论文部分内容阅读
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 very important for Bayesian network to solve practical problems.Based on the method of scoring and searching, it is the most commonly used method of structure learning at present, but the size of structure search space increases with the number of nodes Exponential growth, so generally heuristic search strategy, some methods also need to node order.In the node-based maximum correlation - minimum redundant greedy Bayesian network structure learning algorithm, because it is randomly generated in the order of the initial node , Which increases the uncertainty of the result.In this paper, we propose a method to generate the initial order of the optimized nodes. After obtaining the initial order of the basic ordered nodes, iterative search with the nearest neighbor exchange operator can be performed in a short Time more accurate Bayesian network structure.The experimental results show that the method is effective.