Algebraic Bayesian networks are related to the class of probabilistic graphical models. As a machine learning model they are required to be trained on some data set. This work is dedicated to the frequentist approach to machine learning of a knowledge pattern as a local learning of the Algebraic Bayesian network. The theoretical explanation of approach is provided and the algorithm is described. The algorithm’s pseudocode is presented, its theoretical complexity is calculated. Then an experiment is conducted and real estimates of the algorithm's implementation time of work are received.