SIMCA-P軟件使用指南剖析課件

WhatisMultivariateAnalysisMultivariateanalysisisthebestwaytosummarizeadatatableswithmanyvariablesbycreatingafewnewvariablescontainingmostoftheinformation.Thesenewvariablesarethenusedforproblemsolvinganddisplay,i.e.,classification,relationships,controlcharts,andmore.Thenewvariables,thescores,denotedbyt,arecreatedasweightedlinearcombinationsoftheoriginalvariables.Eachobservationshast-values.PCA,thebasicMVmethod,summarizesonedatatable.Plottingthescores(t’s)givesanoverviewoftheobservations(objects)PLSsummarizessimultaneously2datatables(Xthepredictorvariables)and

(Ytheresponsevariables)inordertodeveloparelationshipbetweenthemPCAandPLSarecalledProjectionmethods7/24/20231SIMCA-PGettingstarted.pptWhatisMultivariateAnalysisMWhatisaProjection?

Reductionofdimensionality,modelinlatentvariablesAlgebraicallySummarizestheinformationintheobservationsasafewnew(latent)variablesGeometricallyTheswarmofpointsinaKdimensionalspace

(K=numberofvariables)isapproximatedbya(hyper)planeandthepointsareprojectedonthatplane.7/24/20232SIMCA-PGettingstarted.pptWhatisaProjection?

ReductioNotation

Eachobshasvaluesoft(andu)–Eachvariablehasvaluesofp(andwandc)t:theXscores;thenewsummarizingvariables(coordinatesinthehyperplaneofX-space)u:theYscoresinPLS;thenewsummarizingvariables(coordinatesinthehyperplaneofY-space,whenYismultidimensional)p:thePCloadings.ThesearetheweightsthatinPCAcombinetheoriginalvariablesinXtoformthenewvariables,scorest.w*:thePLSweights.ThesearetheweightsthatinPLScombinetheoriginalvariablesinXtoformthenewvariables,scorest.c:theweightsusedtocombinetheY'stoformthescoresu.7/24/20233SIMCA-PGettingstarted.pptNotation

EachobshasvaluesoNotation

Eachobshasvaluesoft(andu)–Eachvariablehasvaluesofp(andwandc)OneComponentconsistsofonetandonep(PCA)ort,p,w,u,c(PLS).ThetotalnumberofcomponentsisA.Model:Thedataareapproximatedbyaplaneorhyperplane,(themodel)withasmanydimensionsascomponentsextracted.DModX:alsocalledDistancetothemodel,isthedistanceofagivenobservationtothemodelplane.T2:Hotelling’sT2,isacombinationofallthescores(t)ofallAcomponents.T2measureshowfarawayanobservationisfromthecenterofaPCorPLSmodel.7/24/20234SIMCA-PGettingstarted.pptNotation

EachobshasvaluesoNotationR2X:ThefractionofthevariationoftheXvariablesexplainedbythemodel.R2Y:ThefractionofthevariationoftheYvariablesexplainedbythemodel.Q2X:ThefractionofthevariationoftheXvariablespredictedbythemodel.Q2Y:ThefractionofthevariationoftheYvariablespredictedbythemodel.7/24/20235SIMCA-PGettingstarted.pptNotationR2X:ThefractionofMVA–SIMCARoadMap

MethodsavailablePreprocessing;trimmingandWinsorizing(takeawayextremes)PrincipalComponentsAnalysis(PCA;overviewofdata)ProjectiontoLatentStructures(PLS;relationshipsXY)SimcaclassificationPLS-discriminantanalysis(classification)HierarchicalPCAandPLSPredictionsandclassificationofnewdatausinganymodel7/24/20236SIMCA-PGettingstarted.pptMVA–SIMCARoadMap

MethodsaMVA–SIMCARoadMap

Dataset=alldata;Workset=workingcopyofdataWorkmainmenusfromlefttorightandpop-upmenusfromuptodownPlot/Listallowsyoutoplotorlistanythingnon-standard,notfoundunderAnalysis7/24/20237SIMCA-PGettingstarted.pptMVA–SIMCARoadMap

DatasetStepsinusingSIMCA-PusingthewizardStartanewprojectandimportthedatasetUsetheworksetwizardtoguidethroughbuildingtheworksetandfittingthemodelGeneratethereportwritertowalkthroughthemodelresultsandinterpretationWhendisplayingSimca-PplotsalwaysusetheAnalysisadvisertoguideyou.7/24/20238SIMCA-PGettingstarted.pptStepsinusingSIMCA-PusingWorksetwizardonON7/24/20239SIMCA-PGettingstarted.pptWorksetwizardonON7/24/20239SWorksetwizard7/24/202310SIMCA-PGettingstarted.pptWorksetwizard7/24/202310SIMCAAutotransformvariables

Totransformallvariablesifanyneeded,markthecheckbox7/24/202311SIMCA-PGettingstarted.pptAutotransformvariables

TotraAutomaticcreationofclassesforclassificationordiscrimination7/24/202312SIMCA-PGettingstarted.pptAutomaticcreationofclassesSelectionandFitofmodel7/24/202313SIMCA-PGettingstarted.pptSelectionandFitofmodel7/24Reportwriter

Walksyouthroughthemodelresultswithinterpretation:File|GenerateReport7/24/202314SIMCA-PGettingstarted.pptReportwriter

WalksyouthrouStepsinUsingSIMCA-P,AdvancedModeStartanewprojectandimportthedatasetExploreandpreprocessthedataMakeworkingcopyofselecteddata(workset)formodelbuildingSpecifymodeltypeandfitittotheworksetReviewfit(plots,diagnostics,coefficients,etc.)PredictionsGenerateReport7/24/202315SIMCA-PGettingstarted.pptStepsinUsingSIMCA-P,Advanc1a.FileNew

StartinganewprojectSelectthedatafilecontainingtherawdataoftheprojectdirectory,filetype(XLS,DIF,TXT,…..),filenameAWizardopens(seenextpage)allowingyoutospecify(optionally)therowcontainingtheVariablenames,and(optionally)thecolumnswiththeObs.NumbersandNamesHere(Commands)youcanalsodoadditionalthingssuchastransposingtheinputdatamatrixUsesimplemodewithworksetwizardAtthelastWizardpage,youcan(optionally)specifyanothernameanddirectoryfortheproject.AmapofthemissingdataisshownTheWizardfinishesandputsyouintheSimca-windowAstartingworkset(M1,alldata,allX-s,UV-scaled)isready7/24/202316SIMCA-PGettingstarted.ppt1a.FileNew

Startinganewpr1b.ThesecondscreenoftheWizard7/24/202317SIMCA-PGettingstarted.ppt1b.ThesecondscreenoftheW2.LookingatthedataWiththedatasettableopen(Datasetedit):QuickInfo(bothvarandobswindowscanbeopen)variablesobservationsMovingthecursorinthedatasettableupanddown,orsidewise,changesthedisplayedvariableandobservationInthequickinfooptionsyoucanspecifywhatyouwanttolookat(histograms,auto-correlations,…),aswellaswhichitemsshouldbethebasisfortheplots7/24/202318SIMCA-PGettingstarted.ppt2.LookingatthedataWiththeViewvariablesorObservations,Trim,etc.

QuickInfo7/24/202319SIMCA-PGettingstarted.pptViewvariablesorObservations3.Prepareaworkcopy:TheWorkset

SimpleModewithguidance,orAdvancedModeInWorkset,youprepareaworkingcopyofthepartofthedatayouwillanalyze,i.e.,useasthebasisofyourmodel.Hereyouspecifytransformation,scaling,androlesofvariables(XorYorexcluded).Also,youselecttheobservations(your“trainingset”).Youcanstartwiththepreviousworkset(Workset/Newasmodelxx)andthenmodifyit,e.g.,excludingobservations.WhateveryoudoinWorksetdoesNOTtouchtherawdataNotethatoutliersarejustspecifiedas“notincluded”inthenextworkset(the“polished”data).OutliersareNEVERremovedfromtherawdataset.7/24/202320SIMCA-PGettingstarted.ppt3.Prepareaworkcopy:TheWoWorkset:twoModes,SimpleandAdvanced7/24/202321SIMCA-PGettingstarted.pptWorkset:twoModes,Simpleand4.Analysis

FittheModeltotheWorksetDataEithermenu“Analysis/Autofit”orFastButtonAmodelwithappropriatenumberofcomponentsisfoundIfnothinghappens,getthetwofirstcomponents

(alsomenuorfastbutton)Atableappearsshowingthemodel,componentbycomponent.Morecomponentscanbeadded(menuorfastbutton)Doubleclickonamodeltospecifyatitle7/24/202322SIMCA-PGettingstarted.ppt4.Analysis

FittheModeltot5.Plotresults

Analysis/menu(orfastbuttons)Summary/X/Y-OverviewshowsR2andQ2forallvar.sScores–scatterplot,t1-t2andt1-u1&t2-u2(PLS)Loadings–scatterplot(p1-p2froPCA,wc1-wc2forPLS)DistancetoModel–lineplotContributionplotstointerpretinterestingobservations,e.g.outliers,jumps,…Forallplots,therightmousebutton,propertiesallowschoiceofplotmarkers,andmoreThegraphicaltoolboxallowsfurthermodifications7/24/202323SIMCA-PGettingstarted.ppt5.Plotresults

Analysis/men6a.Outlierswereseeninthescoreplot

(welloutsidetheHotellingellipse)Startanotherworkset (eitherfromWorkset/Newasmodelxx,orusingthegraphicaltool-boxtoremoveoutliersfromthescoreplot)NotethatoutliersshouldNOTbedeletedfromthedatabyEdit/DatasetWhenthenewworksetisall-right,returnto“4.Analysis”tofitanewmodeltothenewworkset (fastbuttonorAnalysis/Autofit)7/24/202324SIMCA-PGettingstarted.ppt6a.Outlierswereseeninthe6b.Nooutlierswereseeninthescoreplots

(ortheyhavebeenexcluded,andthescoreplotsnowlookall-right)Now,interpretthemodelLookat“patterns”,trends,etc.,inthescoreplotsInspecttheloadingplotstointerprettheabovepatternsLookatDModXWhatdothesepatternssayabouttheobjectiveoftheinvestigation?7/24/202325SIMCA-PGettingstarted.ppt6b.NooutlierswereseenintAnalysisAdvisortounderstandandinterpretmodelresults7/24/202326SIMCA-