remi-boutin / sbm_package Goto Github PK

This is a joint work with Pr. Pierre Latouche and Pr. Charles Bouveyron.

In this package, I implemented 2 different estimations : a variational-bayes EM and a variational-EM. The variational-bayes EM can be optimized in two ways : setting the lagrangian to zero, referred to as the VBEM algorithm, or using a natural-conjugate gradient method, referred to as the NCG algorithm.

The main function sbm is only implemented for directed graphs (for now) and can be used this way:

• with no init given : sbm(adj, algo='vbem', init='kmeans')
• with an already computed init : sbm(adj, algo='vbem', tau_init=tau)