yerali
(yerali)
June 15, 2017, 4:34pm
#1
Hello dear members of the mailing list,

I am not sure I am doing the gt.minimize_blockmodel_dl properly, over
directed networks. In the plot
<http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/n4027258/dib_particion_ttt_14.jpg> ;
you can see the yellow nodes 27 and 147 are in the center of the star-like
structures. However, with opposite arrows sense. I would expect the same
kind of connectivity, then characterized by similar incomming and outcomming
connectivity.

My instruction is:
State = gt.minimize_blockmodel_dl (g2, deg_corr = False, overlap = False)

Where the directionality has been already established in the graphml input
file.

Shall I change any option in the gt.minimize_blockmodel_dl instruction?

Thank you very much in advance,

Yérali.

tiago
(Tiago Peixoto)
June 15, 2017, 4:40pm
#2
That is not how the degree-corrected SBM works. It incorporates the
(in,out)-degrees as parameters that are independent of the group memberships.

See this paper for a discussion of this model and some alternative
variations: https://arxiv.org/abs/1205.7009

For this particular network, you might want to consider using the
non-degree-corrected model.

Best,
Tiago

yerali
(yerali)
June 15, 2017, 4:59pm
#3
Dear Tiago,

Thanks for your very fast reply.

I am confused about the following. I understand that if I set: deg_corr =
False, as I did in the last example, then I am using the
non-degree-corrected model, as you suggested. Am I wrong?

Thanks again.

Best,

Yérali.

tiago
(Tiago Peixoto)
June 15, 2017, 5:04pm
#4
I'm sorry, I overlooked that.

Indeed in this case it might be the result of a bad fit. You should call the
function multiple times and use the result with the smallest description
length (obtained via state.entropy()).

Best,
Tiago