Effective and efficient structure learning with pruning and model averaging strategies

Learning the structure of a Bayesian Network (BN) with score-based solutions involves exploring search space possible graphs and moving towards graph that maximises given objective function. Some algorithms offer exact guarantee to return highest score, while others approximate in exchange for reduced computational complexity. This paper describes an BN learning algorithm, which we call Model Averaging Hill-Climbing (MAHC), combines two novel strategies hill-climbing search. The algorithm starts by pruning graphs, where strategy can be viewed as aggressive version are typically applied combinatorial optimisation problems. It then performs model averaging process moves neighbouring function, on average, over all its valid graphs. Comparisons other spanning different classes suggest combination is both effective efficient, particularly presence data noise.