Program and information processing apparatus for Bayesian network structure learning

摘要

To perform a Bayesian network structure learning in high-speed and stable in a situation where there is a large amount of data and a large number of random variables. Per pair of random variable ε or more, add the edge if the graph is still in the tree structure, the information processing apparatus to produce a graph of the tree structure for the input data is the amount of mutual information. For a pair that has not been added edge mutual information while ε is over, I want to add an edge if necessary. Compute the conditional mutual information as a condition sets a set of probability variable included in the set of nodes adjacent to one of the nodes located on the path of a random variable nodes constituting a pair, the value and ε than I want to remove the edge to random variables between the two in the case where there is set to be. If, in the calculation of the conditional mutual information, and the threshold δ is less than the joint probability distribution of the state set corresponding to the state of the random variables of the two is ε or less, is omitted calculation of associated components.