Simulating Text With Markov Chains in Python – Towards Data Science

《Simulating Text With Markov Chains in Python – Towards Data Science》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、ApplausefromLudovicBenistantand60othersBenShaverFollowDataScienceImmersivestudent @GADCDec22·3minreadSimulatingTextWithMarkovChainsinPythonInmylastpost,IintroducedMarkovchainsinthecontextofMarkovchainMonteCarlomethods.Thispostisasmalladdendumtothatone,dem

2、onstratingonefunthingyoucandowithMarkovchains:simulatetext.PixabayAMarkovchainisasimulatedsequenceofevents.Eacheventinthesequencecomesfromasetofoutcomesthatdependononeanother.Inparticular,eachoutcomedetermineswhichoutcomesarelikelytooccurnext.InaMarkovcha

3、in,alloftheinformationneededtopredictthenexteventiscontainedinthemostrecentevent.ThatmeansthatknowingthefullhistoryofaMarkovchaindoesn’thelpyoupredictthenextoutcomeanybetterthanonlyknowingwhatthelastoutcomewas.Markovchainsaren’tgenerallyreliablepredictors

4、ofeventsinthenearterm,sincemostprocessesintherealworldaremorecomplexthanMarkovchainsallow.Markovchainsare,however,usedtoexaminethelong-runbehaviorofaseriesofeventsthatarerelatedtooneanotherbyxedprobabilities.Foranysequenceofnon-independenteventsintheworl

5、d,andwherealimitednumberofoutcomescanoccur,conditionalprobabilitiescanbecomputedrelatingeachoutcometooneanother.Oftenthissimplytakestheformofcountinghowoftencertainoutcomesfollowoneanotherinanobservedsequence....Togenerateasimulationbasedonacertaintext,co

6、untupeverywordthatisused.Then,foreveryword,storethewordsthatareusednext.Thisisthedistributionofwordsinthattextconditionalontheprecedingword.InordertosimulatesometextfromDonaldTrump,let’suseacollectionofhisspeechesfromthe2016campaignavailablehere.Firstimpo

7、rtnumpyandthetextlecontainingTrump’sspeeches:importnumpyasnptrump=open('speeches.txt',encoding='utf8').read()Then,splitthetextleintosinglewords.Notewe’rekeepingallthepunctuationin,sooursimulatedtexthaspunctuation:corpus=trump.split()Then,wedeneafunctio

8、ntogiveusallthepairsofwordsinthespeeches.We’reusinglazyevaluation,andyieldingageneratorobjectinsteadofactuallyllingupourmemorywitheverypairofwords:defmake_pairs(corpus):foriinrange(len(corpus)-1):yield(corpus[i],co