资源描述:
《A Tutorial on Principal Component Analysis (Jonathon Shlens)》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、ATutorialonPrincipalComponentAnalysisJonathonShlens∗CenterforNeuralScience,NewYorkUniversityNewYorkCity,NY10003-6603andSystemsNeurobiologyLaboratory,SalkInsituteforBiologicalStudiesLaJolla,CA92037(Dated:April22,2009;Version3.01)Principalcomponentanalysis(PCA)isamainstayofmoder
2、ndataanalysis-ablackboxthatiswidelyusedbut(sometimes)poorlyunderstood.Thegoalofthispaperistodispelthemagicbehindthisblackbox.Thismanuscriptfocusesonbuildingasolidintuitionforhowandwhyprincipalcomponentanalysisworks.Thismanuscriptcrystallizesthisknowledgebyderivingfromsimpleint
3、uitions,themathematicsbehindPCA.Thistutorialdoesnotshyawayfromexplainingtheideasinformally,nordoesitshyawayfromthemathematics.Thehopeisthatbyaddressingbothaspects,readersofalllevelswillbeabletogainabetterunderstandingofPCAaswellasthewhen,thehowandthewhyofapplyingthistechnique.
4、I.INTRODUCTIONII.MOTIVATION:ATOYEXAMPLEPrincipalcomponentanalysis(PCA)isastandardtoolinmod-Hereistheperspective:weareanexperimenter.Wearetryingerndataanalysis-indiversefieldsfromneurosciencetocom-tounderstandsomephenomenonbymeasuringvariousquan-putergraphics-becauseitisasimple,
5、non-parametricmethodtities(e.g.spectra,voltages,velocities,etc.)inoursystem.forextractingrelevantinformationfromconfusingdatasets.Unfortunately,wecannotfigureoutwhatishappeningbe-WithminimaleffortPCAprovidesaroadmapforhowtore-causethedataappearsclouded,unclearandevenredundant.d
6、uceacomplexdatasettoalowerdimensiontorevealtheThisisnotatrivialproblem,butratherafundamentalobstaclesometimeshidden,simplifiedstructuresthatoftenunderlieit.inempiricalscience.Examplesaboundfromcomplexsys-temssuchasneuroscience,webindexing,meteorologyandThegoalofthistutorialisto
7、providebothanintuitivefeelforoceanography-thenumberofvariablestomeasurecanbePCA,andathoroughdiscussionofthistopic.Wewillbeginunwieldyandattimesevendeceptive,becausetheunderlyingwithasimpleexampleandprovideanintuitiveexplanationrelationshipscanoftenbequitesimple.ofthegoalofPCA.
8、Wewillcontinuebyaddingmathemati-calrigortoplaceitwithinthefra