dss 第九組期末報告 7.4 p.401~p.408

39
DSS 第第第第第第第 7.4 P.401~P.408 第第93156204 第第第 93156206 第第第 93156208 第第第 93156219 第第第 DSS 第第第第第第第 1

Upload: kaili

Post on 23-Jan-2016

81 views

Category:

Documents


0 download

DESCRIPTION

DSS 第九組期末報告 7.4 P.401~P.408. 組員: 93156204 盧宗佑 93156206 王雅玲 93156208 許韶玲 93156219 詹伯為. DSS 第九組期末報告. 7-4 Partitioning Methods. DSS 第九組期末報告. 目錄. Cluster Partitional clustering algorithm Partitioning algorithm K-Means K-Medoids 大型資料庫處理 綜合比較 總結. 目錄. Cluster - PowerPoint PPT Presentation

TRANSCRIPT

  • DSS

    7.4 P.401~P.408 93156204 93156206 93156208 93156219

    DSS*

    DSS

  • 7-4 PARTITIONING METHODS DSS*

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • CLUSTERclustercluster

    DSS*

  • CLUSTERDSS*

  • *

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • PARTITIONAL CLUSTERING ALGORITHM

    (Partitioning Algorithm)

    (Hierarchical Algorithm)

    (Density-Based Algorithm) *

  • PARTITIONING ALGORITHMK (partition)(cluster) (partitioning criterion) (similarity function)

    K-MeansK-Medoids (heuristic) *

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • K-MEANS1967J.B.MacQueen

    K-means

    :

    (spherical-shaped) data mining

    *

  • K-MEANS k(cluster)

    k

    k

    *

  • K-MEANS Input K- D-nOutputK

    Method:nK 1n ()*

  • K-MEANS 4. 3 ((cluster) )5.

    () square-error criterion()

    miipiCii*

  • EXAMPLE1. 6 INSTANCES, 2 ATTRIBUTES, 2 CLUSTERS2. RANDOMLY SELECT INSTANCE 1&3 IN INITIAL*

  • *

  • K-MEANS COMPLEXITYK-means K,square error function clusters,clusters.methodO(nkt),nobject,kcluster,titeration(k
  • K-MEANS k- (scalable)

    *

    DSS

  • K-MEANS1.Hierarchical agglomerationDetermines the number of cluster and finds an initial clustering , and the then use iterative relocation to improve the clustering.

    *

    DSS

  • K-MEANS()2.K-modesExtend the k-means paradigm to cluster categorical data by replacing the means of cluster with modes. using new dissimilarity measure.Using new dissimilarity measures to deal with categorical object and a frequency-based method to update modes of clusters.*

    DSS

  • K-MEANS()3.EM(Expectation-Maximization)Each object is assigned to each cluster according to a weight representing its probability of membership.

    *

    DSS

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • K-MEDOIDS METHOD k-k-k-k-k-E

    *

    DSS

  • ABSOLUTE-ERROR CRITERION

    Eabsolute errorCjclusterPCjMiCjobject*

    DSS

  • K-MEDOIDSk-k-k-k-k-E(Partitioning Around Medoids, PAM) *

  • PAMPAM Kaufman and Rousseeuw

    K-means

    medoidmean

    *

  • PAMPAMk(representative objects)medoidmedoid(Euclidean distance)d(Oa, Ob)OaObOimedoidOjmedoidd(Oj,Oi)=min{d(Oj, Oe)} OemedoidsOjOi

    *

  • PAMmedoidOj,medoid OimedoidOhCjihCjih= d(Oj, Om) d(Oj, On) Oh Oi medoid TCih= Cjih

    TCih>0OhOiOiOhTCih*

  • K-(PAM)kk 1kk 2k 3E 4E2 5 *

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • K-medoidsPAM CLARA (Clustering LARge Application)CLARA applies PAMO(ks^2+k(n-k))*

    DSS

  • CLARA -- PAM CLARA Trade-off*

    DSS

  • CLARA CLARANS (Clustering Large Application based upon RANdomized Search) graphO(n^2) *

    DSS

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • *

    K meansK medoidsCLARACLARANS O(nkt)O(k(n-k)^2)O(ks^2+k(n-k))O(n^2)

    DSS

  • ClusterPartitional clustering algorithmPartitioning algorithmK-MeansK-Medoids*

    DSS

  • Supervised clustering method K supervised clustering method

    *

    DSS

  • --k, http://vega.cs.tku.edu.tw/~cyh/data_mining/F7899-Ch06.pptbidm.stat.fju.edu.tw:81/STATISTICA-WEBCAST/STATISTICA-DM/DM1/K-Means%20Cluster.ppt http://mathworld.wolfram.com/K-MeansClusteringAlgorithm.html -- wolfram mathworld

    *

    DSS

    **