Learning Bayesian Network Structures When Discrete and Continuous Variables Are Present

摘要

In any database, some fields are discrete and others continuous in each record. We consider learning Bayesian network structures when discrete and continuous variables are present. Thus far, most of the previous results assumed that all the variables are either discrete or continuous. We propose to compute a new Bayesian score for each subset of discrete and continuous variables, and to obtain a structure that maximizes the posterior probability given examples. We evaluate the proposed algorithm and make experiments to see that the error probability and Kullback-Leibler divergence diminish as n grows whereas the computation increases linearly in the logarithm of the number of bins in the histograms that approximate the density.