View article

This report describes a procedure for analysis of the conditional probability tables (CPT) using the KBM2L framework. A CPT is a table for a discrete random variable X conditioned to a given set of discrete random variables Y. The CPT has n columns, the domain size of X, and m rows, the cartesian product of domain size of Y. One position (i, j) at the CPT is P (X= xj| Y=(Yi= yi)). Every row is a probability distribution, P (X| Y=(Yi= yi)). The index list of the KBM2L is the vector of Y values, and the response is the probability distribution. We use a cluster (pam: Partitioning (clustering) of the data into k clusters “around medoids”) over the CTP’s rows, to generate a few (∼ 5) profiles of probabilistic behaviour, medoids, and after the KBM2L optimization process we derive the CPT explanation.