albchim / r-k2algorithm

A naive implementation in Rcpp of K2 Bayesian Net Algorithm

Learning the Topology of a Bayesian Network Using K2 Algorithm

Prerequisites

Python versions supported:

Authors:

• Alberto Chimenti (University of Padova)
• Paolo Frazzetto (University of Padova)
• Vincenzo Schimmenti (University of Padova)

Brief Introduction

A Byesian belief-network structure $B_s$ is a directed acyclic graph in which nodes represent domain variables and arcs between nodes represent probabilistic dependanceies. A variable in such a structure may be continuous or discrete. In the following approach we will consider the discrete case. A BBN structure gives us informations about dependency relationships among the nodes and can be used for probabilistic inference over the system, providing a direct way to compute joint probabilities, given the conditional ones associated with the graph's arcs.

Using a Bayesian approach such a network can be constructed starting from a database which presents several records of the values combination of the system variables. One can find the most probable belief-network structure, given the dataset.

The K2 Algorithm gives a computationally optimized way to search for the most probable structure.

Instead of maximizing the probability over all structures, one assumes that a node has no parents and incrementally adds that parent whose addition most increases the probability of the resulting structure.

A complete explanation of the methods and results obtained can be found in the notebook Project3_CFS_final.ipynb

Sources:

• G.F.Cooper and E. Herskovits
• C. Ruiz
• A. Franzin et al.
• bnstruct
• bnlearn