摘要
针对爬山法容易陷入局部最优,而随机重复爬山法时间开销过大的问题,将互信息与爬山法相结合,提出了MI&HC贝叶斯网络结构学习算法。首先利用互信息构建初始网络结构,再从该网络结构开始利用爬山法进行贝叶斯网络结构学习。仿真结果表明:MI&HC算法,对小型稀疏网络结构的学习效果非常好,对较大型的网络结构的学习也能得到令人满意的结果;该算法不需要节点顺序这一先验信息,却能获得与K2算法相当的学习效果。
Aiming at the problems that the hill climbing algorithm is easy to get into local optimisation and the random repeated hill climbing algorithm costs too much time,a Bayesian network structure learning algorithm,MIHC,is presented,which combines the mutual information with the hill climbing algorithm.First the initial network is constructed using mutual information,and then the Bayesian network structure learning begins proceeding from the initial network structure using hill climbing algorithm.Simulation results show that the learning effect of the algorithm of MIHC is very good for small sparse networks,and is satisfactory as well for network structure with bigger size.It can achieve the learning effect equivalent to algorithm of K2＇s without priori information of the nodes order.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第9期122-125,共4页
Computer Applications and Software
基金
军队科研基金资助项目
关键词
互信息
爬山法
贝叶斯网络
结构学习
Mutual information, Hill climbing algorithm, Bayesian network ,Structure learning
