Thank you very much! I experimented a little bit with overlap = True in the LayeredBlockState and I don't know if this is a common issue or it is just me but 1) solver takes way longer to equilibrate compared to BlockState(and networks are not very large, around N=100 with 4 layers on average, which I also threshold the weighted network edges) and 2) more interestingly, partitions in the layers that are towards the end of the temporal network are found perfectly whereas the communities I want to find in the first a few layers are partitioned into almost singleton communities. This is what I mean:
<https://nabble.skewed.de/file/t496292/Screen_Shot_2021-03-17_at_11.png> <https://nabble.skewed.de/file/t496292/Screen_Shot_2021-03-17_at_11.png> I wasn't sure if this is because of the low number of iterations which I kept 100<niter<1000 (I guess reasonable given the network size and number of edges) but I kept getting similar results over different trials. One other thing I noticed is that DSBM fits non-communities(the upper left stairs in the ground truth) with %100 accuracy. I think I'm going to try fitting a different sbm to each layer as next step and share how it goes. Sorry for keep bugging you!! Best -- Sent from: https://nabble.skewed.de/ _______________________________________________ graph-tool mailing list [email protected] https://lists.skewed.de/mailman/listinfo/graph-tool
