Learning Bayesian networks with mixed variables

This project considers conditional Gaussian (CG) networks. We learn the parameters in the network by using conjugate Bayesian analysis. As conjugate local priors, we apply the Dirichlet distribution for discrete variables and the Gaussian-inverse gamma distribution for continuous variables, given a configuration of the discrete parents. We assume parameter independence and complete data. Further, to learn the structure of the network, the network-score is deduced. We then develop a local master prior procedure, for deriving parameter priors in CG networks. This procedure satisfies parameter independence, parameter modularity and likelihood equivalence. Bayes' factors to be used in model search are introduced. Finally the methods derived are illustrated by a simple example. This project is supportet by the ESPRIT project P29105 (BaKE)