An Improved Algorithm Based on BBES for Learning Bayesian Network Structure from Data

Abstract

In this paper we look at some approaches to improve the computational performance of BBES (van Ommen, 2018) algorithm, which aims to learn Bayesian network structure from data. We discuss about how to choose AIC (Akaike information criterion) and BIC (Bayesian information criterion) score in different situations. We give a new branching heuristic based on Graph Attention Network and evaluate it on both simulated continuous data and real-life data.