Once I corrected
that error by only sampling edges which don't have a reciprocal the
performance between undirected/directed mostly disappeared. So,
performance still bad. To ensure it wasn't just my network or something
to do with directedness I tried several simple undirected graphs from
graph-tool's collection (where I don't need to worry about multiedges).
All performed poorly.
Regardless of your claimed performance, I still think it's pointless to
comment on unseen code.
The directed SBM does not generate reciprocal edges very well, hence it is not a good
predictor in this case.
Where can I read more about this? I assumed the SBM handles the directed
case just fine, even if there are many reciprocal edges. Does this have
implications for community detection as well?
The SBM does not distinguish between nodes that belong to the same group
(and have the same in/out-degrees in the degree-corrected version), and
places edges in a conditionally independent manner. Therefore even if
the probabilities are symmetric between two groups in both directions,
the probability that exactly the same edges are placed in both
directions is very small.
The consequences for community detection are the same for other
small-scale structure not captured by the SBM (e.g. triangles): edge
reciprocity is not used to find comnunities.
Best,
Tiago