摘要
"No Free Lunch"定理表明:若无任何先验假设,则没有理由认为一种算法优于另一种算法.算法的性能与问题的元特征密切相关.目前的元特征提取方法只关注从数据集中提取元特征,而忽略了候选算法元特征的提取.为此,在原有元特征集合的基础上提出基于决策树桩的元特征提取方法,将候选算法信息纳入新的元特征集合中.实验表明:在传统元特征集合中加入基于决策树桩的元特征后,算法排序的预测准确率能够得到显著提高.
The"No Free Lunch"theorem shows that there is no reason to think that one algorithm is superior to the other one without any prior assumptions.The performance of algorithm is closely related to the meta-feature of problem.The current meta-feature extraction method is only concerned with extracting meta-feature from the data set,while ignoring the meta-feature extraction of candidate algorithms.Therefore,an extraction method based on decision stump is proposed,which can effectively reflect the information of candidate algorithms.Experiments show that the new meta-feature sets significantly increase the prediction accuracy of algorithm ranking.
作者
曾子林
陈建军
ZENG Zilin;CHEN Jianjun(Army Infantry College of People′s Liberation Army,Nanchang Jiangxi 330103,China;Shangrao Vocational and Technical College,Shangrao Jiangxi 334109,China)
出处
《江西师范大学学报（自然科学版）》
CAS
北大核心
2018年第6期616-620,共5页
Journal of Jiangxi Normal University(Natural Science Edition)
基金
国家自然科学基金(11501281)
装备军内科研课题(面向作战任务的分队战斗体能数据分析评估系统建设)
江西省社科"十二五"规划课题(15GL44)资助项目
关键词
元特征
算法性能
算法排序
决策树桩
meta-feature
performance of algorithms
ranking of algorithms
decision stump
