GML_Glauber_Dynamics.jl

GML_Glauber_Dynamics is a julia package for learning graphical models from time correlated samples generated through Gibbs sampling (aka Glauber dynamics). It is built on top of the julia package GraphicalModelLearning.jl (see package here).

Installation

Install with Pkg, just like any other registered Julia package:

pkg> add GML_Glauber_Dynamics  # Press ']' to enter the Pkg REPL mode.

Getting started

Let's start with a simple example where we generate samples through Glauber dynamics from an Ising model defined on a three node graph. The goal is to then check if the learned graph is close to the true graph from which the samples were generated.

using GML_Glauber_Dynamics

model = FactorGraph([0.0 0.9 0.1; 0.9 0.0 0.1; 0.1 0.1 0.0])
n_samples = 100000
samples_T, samples_mixed_T = gibbs_sampling(model, n_samples, T_regime())
learned_gm = learn_glauber_dynamics(samples_T)

err = abs.(convert(Array{Float64,2}, model) - learned_gm)

Reference

Watch this space! Link to preprint on arxiv coming soon!