In naive Bayesian classification, What is the main difference in how discrete categorical and continuous numerical outcomes are handled?