximizing modularity ignores the statistical significance of the partition. Because of this, by maximizing modularity one can find partitions of fully random graphs with very high modularity, but which are completely meaningless: https://journals.aps.org/pre/abstract/10.1103/PhysRevE.70.025101 This is not a problem with the SBM inference approach implemented in graph-tool, which is meant to solve this problem in a principled manner. So when the algorithm finds only one community in your data, it is telling you something important: You graph cannot be distinguished from a fully random one with the same size and density (or degree sequence). It is a bad idea (and bad practice) to ignore this, and use some heuristic just because it finds some partition. This is a recipe for overfitting. Best, Tiago -- Tiago de Paula Peixoto <[email protected]>
signature.asc
Description: OpenPGP digital signature
_______________________________________________ graph-tool mailing list [email protected] https://lists.skewed.de/mailman/listinfo/graph-tool
