基于MCMC算法贝叶斯网络的学习  

Learning of Bayesian network based on MCMC algorithm

作　　者：史会峰[1] 谷根代[1] 姜波[1] 

机构地区：[1]华北电力大学数理学院,河北保定071003

出　　处：《华北电力大学学报（自然科学版）》2004年第4期109-112,共4页Journal of North China Electric Power University：Natural Science Edition

基　　金：华北电力大学青年基金资助项目 (060203).

摘　　要：对于给定的阈值,通过计算变量之间的互信息,设计了一种构造贝叶斯网络结构的方法。改进了关于图模结构学习中常见的 MCMC 算法。将这种方法构造的贝叶斯网络作为马尔可夫链初始状态的网络结构,利用改进后的 MCMC 算法,构造一个关于贝叶斯网络结构的马尔可夫链。迭代给定次数后,得到关于变量组的贝叶斯网络结构。实验结果表明:改进前和改进后的两种方法得到的贝叶斯网络结构基本一致,网络结构的接受率也相近。For the given threshold, a new method of constructing the Bayesian network is proposed by computing the mutual information between two variables. The traditional Markov Chain Monte Carlo method for structural learning in graphical models MCMC algorithm is improved. Based on the improved algorithm, Markov Chain of the Bayesian network is got. The result of the experiment show that the Bayesian network learned by the improved method is similar to that learned by the old algorithm, and their accepted ratio is also very similar.

关 键 词：贝叶斯网络 MCMC算法 学习方法 随机变量 

分 类 号：O211.5[理学—概率论与数理统计]