MASTINO: Learning Bayesian Networks Using R

Bayesian Networks are increasingly used to represent conditional independence relations among variables and causal information in problem domains in which decisions are based on probabilistic reasoning. Structural learning is NP-hard therefore the database of observed cases must be often supplemented with search heuristics based on prior infor- mation. In this paper we present a software package for R, called MASTINO, that extends the existing DEAL package by providing new tools for learning Bayesian Networks and Conditional Gaussian networks in a score-and-search framework, such as the score func- tion P-metric and the M-GA genetic algorithm. MASTINO is freely available under the terms of the GNU General Public License Version 2, and it has been recently submitted to be part of the CRAN repository. Meanwhile it can be downloaded from the website: http://statind.jrc.it/mastino.