Estimating the parameters of mixed Bayesian networks fromincomplete data

摘要

Under complete data, there are closed-form maximum likelihoodestimators for mixed Bayesian networks composed of discrete models,conditional Gaussian models and conditional Gaussian regression models.We describe an extension to Lauritzen' expectation-maximisationalgorithm, which estimates the parameters of discrete networks fromincomplete data, to the more general case of mixed continuous anddiscrete variable networks. A simple mixed network that is easy tomanipulate is the leaf node continuous Bayesian network (LNCBN). Fastalgorithms for estimation and marginalisation of LNCBNs are described