Fast Algorithm for Learning the Bayesian Networks From Data

摘要

The new constraint-based algorithm for learning dependency structures from data is developed. The novelty of the proposed algorithm is conditioned by the rules of acceleration of inductive inference, which drastically reduce the search area of separators while derivation of the model skeleton. On examples of the Bayesian networks of moderate saturation we have demonstrated that proposed algorithm learns Bayesian nets (of moderate density) multiple times faster than well-known PC algorithm.