A comparison of structural distance measures for causal Bayesian network models
Martijn de Jongh, Marek J. Drużdżel · D-Scholarship@Pitt (University of Pittsburgh) · 2009
We compare measures of structural distance between both, Bayesian networks Aand equivalence classes of Bayesian networks. The main application of these measures is in learning algorithms, where typically the interest is in how accurately a gold standard structure is retrieved by a learning algorithm. Structural distance measures can be especially useful when looking for causal structures. We discuss desirable properties of measures, review existing measures, and show some of our empirical findings concerning the performance of these metrics in practice.