Python的隐马尔科夫HMMLearn库的应用教学.doc

《Python的隐马尔科夫HMMLearn库的应用教学.doc》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、PythonHMMLearnTutorialEditedBy毛片物语hmmlearn implementstheHiddenMarkovModels(HMMs).TheHMMisagenerativeprobabilisticmodel,inwhichasequenceofobservable (mathbf{X}) variablesisgeneratedbyasequenceofinternalhiddenstates (mathbf{Z}).Thehiddenstatesarenotbeobserveddirectly.Thetrans

2、itionsbetweenhiddenstatesareassumedtohavetheformofa(first-order)Markovchain.Theycanbespecifiedbythestartprobabilityvector (boldsymbol{pi}) andatransitionprobabilitymatrix (mathbf{A}).Theemissionprobabilityofanobservablecanbeanydistributionwithparameters (boldsymbol{thet

3、a}) conditionedonthecurrenthiddenstate.TheHMMiscompletelydeterminedby (boldsymbol{pi}), (mathbf{A}) and (boldsymbol{theta}).TherearethreefundamentalproblemsforHMMs:·Giventhemodelparametersandobserveddata,estimatetheoptimalsequenceofhiddenstates.·Giventhemodelparameter

4、sandobserveddata,calculatethelikelihoodofthedata.·Givenjusttheobserveddata,estimatethemodelparameters.ThefirstandthesecondproblemcanbesolvedbythedynamicprogrammingalgorithmsknownastheViterbialgorithmandtheForward-Backwardalgorithm,respectively.ThelastonecanbesolvedbyaniterativeEx

5、pectation-Maximization(EM)algorithm,knownastheBaum-Welchalgorithm.References:[Rabiner89]LawrenceR.Rabiner“AtutorialonhiddenMarkovmodelsandselectedapplicationsinspeechrecognition”,ProceedingsoftheIEEE77.2,pp.257-286,1989.[Bilmes98]JeffA.Bilmes,“AgentletutorialoftheEMalgorithmandit

6、sapplicationtoparameterestimationforGaussianmixtureandhiddenMarkovmodels.”,1998.Availablemodelshmm.GaussianHMMHiddenMarkovModelwithGaussianemissions.hmm.GMMHMMHiddenMarkovModelwithGaussianmixtureemissions.hmm.MultinomialHMMHiddenMarkovModelwithmultinomial(discrete)emissionsReadon

7、 fordetailsonhowtoimplementanHMMwithacustomemissionprobability.BuildingHMMandgeneratingsamplesYoucanbuildanHMMinstancebypassingtheparametersdescribedabovetotheconstructor.Then,youcangeneratesamplesfromtheHMMbycalling sample.>>>importnumpyasnp>>>fromhmmlearnimporthmm>>>np.random.s

8、eed(42)>>>model=hmm.GaussianHMM(n_compon