On Using The Pc Algorithm For Learning Continuous Bayesian Networks: An Experimental Analysis

Mixtures of truncated basis functions (MoTBFs) have been recently proposed as a generalisation of mixtures of truncated exponentials and mixtures of polynomials for modelling conditional distributions in hybrid Bayesian networks. However, no structural learning algorithm has been proposed so far for such models. In this paper we investigate the use of the PC algorithm as a means of obtaining the underlying network structure, that is finally completed by plugging in the conditional MoTBF densities. We show through a set of experiments that the approach is valid and competitive with current alternatives of discretizing the variables or adopting a Gaussian assumption. We restrict the scope of this work to continuous variables.