Choosing number of blocks in each side of the bi-partite SBM

graph-tool
1

Hi,

I am aware of a post asking for the same topic, but it was dated 2 years
ago, but I wonder if there was any development on this front?

In the function "model.minimize_nested_blockmodel_dl" it is possible to
specify B_min=G and B_max=G in order to get exactly G blocks for the whole
clustering. I wonder if it is possible to specify separately the number of
blocks on each side of the bi-partite network, say G1 for one side of the
network and G2 for the other side? Even maybe by running 2 different
estimations?

I also noticed that Tzu-Chi Yen has an alternative code for this, but I
wonder how comparable it is to graph-tool in terms of efficiency?

Code: https://github.com/junipertcy/bipartiteSBM Related doc:
https://docs.netscied.tw/bipartiteSBM/usage/explore-consistency.html

Thank you all!

attachment.html (4.33 KB)

2

Dear Bernardo,

Hi,

I am aware of a post asking for the same topic, but it was dated 2 years
ago, but I wonder if there was any development on this front?

In the function "model.minimize_nested_blockmodel_dl" it is possible to
specify B_min=G and B_max=G in order to get exactly G blocks for the
whole clustering. I wonder if it is possible to specify separately the
number of blocks on each side of the bi-partite network, say G1 for one
side of the network and G2 for the other side? Even maybe by running 2
different estimations?

Unfortunately, this functionality has not been implemented yet.

I also noticed that Tzu-Chi Yen has an alternative code for this, but I
wonder how comparable it is to graph-tool in terms of efficiency?

Code: GitHub - junipertcy/bipartiteSBM: A Bayesian model+algorithm for community detection in bipartite networks
<https://github.com/junipertcy/bipartiteSBM&gt;Related
doc:Explore the consistency of results — junipertcy/bipartiteSBM 0.0.1 documentation
<https://docs.netscied.tw/bipartiteSBM/usage/explore-consistency.html&gt;

As you can see for yourself, that implementation is in pure Python,
whereas graph-tool is implemented in C++. Therefore it should be much
slower than graph-tool.

Best,
Tiago