Skip to Main Content
Learning the structure of a Bayesian network from a data set is NP-hard. In this paper, we discuss a novel heuristic called polynomial max-min skeleton (PMMS) developed by Tsamardinos et al. in 2005. PMMS was proved by extensive empirical simulations to be an excellent trade-off between time and quality of reconstruction compared to all constraint based algorithms, especially for the smaller sample sizes. Unfortunately, there are two main problems with PMMS : it is unable to deal with missing data nor with datasets containing functional dependencies between variables. In this paper, we propose a way to overcome these problems. The new version of PMMS is first applied on standard benchmarks to recover the original structure from data. The algorithm is then applied on the nasopharyngeal carcinoma (NPC) made up from only 1289 uncomplete records in order to shed some light into the statistical profile of the population under study.