Why should I use bipartiteSBM?

The biSBM is a replacement for Graph-tool’s minimize_blockmodel_dl function, specifically tailored for bipartite networks, with the following advantages:

• control directly the numbers of communities to infer for a bipartite network. There are 2 numbers that we can specify; i.e., \(K_a\) and \(K_b\), one for each node type.

• conclude a different partition with a smaller description length (and a higher AMI on tested synthetic dataset).

And, similar to minimize_blockmodel_dl and minimize_nested_blockmodel_dl, it shares with many of their good properties:

• converge to consistent partitions.

• estimate the SBM parameters parsimoniously (without over-fitting or under-fitting).

However, there are also some disadvantages of this program:

• It’s slower than minimize_blockmodel_dl.

• It is not guaranteed to find the globally optimal partition.