Markov Chain Monte Carlo (MCMC)

《Markov Chain Monte Carlo (MCMC)》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。

1、1MarkovChainMonteCarlo(MCMC)ByStevenF.ArnoldProfessorofStatistics-PennStateUniversitySomereferencesforMCMCare1.Tanner,M.(1993)ToolsforStatisticalInference,MethodforExplorationofPosteriorDistributionsandLikelihoodFunc-tions.2.Gilks,W.,Richardson,S.andSpiegelhalter,D.(1996)M

2、arkovChainMonteCarloinPractice.3.Gelman,A.,Carlin,J.,Stern,HandRubin,D.(1995)BayesianDataAnalysis.AreferenceforMarkovChainsis1.Ross,Sheldon,(1989)IntroductiontoProbabilitymodels4thEdit.11.1MCMCandBayesianStatisticsInthelast15yearstherehasbeenanexplosionofworkinBayesianstat

3、istics.AsyourecallaBayesianstatisticianchoosesapriordistributionovertheparameterspace.Hethendeterminestheposteriordistribution.AsDr.Leonardobserved,onceweknowtheposteriordistribution,Bayesiananalysisisoftenfairlyeasy.Oftenchoosingthepriorandcomputingtheposteriorarethehardp

4、arts.Inthepast,oneoftheproblemswithBayesianstatisticshasbeenˉndingtheposteriordistribution.InrecentyearsthisproblemhasbeencontrolledbyusingMCMCtosimulatetheposterior.21.2MarkovchainsAdiscretetimeMarkovChainisasequenceofrandomvariablesinwhichtheconditionaldistributionofapre

5、sentobservationsgivenasetofpastobservationsonlydependsonthepastthroughthemostrecentobservation.Insymbols3′k1

6、Markovchainsaretime-homogenious.3Example(symmetricrandomwalk(drunkardswalk))ThisisaMarkovchainonthesetofallintegersinwhich(Xt?1+1withp=:5XtjXt?1=Xt?1?1withp=:5ThepossiblevaluesfortheMarkovchainarecalledthestatesoftheMarkovchain.Astationarydistribution?foraMarkovchainisadis

7、tributionoverthestatessuchthatifwestarttheMarkovchainin?,westayin?.Alimitingdistribution?;isadistributionoverthestatessuchthatwhateverthestartingthedistribution?0,theMarkovchainconvergesto?:Itiseasilyseenthatifthereisalimitingdistribution?;thenitisunique,anditistheonlystat

8、ionarydistribution.Itiseasiertoˉndastationarydistributionthanalimitingdistribution.Sotoˉn