层次聚类课件.pptVIP

层次聚类

HierarchicalClustering

HierarchicalClustering

•Producesasetofnestedclusters

organizedasahierarchicaltree

•Canbevisualizedasadendrogram

–Atreelikediagramthatrecordsthe

sequencesofmergesorsplits

65

0.24

34

2

0.155

2

0.1

1

0.0531

0

132546

StrengthsofHierarchical

Clustering

•Donothavetoassumeanyparticular

numberofclusters

–Anydesirednumberofclusterscanbe

obtainedby‘cutting’thedendogramatthe

properlevel

•Theymaycorrespondtomeaningful

taxonomies

–Exampleinbiologicalsciences(e.g.,animal

kingdom,phylogenyreconstruction,…)

•Twomaintypesofhierarchicalclustering

–Agglomerative(凝聚):

•Startwiththepointsasindividualclusters

•Ateachstep,mergetheclosestpairofclustersuntilonlyone

cluster(orkclusters)left

–Divisive(分裂的):

•Startwithone,all-inclusivecluster

•Ateachstep,splitaclusteruntileachclustercontainsa

point(ortherearekclusters)

•Traditionalhierarchicalalgorithmsusea

similarityordistancematrix

–Mergeorsplitoneclusteratatime

AnAgglomerativeClusteringAlgorithm

•Basicalgorithmisstraightforward

1.Computetheproximity(接近程度）matrix

2.Leteachdatapointbeacluster

3.Repeat

4.Mergethetwoclosestclusters

5.Updatetheproximitymatrix

6.Untilonlyasingleclusterremains

•Keyoperationisthecomputationoftheproximityof

twoclusters

–Differentapproachestodefiningthedistance

betweenclustersdistinguishthedifferent

algorithms

StartingSituation

•Startwithclustersofindividualpointsand

aproximitymatrix