Am 20.04.21 um 12:56 schrieb arttu.malkam...@helsinki.fi:
Dear Community,
I've been working on community detection for some time already (Louvain/Leiden
algorithms), but I got started with graph-tool only recently. After spending
some time scrolling the HOWTO, I still haven't found answers to my two concerns
that are related to the applicability of SBM to my data.
I'm analysing political coalitions by combining data from a survey, newspapers,
and Twitter. To study coalitions across the three weighted networks (or more if
I include temporal slices), I am using LayeredBlockState through
NestedBlockState.
Question 1:
I haven't found an answer to whether it is possible to run SBM for layered
networks with edge weights? I can get around this problem by extracting the
noise-corrected 'backbone' of each network, which yields simple graphs, but
ideally I wouldn't have to do too much violence to the data but be able to use
the raw weighted networks as inputs. Is it possible to run LayeredBlockState
with edge weights?
Yes, it's perfectly possible. Just pass the edge covariate parameters
(recs, rec_types, rec_params) to LayeredBlockState as you would
otherwise to BlockState.
Question 2:
Even if the LayeredBlockState would not (yet) support weighted networks, I also
have another, more fundamental question. As my three layers come from different
data generation processes, they do not share the exactly same set of nodes. For
example, one organisation responded the survey but doesn't necessarily appear
in the newspaper data. Is possible to determine constraints for certain nodes
in certain layers that would tell the LayeredBlockState to not consider
layer-specific isolates?
If you used the "layers=True" version of the model, then the nodes of
degree zero in a given layer are considered not to be long to that
layer, i.e. they have a probability of zero of receiving edges of that type.
Best,
Tiago
--
Tiago de Paula Peixoto <ti...@skewed.de>
_______________________________________________
graph-tool mailing list -- graph-tool@skewed.de
To unsubscribe send an email to graph-tool-le...@skewed.de
