粗等价类双边递减下多次Hash的渐增式求核与约简算法

发布时间：2018-05-02 08:43

本文选题：粗糙约简 + 核　；参考：《系统工程理论与实践》2017年02期

【摘要】：为设计高效约简算法,首先以全局等价类为最小计算单位提出粗等价类概念,证明粗等价类下约简与原信息系统等价;然后深入剖析1,0,-1三类粗等价类的性质,把求正区域等价转化为0-粗等价类双边递减下的渐增式计算,结合1和-1-粗等价类的传递性,设计双边横向删减实体和纵向删减属性的优化规则,可在每一轮增量计算中缩减计算域,基于此设计多次Hash的属性增量划分方法;最后给出新的渐增式快速求核与约简算法,其中求核基于纵向优化规则,可在一次计算中求得多个非核属性,无需遍历全部属性.基于UCI、海量和超高维3类数据集进行多个实验,实验结果证明本文求核与约简算法是高效完备的,在海量数据与超高维数据集下有较大优势.
[Abstract]:In order to design an efficient reduction algorithm, the concept of rough equivalence class is first put forward with the global equivalence class as the minimum unit of calculation, and it is proved that the reduction under the rough equivalence class is equivalent to the original information system. In this paper, the equivalence of positive region is transformed into the incremental calculation under the condition of bilateral decline of 0-rough equivalence class. Combining with the transitivity of 1 and -1-rough equivalence classes, the optimization rules for the horizontal and vertical deletion properties of the two sides are designed. In each round of incremental computation, the computing domain can be reduced and the attribute increment partition method of Hash can be designed several times. Finally, a new incremental fast kernel-finding and reduction algorithm is presented, in which the kernel is based on longitudinal optimization rules. Multiple non-core attributes can be obtained in one calculation without traversing all of them. Based on UCI, three kinds of data sets, massive and ultra-high dimensional, are experimented. The experimental results show that the kernel and reduction algorithms are efficient and complete, and have a great advantage in massive data and ultra-high dimensional data sets.
【作者单位】：广东工业大学管理学院管科系;中国科学院信息工程研究所;
【基金】：国家自然科学基金(71401045,71571052) 广东省自然科学基金(2016A030310300)~~
【分类号】：TP18

【相似文献】