机器学习十大算法：K-means

《机器学习十大算法：K-means》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Chapter2K-MeansJoydeepGhoshandAlexanderLiuContents2.1Introduction............................................................212.2Thek-meansAlgorithm.................................................222.3AvailableSoftware......................................................262.4Examples............

2、...................................................272.5AdvancedTopics........................................................302.6Summary...............................................................322.7Exercises...............................................................33References.........

3、..........................................................342.1IntroductionInthischapter,wedescribethek-meansalgorithm,astraightforwardandwidelyusedclusteringalgorithm.Givenasetofobjects(records),thegoalofclusteringorsegmentationistodividetheseobjectsintogroupsor“clusters”suchthatobjectswithinagrou

4、ptendtobemoresimilartooneanotherascomparedtoobjectsbelongingtodifferentgroups.Inotherwords,clusteringalgorithmsplacesimilarpointsinthesameclusterwhileplacingdissimilarpointsindifferentclusters.Notethat,incontrasttosupervisedtaskssuchasregressionorclassificationwherethereisanotionofatargetvalueorclas

5、slabel,theobjectsthatformtheinputstoaclusteringproceduredonotcomewithanassociatedtarget.Therefore,clusteringisoftenreferredtoasunsupervisedlearning.Becausethereisnoneedforlabeleddata,unsupervisedalgorithmsaresuitableformanyapplicationswherelabeleddataisdifficulttoobtain.Unsupervisedtaskssuchascluste

6、ringarealsooftenusedtoexploreandcharacterizethedatasetbeforerunningasupervisedlearningtask.Sinceclusteringmakesnouseofclasslabels,somenotionofsimilaritymustbedefinedbasedontheattributesoftheobjects.Thedefinitionofsimilarityandthemethodinwhichpointsareclustereddifferbasedontheclusteringalgorithmbeinga

7、pplied.Thus,differentclusteringalgorithmsaresuitedtodifferenttypesofdatasetsanddifferentpurposes.The“best”clusteringalgorithmtousethereforedependsontheapplication.Itisnotuncommontotryseveraldifferentalgorit