birch

Package‘birch’

August29,2013

TypePackage

DependsR(>=2.10),ellip

SuggestsMASS

TitleDealingwithverylargedatatsusingBIRCH

Version1.2-3

Date2012-05-03

AuthorLysianeCharest,JustinHarrington,MatiasSalibian-Barrera

MaintainerLysianeCharest

Descriptionbirchisanimplementationofthealgorithmsdescribedin

Zhangetal(1997),andprovidesfunctionsforcreating

CF-trees,alongwithalgorithmsfordealingwithsome

combinatorialproblems,ry

wellsuitedfordealingwithverylargedatats,anddoesnot

requirethatthedatacanfitinphysicalmemory.

LicenGPL

RepositoryCRAN

Date/Publication2012-05-0405:24:11

NeedsCompilationyes

Rtopicsdocumented:

birch-package........................................2

birch.............................................3

birchObj...........................................6

........................................7

..........................................9

genericmethods.......................................10

1

2birch-package

kendall............................................13

........................................15

...........................................16

..........................................18

spearman..........................................20

Index23

birch-packageWorkingwithverylargedatatsusingBIRCH.

Description

Thefunctionsinthispackagearedesignedforworkingwithverylargedatatsbypre-processing

thedatatwithanalgorithmcalledBIRCH(BalancedIterativeReducingandClusteringusing

Hierarchies),whichtransformsthedatatintocompact,locallysimilarsubclusters,eachwith

summarystatisticsattached(calledclusteringfeatures).Then,insteadofusingthefulldatat,

thesummarystatisticscanbeud.

Thisapproachismostadvangeousintwosituations:whenthedatacannotbeloadedintomemory

duetoitssize;and/orwhensomeformofcombinatorialoptimizationisrequiredandthesizeofthe

solutionspacemakesfindingglobalmaximums/minimumsdifficult.

AcompleteexplanationofthispackageisgiveninHarringtonandSalibian-Barrera(2008),and

discussionoftheunderlyingalgorithmscanbefoundinHarringtonandSalibian-Barrera(2010).

Doc,thesource

codecontainsdOxygentagsforfurtherinformation.

Details

Themainfunctionis

birchtakesadatat(anRobject,textfile,etc),andcreatesabirchobject

Variousgenericmethodsareprent,including

print

summary

plot

Methodsforestimatingthecorrelationmatrixarealsoavailable:

spearman

kendall

birch3

Finally,nclude:

inimumCovarianceDeterminant(robustestimatorforlocationanddispersion)

eastTrimmedSquares(robustregressionestimator)

obustLinearGroupingAnalysis(robustclusteringabouthyperplanes)

-means

Author(s)

LysianeCharest,JustinHarringtonand

MatiasSalibian-Barrera

References

Harrington,ibian-Barrera,M.(2010),“FindingApproximateSolutionstoCombinatorial

ProblemswithVeryLargeDatatsusingBIRCH”,ComputationalStatisticsandDataAnalysis54,

655-667.

Harrington,ibian-Barrera,M.(2008),“birch:Workingwithverylargedatats”,working

paper.

SeeAlso

birch,,,,,kendall,spearman.

birchCreateabirchobject

Description

ThisfunctioncreatesabirchobjectusingthealgorithmBIRCH.

Usage

birch(x,radius,compact=radius,keeptree=FALSE,columns=NULL,...)

ree(x,birchObject,updateDIM=TRUE,...)

e(birchObject)

ee(birchObject)

Arguments

xanumericmatrixofatleasttwocolumns,afilenameoraconnectionthatis

.

radiustheclonesscriterion

compactthecompactnesscriterion

keeptreeABoolean,whethertokeeptheCFtreeinmemory.

4birch

columnsthechoiceofvariablestouifxisafilenameoraconnection(defaultisall

variables)

...orloadingafileorconnection

updateDIMUpdatethedimensionoftheobject?DefaultstoTRUE(whichisdesirable!).

birchObjectTheoutputfrombirch.

Details

ThisfunctioncreatesaCF-TreenotunlikethatdescribedinZhangetal.(1997),andudinHar-

ringtonandSalibian-Barrera(2010).AcompleteexplanationofthispackageisgiveninHarrington

andSalibian-Barrera(2008).

Afulltreestructureisud,asisthesplittingofnodes(asdescribedintheoriginalarticle).How-

ever,the‘MergingRefinement’o-

maticrebuildingbadonpagesizeisnotimplemented.

Theargumentkeeptreeallowsforthetreetobekeptinmemoryaftertheinitialprocessing

lowsforadditionalinformationtobeaddedatalatterstagewith

ree,r,itshould

benotedthattheCFdata(thesummarystatisticsofeachsubcluster,asudbythesubquental-

gorithms)isnotreturned,ecommand

rwords,

##Createthetree

myobject<-birch(x,