Research On High-Efficiency Bayesian Network Structure Learning Algorithm

Bayesian network is one of the most important topics in the field of machine learning.Bayesian network structure learning refers to the method of finding the network structure with the highest degree of fitting to training samples.With the advent of the Big Data era,the traditional structure learning methods face enormous challenges.On the one hand,the accuracy of traditional algorithms such as MMHC has become increasingly difficult to meet the requirements of practical scenarios;on the other hand,the traditional algorithm implementation speed is very slow.In this paper,we study how to improve the precision of the algorithm and how to improve the efficiency of the algorithm in Bayesian network structure learning.First,the algorithm of Topology Sequence Heuristic Network Construction based on topology sequence is proposed to deal with the drawback that traditional algorithm is easy to fall into local maximum in the second stage.The basic idea is that each Bayesian network corresponds to a set of topological sequences,and by constructing the optimal topology sequence heuristically,the target network is generated.The algorithm transforms the search of networks into the search of a linear sequence,reducing the difficulty of the search.In this algorithm,the residual climbing network is able to converge quickly,and the climbing network is used as the heuristic function to construct the topology sequence,which greatly improves the accuracy of the results.Experimental results show that the proposed algorithm has a higher accuracy.Secondly,a distributed Bayesian network structure learning algorithm is proposed,which is a distributed algorithm based on topology sequence heuristic construction algorithm.This paper proposes a multi-level k-partitioning tree-based distributed framework,which can balance the accuracy of results and the efficiency of execution.The algorithm divides the set of nodes into multiple sub-sets with common nodes,and then generates the topological sequences of the sub-sets.Then,the master nodes heuristically combine the topological sequences of the sub-sets to generate the target network.Experimental results show that the distributed algorithm has a significant acceleration effect compared with the stand-alone algorithm.