Formation of evaluation functions for solving the task of constructing directional dynamic Bayesian network

The formation and description of evaluation functions is an important task for determining the necessary and sufficient conditions for the existence of directional links between Bayesian network nodes. The use of evaluation functions of various types allows you to evaluate and optimize the network training procedure and determine the function that is most adapted for a particular network with a fixed set of training samples. The work examines evaluation functions based on the logarithm of likelihood, mutual entropy, as well as the Bayes - Dirichlet asymptotic metric. These functions are widely used in the process of solving problems of training the structure of probabilistic models and can be adapted to determine the direction of connections between nodes of temporary models built on the basis of dynamic Bayesian networks.