一种基于预排序的贝叶斯网络结构学习算法  

A Presorting-based Algorithm on Bayesian Network Structure Learning

作　　者：陈坤[1] 姚望舒[1] 

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006

出　　处：《小型微型计算机系统》2013年第9期2135-2138,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61033013)资助;江苏省基础研究计划(自然科学基金)企业博士创新项目(BK2009583)资助;校青年基金项目(Q3118814)资助

摘　　要：贝叶斯网络结构学习对贝叶斯网络解决实际问题至关重要.基于评分与搜索的方法是目前比较常用的结构学习方法,但该类方法中结构搜索空间的大小随结点个数增加而指数增长,因此一般采用启发式搜索策略,有些方法还需要结点次序.在基于结点次序的最大相关-最小冗余贪婪贝叶斯网络结构学习算法中,由于是随机产生初始结点的次序,这增大了结果的不确定性.本文提出一种生成优化结点初始次序的方法,在得到基本有序的结点初始次序后,再结合近邻交换算子进行迭代搜索,能够在较短的时间内得到更加正确的贝叶斯网络结构.实验结果表明了该方法的有效性.Bayesian network structure learning is critically important when using Bayesian network to solve practical problem. Al- though the method at present that based on Search ＆ Score is more and more popular, the search space of structure increased exponen- tially with the number of nodes, so usually adopt heuristic search methods and some algorithm require node ordering. When learning structure of Bayesian network by Ordering-based Max-Relevance and Min-Redundaney Greedy algorithm, node＇s initial order pro- duced at random will increase the uncertainty of the result greatly. In this paper, we propose a new method that can find a basic order- ly order. This method combined with neighbor-swap operator to search iteratively, can obtain a better result in a shorter time. Experi- mental results show that the new approach is correct and effective.

关 键 词：最大相关-最小冗余 贝叶斯网络 结构学习 结点次序 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]