Graph clustering by Bayesian inference with cross-validation model assessment.
This is free software, you can redistribute it and/or modify it under the terms of the GNU General Public License, version 3 or above. See LICENSE.txt for details.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- sbm.jl Bayesian inference of the stochastic block model (with full degrees of freedom) using EM algorithm + belief propagation with the leave-one-out cross-validation.
- mod.jl Bayesian inference of the stochastic block model restricted to community structure using EM algorithm + belief propagation with the leave-one-out cross-validation.
To start, the following package needs to be imported:
using DocOpt
using PyPlot
For a given edgelist file, e.g. edgelist.txt,
julia sbm.jl edgelist.txt
generates the following outputs:
- Summary of model assessments (summary.txt)
Input parameters / actual number of clusters & the number of iteration until convergence for eachq. - Detailed results of model assessments (assessment.txt):
Values of the cluster sizes and the affinity matrices learned. - Cluster assignments (assignment.txt):
(i,q)-element indicates the cluster assignments of vertexiwith the input number of clustersq. - Plot of model assessments (assessment_"dataset".pdf)
- [optional]
.smapfiles for the alluvial diagram
julia sbm.jl -help
shows the options and more details.
sbm.jl: Tatsuro Kawamoto and Yoshiyuki Kabashima, "Cross-validation estimate of the number of clusters in a network", Scientific Reports, 7, 3327 (2017).
mod.jl: Tatsuro Kawamoto and Yoshiyuki Kabashima, "Comparative analysis on the selection of number of clusters in community detection", Phys. Rev. E 97, 022315 (2018).
labeled_sbm: Tatsuro Kawamoto, "Algorithmic detectability threshold of the stochastic block model", Phys. Rev. E 97, 032301 (2018).
============
Author: Tatsuro Kawamoto: [email protected]