A Primer on the Evolution of Equivalence Classes of Bayesian-Network
C. Cotta, J. Muruzábal
Proceedings of the Second European Workshop on Probabilistic Graphical Models, P. Lucas (ed.),
pp. 65-72, Leiden (The Netherlands), 2004
Due to their intuitive interpretation, Bayesian Network
models are often sought as useful descriptive and predictive
models for the available data. Learning algorithms trying to
ascertain the best BN model (graph structure) are therefore of the
greatest interest. In this paper we examine a number of
Evolutionary Programming algorithms for this network induction
problem. Our algorithms build on recent advances in the field and
are based on selection and various kinds of mutation operators
(working at both the directed acyclic and essential graph level).
We carefully measure the merit and computational toll of these EP
variants in a couple of benchmark tasks. Some preliminary
conclusions are outlined.