Equivalence class selection of categorical graphical models

TL;DR

This paper introduces a Bayesian approach for learning the structure of categorical graphical models by focusing on equivalence classes, providing a computationally efficient method with strong theoretical backing and good empirical performance.

Contribution

It develops a novel Bayesian methodology for structure learning of categorical essential graphs, including a closed-form marginal likelihood and an MCMC inference scheme.

Findings

Effective on simulated data

Competitive with state-of-the-art methods

Strong theoretical properties

Abstract

Learning the structure of dependence relations between variables is a pervasive issue in the statistical literature. A directed acyclic graph (DAG) can represent a set of conditional independences, but different DAGs may encode the same set of relations and are indistinguishable using observational data. Equivalent DAGs can be collected into classes, each represented by a partially directed graph known as essential graph (EG). Structure learning directly conducted on the EG space, rather than on the allied space of DAGs, leads to theoretical and computational benefits. Still, the majority of efforts in the literature has been dedicated to Gaussian data, with less attention to methods designed for multivariate categorical data. We then propose a Bayesian methodology for structure learning of categorical EGs. Combining a constructive parameter prior elicitation with a graph-driven likelihood decomposition, we derive a closed-form expression for the marginal likelihood of a categorical EG model. Asymptotic properties are studied, and an MCMC sampler scheme developed for approximate posterior inference. We evaluate our methodology on both simulated scenarios and real data, with appreciable performance in comparison with state-of-the-art methods.