A Framework for the Scoring of Operators on the Search Space of Equivalence Classes of Bayesian Network Structures

Abstract

A method is proposed, whereby a particular application of an operator, applied to a structure representing a Bayesian network equivalence class can be scored in a generic fashion. This is achieved by representing a particular compound operator in terms of a finite set of primitive operators and finding the score of the compound operator through the influence of the primitive operators on the equivalence class. This method could be used in a Bayesian network structure learning framework which allows arbitrary definition of operators at runtime, by the composition of primitive operators.