基于贝叶斯-遗传算法的多值无环CP-nets学习

Learning of multi-valued acyclic CP-nets based on Bayesian-genetic method

条件偏好网(Conditional Preference networks,CP⁃nets)是描述属性间条件偏好的图模型,多值无环CP⁃nets学习是重要的研究方向之一.区别于传统的CP⁃nets学习方法,提出基于贝叶斯方法和遗传算法的多值无环CP⁃nets学习.在偏好处理上以多值属性的完整偏序关系作为条件偏好,进行相关性关系判定.随后,基于贝叶斯方法,以单一父属性推出多父属性下的相关性关系,进行CP⁃nets结构学习.采用遗传算法在CP⁃nets结构搜索空间中进行搜索,求解最优结构.通过Delink算法进行去环,完成无环CP⁃nets学习.在寿司数据集上验证算法的有效性,实验结果表明,基于贝叶斯⁃遗传算法的CP⁃nets学习算法能够在有限时间内学习得到局部最优无环CP⁃nets.

Conditional preference network(CP⁃net)is a graph model describing conditional preferences among attributes.Multi⁃valued acyclic CP⁃nets learning is an important research direction.Different from the traditional CP⁃nets learning method,a multi⁃valued acyclic CP⁃nets learning method based on Bayesian method and genetic algorithm is proposed.In preference processing,the complete partial order relationship of multi⁃valued attributes is used as conditional preferences to determine the correlation relationship.This method uses single parent attribute to deduce the relativity under multi⁃parent attributes and to learn CP⁃nets structure.Genetic algorithm is used to search in the search space of CP⁃nets structure to find the optimal structure.Delink algorithm is used to de⁃loop and complete acyclic CP⁃nets learning.The validity of the algorithm is validated on sushi data set.The experimental results show that the algorithm based on Bayesian⁃genetic algorithm is effective.CP⁃nets learning algorithm can obtain local optimal acyclic CP⁃nets in limited time.