On 22.03.2017 15:22, yerali wrote:
Thanks for your reply.
The input is test2.graphml: test2.graphml <http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/n4027136/test2.graphml>
and the final partition (using the commands explained before) can be seen in: <http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/n4027136/dib.jpg>
I was expecting each community to have connected nodes, but that is not the case in the blue one, for example.
I don't see anything wrong. The approach is based on fitting the stochastic block model, which accepts any type of mixing pattern, not only the assortative one you are expecting. If you use the non-degree-corrected model (deg_corr=False), the model expects that nodes inside the same group will receive typically the same number of edges, and thus the inference procedure will often just split the nodes according to degree, if there is no structures that are more salient. Indeed, this is why the degree-corrected model (deg_corr=True) is often preferred. Best, Tiago -- Tiago de Paula Peixoto <tiago@skewed.de>