基于量化误差与分形理论的高计算效率无监督聚类研究  

Quantization error and fractal theory based high computation efficiency unsupervised clustering algorithm

作　　者：胡国生[1] 杨海涛[2] Hu Guosheng;Yang Haitao(School of Software, Guangdong Food & Drug Vocational College, Guangzhou 510520 , China;School of Mathematics, Zhejiang University,Hangzhou 310021, China)

机构地区：[1]广东食品药品职业学院软件学院,广州510520 [2]浙江大学数学学院,杭州310027

出　　处：《计算机应用研究》2016年第10期2919-2922,共4页Application Research of Computers

基　　金：浙江省自然科学基金资助项目(Y1090416);浙江省自然科学基金资助项目(Y1091084)

摘　　要：已有的矢量聚类算法需学习较多的复杂数据方可获得较好的聚类效果,而对于多维的大数据性能较弱,为此提出一种基于量化误差与分形理论的高计算效率无监督聚类算法。首先,为数据集建立量化误差的参数化模型,基于数据集的空间结构获得数据集的率失真曲线;然后通过对率失真曲线的估算,获得数据空间的有效维度;最终利用分形理论,通过搜索数据集的量化模型参数获得目标数据集的最优类簇数量。实验结果表明,该量化误差参数化模型可较好地估算数据集的有效维度,同时,本算法对数值型数据集的最优类簇估算与计算效率优于已有的矢量聚类算法。The existing vector clu ste rin g a lgo rithm need to learn a lo t o f com plex data in order to get a good perform ance forc lu s te rin g , and it does not have good perform ance fo r b ig data. T h is paper proposed a quantization error and fra cta l theory basedh igh com putation e fficie n cy unsupervised clu ste rin g algo rithm to solve that pro b le m . F ir s tly ,it constructed a param etric m odeling o f the quantization e rror fo r data s e t,g o t the ra te -d isto rtio n curve based on the space structure o f the data set. T h e n ,it com putedthe e ffic ie n t d im e nsion ality o f the data set by estim ation o f the rate disto rtion cu rve . L a s tly ,it obtained the op tim al clustering num ber o f the target data set by fra cta l theory. E xperim ents re su lt shows th a t the proposed qu antization error m odeling canestim ate the qu antization error very w e ll and the proposed a lgo rithm has be tter perform ance in search the best clu ste rin g num berand com putation e fficie n cy than the existing vector clu ste rin g a lg o rith m .

关 键 词：分形理论 量化误差 率失真曲线 无监督聚类 多维数据 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]