Python機(jī)器學(xué)習(xí)教程

想要理解和研究機(jī)器學(xué)習(xí),首先你應(yīng)該要掌握Python或者R,都是和C,Java,PHP差不多的語言(譯:差太多了好吧).不過呢,Python和R都是比較年輕(譯:不懂,Python可并不年輕吧),而且呢更高級,完全不用理解底層(譯:?),所以他倆都很容易學(xué).Python更牛逼的地方在于她能夠處理更多的問題,比如,機(jī)器學(xué)習(xí),算法,圖像等,而不像R只能是進(jìn)行數(shù)據(jù)處理和分析.Python有著更廣泛的應(yīng)用領(lǐng)域,比如后端框架Django(譯:原文是,'Hostingwebsites:Jango'),自然語言處理(譯:原文是,'naturallanguageproecssing',作者太不認(rèn)真,NLP),網(wǎng)站接入等,而且Python更像C語言(譯:扯淡),所以她現(xiàn)在很流行.毛子的原文里面有不少錯(cuò)誤,我以自己的理解加以修正,僅供參考.語法文法錯(cuò)誤我就直接修改,原文作者的表達(dá)內(nèi)容錯(cuò)誤會依據(jù)原文不變,在()內(nèi)說明.新手用Python進(jìn)行機(jī)器學(xué)習(xí)的四個(gè)步驟Python基礎(chǔ)知識學(xué)習(xí),有書,Mooc,視頻.處理數(shù)據(jù),你得了解一些模塊,如:Pandas,Numpy,Matplotlib和NaturalLanguageProcessing.接著你就得爬取數(shù)據(jù),可以通過API,也可以直接到網(wǎng)站上去爬取.網(wǎng)站爬蟲模塊:BeautifulSoup(譯:應(yīng)該是Scrapy,BS是HTML/XML解析器).我們用拿到的數(shù)據(jù)來訓(xùn)練算法.最后一步,就是要學(xué)習(xí)ML的相關(guān)算法,以及工具Scikit-learn.1.學(xué)習(xí)Python學(xué)習(xí)Python最簡單粗暴的法子就是到Codecademy上去注冊個(gè)賬號來學(xué)習(xí)基礎(chǔ)知識.一個(gè)被好多碼農(nóng)推薦的很經(jīng)典的網(wǎng)站LearnPythonTheHardWay.ByteofPython這篇文章是非常值得去學(xué)習(xí)的.Python社區(qū)還為新手給出了一個(gè)Python學(xué)習(xí)資源列表.O’Reilley出版的一本書ThinkPython,這里可以免費(fèi)下載.最后還有一個(gè)IntroductiontoPythonforEconometrics,StatisticsandDataAnalysis也講了好多Python的基礎(chǔ)知識.2.導(dǎo)入模塊做機(jī)器學(xué)習(xí)很重要的幾個(gè)模塊和工具是NumPy,Pandas,Matplotlib和IPython.DataAnalysiswithOpenSourceTools這本書里面都有涉及這些內(nèi)容.上面提到的IntroductiontoPythonforEconometrics,StatisticsandDataAnalysis也涵蓋了這些東西.還有一本書PythonforDataAnalysis:DataWranglingwithPandas,NumPy,andIPython.下面還有一些免費(fèi)的資源:10minutestoPandasPandasformachinelearning100NumPyexercises3.爬取挖掘數(shù)據(jù)一旦你掌握了Python的基礎(chǔ),下面就要學(xué)會怎么去爬取數(shù)據(jù).也就是網(wǎng)頁爬蟲.像Twitter和LinkedIn這些網(wǎng)站都給出了APIs接口,讓我們?nèi)カ@得文本數(shù)據(jù).關(guān)于這方面下面有幾本書不錯(cuò)的書:MiningtheSocialWeb(免費(fèi)),WebScrapingwithPython和WebScrapingwithPython:CollectingDatafromtheModernWeb.最后這些文本數(shù)據(jù)要由NLP技術(shù)處理成數(shù)值化數(shù)據(jù):NaturallanguageprocessingwithPython.圖像和視頻要用圖像處理CV,下面有幾個(gè)不錯(cuò)的資源:ProgrammingComputerVisionwithPython(免費(fèi)),ProgrammingComputerVisionwithPython:Toolsandalgorithmsforanalyzingimages和PracticalPythonandOpenCV.Python爬蟲的一些例子:Mini-Tutorial:SavingTweetstoaDatabasewithPythonWebScrapingIndeedforKeyDataScienceJobSkillsCaseStudy:SentimentAnalysisOnMovieReviewsFirstWebScraperSentimentAnalysisofEmailsSimpleTextClassificationBasicSentimentAnalysiswithPythonTwittersentimentanalysisusingPythonandNLTKSecondTry:SentimentAnalysisinPythonNaturalLanguageProcessinginaKaggleCompetitionforMovieReviews4.機(jī)器學(xué)習(xí)機(jī)器學(xué)習(xí)可以分為四部分:分類,聚類,回歸和降維.MachinelearninginPythonScikit-learn官網(wǎng)上有很多指南,下面列一些其它的:IntroductiontoMachineLearningwithPythonandScikit-LearnDataScienceinPythonMachineLearningforPredictingBadLoansAGenericArchitectureforTextClassificationwithMachineLearningUsingPythonandAItopredicttypesofwineAdviceforapplyingMachineLearningPredictingcustomerchurnwithscikit-learnMappingYourMusicCollectionDataScienceinPythonCaseStudy:SentimentAnalysisonMovieReviewsDocumentClusteringwithPythonFivemostpopularsimilaritymeasuresimplementationinpythonCaseStudy:SentimentAnalysisonMovieReviewsWillitPython?TextProcessinginMachineLearningHackinganepicNHLgoalcelebrationwithahuelightshowandreal-timemachinelearningVancouverRoomPricesExploringandPredictingUniversityFacultySalariesPredictingAirlineDelays書:CollectionofbooksonredditBuildingMachineLearningSystemswithPythonBuildingMachineLearningSystemswithPython,2ndEditionLearningscikit-learn:MachineLearninginPythonMachineLearningAlgorithmicPerspectiveDataSciencefromScratch–FirstPrincipleswithPythonMachineLearninginPython機(jī)器學(xué)習(xí)相關(guān)的Blog和課程在線課程:Collectionoflinks.MOOC:machinelearning和DataAnalystNanodegree.

這里是一些Blog.機(jī)器學(xué)習(xí)理論TheElementsofstatisticalLearningIntroductiontoStatisticalLearning書:IntroductiontomachinelearningACourseinMachineLearning.還有一些Watch15hourstheoryofmachinelearning!越看越懶得翻,著實(shí)沒什么營養(yǎng),索性直接列出資源.下面是美國麻省理工學(xué)院（MIT）博士林達(dá)華老師(ML大牛)推薦的書單.MachineLearningPatternRecognitionandMachineLearningByChristopherM.Bishop

Anewtreatmentofclassicmachinelearningtopics,suchasclassification,regression,andtimeseriesanalysisfromaBayesianperspective.ItisamustreadforpeoplewhointendstoperformresearchonBayesianlearningandprobabilisticinference.GraphicalModels,ExponentialFamilies,andVariationalInferenceByMartinJ.WainwrightandMichaelI.Jordan

Itisacomprehensiveandbrilliantpresentationofthreecloselyrelatedsubjects:graphicalmodels,exponentialfamilies,andvariationalinference.ThisisthebestmanuscriptthatIhaveeverreadonthissubject.Stronglyrecommendedtoeveryoneinterestedingraphicalmodels.Theconnectionsbetweenvariousinferencealgorithmsandconvexoptimizationisclearlyexplained.Note:pdfversionofthisbookisfreelyavailableonline.BigData:ARevolutionThatWillTransformHowWeLive,Work,andThinkViktorMayer-Schonberger,andKennethCukier

Ashortbutinsightfulmanuscriptthatwillmotivateyoutorethinkhowweshouldfacetheexplosivegrowthofdatainthenewcentury.StatisticalPatternRecognition(2nd/3rdEdition)ByAndrewR.Webb,andKeithD.Copsey

Awellwrittenbookonpatternrecognitionforbeginners.Itcoversbasictopicsinthisfield,includingdiscriminantanalysis,decisiontrees,featureselection,andclustering--allarebasicknowledgethatresearchersinmachinelearningorpatternrecognitionshouldunderstand.LearningwithKernels:SupportVectorMachines,Regularization,Optimization,andBeyondByBernhardSchlkopfandAlexanderJ.Smola

Acomprehensiveandin-depthtreatmentofkernelmethodsandsupportvectormachine.Itnotonlyclearlydevelopsthemathematicalfoundation,namelythereproducingkernelHilbertspace,butalsogivesalotofpracticalguidance(e.g.howtochooseordesignkernels.)MathematicsTopology(2ndEdition)ByJamesMunkres

Aclassicontopologyforbeginners.Itprovidesaclearintroductionofimportantconceptsingeneraltopology,suchascontinuity,connectedness,compactness,andmetricspaces,whicharethefundamentalsthatyouhavetograspedbeforeembarkingonmoreadvancedsubjectssuchasrealanalysis.IntroductoryFunctionalAnalysiswithApplicationsByErwinKreyszig

ItisaverywellwrittenbookonfunctionalanalysisthatIwouldliketorecommendtoeveryonewhowouldliketostudythissubjectforthefirsttime.Startingfromsimplenotionssuchasmetricsandnorms,thebookgraduallyunfoldsthebeautyoffunctionalanalysis,exposingimportanttopicsincludingBanachspaces,Hilbertspaces,andspectraltheorywithareasonabledepthandbreadth.Mostimportantconceptsneededinmachinelearningarecoveredbythisbook.Theexercisesareofgreathelptoreinforceyourunderstanding.RealAnalysisandProbability(CambridgeStudiesinAdvancedMathematics)ByR.M.Dudley

ThisisadensetextthatcombinesRealanalysisandmodernprobabilitytheoryin500+pages.WhatIlikeaboutthisbookisitstreatmentthatemphasizestheinterplaybetweenrealanalysisandprobabilitytheory.Alsotheexpositionofmeasuretheorybasedonsemi-ringsgivesadeepinsightofthealgebraicstructureofmeasures.ConvexOptimizationByStephenBoyd,andLievenVandenberghe

Aclassiconconvexoptimization.EveryonethatIknewwhohadreadthisbooklikedit.Thepresentationstyleisverycomfortableandinspiring,anditassumesonlyminimalprerequisiteonlinearalgebraandcalculus.Stronglyrecommendedforanybeginnersonoptimization.Note:thepdfofthisbookisfreelyavailableontheProf.Boyd'swebsite.NonlinearProgramming(2ndEdition)ByDimitriP.Bersekas

Athoroughtreatmentofnonlinearoptimization.Itcoversgradient-basedtechniques,Lagrangemultipliertheory,andconvexprogramming.PartofthisbookoverlapswithBoyd's.Overall,itgoesdeeperandtakesmoreeffortstoread.IntroductiontoSmoothManifoldsByJohnM.Lee

ThisisthebookthatIusedtolearndifferentialgeometryandLiegrouptheory.Itprovidesadetailedintroductiontobasicsofmoderndifferentialgeometry--manifolds,tangentspaces,andvectorbundles.TheconnectionsbetweenmanifoldtheoryandLiegrouptheoryisalsoclearlyexplained.ItalsocoversDeRhamCohomologyandLiealgebra,whereaudienceisinvitedtodiscoverthebeautybylinkinggeometrywithalgebra.ModernGraphTheoryByBelaBollobas

Itisamoderntreatmentofthisclassicaltheory,whichemphasizestheconnectionswithothermathematicalsubjects--forexample,randomwalksandelectricalnetworks.Ifoundsomemessagesconveyedbythisbookisenlighteningformyresearchonmachinelearningmethods.ProbabilityTheory:AComprehensiveCourse(Universitext)ByAchimKlenke

Thisisacompletecoverageofmodernprobabilitytheory--notonlyincludingtraditionaltopics,suchasmeasuretheory,independence,andconvergencetheorems,butalsointroducingtopicsthataretypicallyintextbooksonstochasticprocesses,suchasMartingales,Markovchains,andBrownianmotion,Poissonprocesses,andStochasticdifferentialequations.Itisrecommendedasthemaintextbookonprobabilitytheory.AFirstCourseinStochasticProcesses(2ndEdition)BySamuelKarlin,andHowardM.Taylor

AclassictextbookonstochasticprocesswhichIthinkareparticularlysuitableforbeginnerswithoutmuchbackgroundonmeasuretheory.Itprovidesacompletecoverageofmanyimportantstochasticprocessesinanintuitiveway.ItsdevelopmentofMarkovprocessesandrenewalprocessesisenlightening.PoissonProcesses(OxfordStudiesinProbability)ByJ.F.C.Kingman

IfyouareinterestedinBayesiannonparametrics,thisisthebookthatyoushoulddefinitelycheckout.Thismanuscriptprovidesanunparalleledintroductiontorandompointprocesses,includingPoissonandCoxprocesses,andtheirdeeptheoreticalconnectionswithcompleterandomness.ProgrammingStructureandInterpretationofComputerPrograms(2ndEdition)ByHaroldAbelson,GeraldJaySussman,andJulieSussman

Timelessclassicthatmustbereadbyallcomputersciencemajors.WhilesometopicsandtheuseofSchemeastheteachinglanguageseemsoddatfirstglance,thepresentationoffundamentalconceptssuchasabstraction,recursion,andmodularityissobeautifulandinsightfulthatyouwouldneverexperiencedelsewhere.ThinkinginC++:IntroductiontoStandardC++(2ndEdition)ByBruceEckel

Whileitiskindofold(writtenin2000),IstillrecommendthisbooktoallbeginnerstolearnC++.Thethoughtsunderlyingobject-orientedprogrammingisveryclearlyexplained.ItalsoprovidesacomprehensivecoverageofC++inawell-tunedpace.EffectiveC++:55SpecificWaystoImproveYourProgramsandDesigns(3rdEdition)ByScottMeyers

TheEffectiveC++seriesbyScottMeyersisamustforanyonewhoisseriousaboutC++programming.Theitems(rules)listedinthisbookconveystheauthor'sdeepunderstandingofbothC++itselfandmodernsoftwareengineeringprinciples.ThiseditionreflectslatestupdatesinC++development,includinggenericprogrammingtheuseofTR1library.AdvancedC++MetaprogrammingByDavideDiGennaro

Likeitorhateit,meta-programminghasplayedanincreasinglyimportantroleinmodernC++development.IfyouaskedwhatisthekeyaspectsthatdistinguishesC++fromallotherlanguages,IwouldsayitistheunparalleledgenericprogrammingcapabilitybasedonC++templates.Thisbooksummarizesthelatestadvancementofmetaprogramminginthepastdecade.IbelieveitwilltaketheplaceofLoki's"ModernC++Design"tobecomethebibleforC++meta-programming.IntroductiontoAlgorithms(2nd/3rdEdition)ByThomasH.Cormen,CharlesE.Leiserson,RonaldL.Rivest,andCliffordStein

Ifyouknownothingaboutalgorithms,youneverunderstandcomputerscience.Thisisbookisdefinitelyaclassiconalgorithmsanddatastructuresthateveryonewhoisseriousaboutcomputersciencemustread.Thiscontentsofthisbookrangesfromelementarytopicssuchasclassicsortingalgorithmsandhashtabletoadvancedtopicssuchasmaximumflow,linearprogramming,andcomputationalgeometry.Itisabookforeveryone.EverytimeIreadit,Ilearnedsomethingnew.DesignPatterns:ElementsofReusableObject-OrientedSoftwareByErichGamma,RichardHelm,RalphJohnson,andJohnVlissides

TextbooksonC++,Java,orotherlanguagestypicallyusetoyexamples(animals,students,etc)toillustratetheconceptofOOP.Thisway,however,doesnotreflectthefullstrengthofobjectorientedprogramming.Thisbook,whichhasbeenwidelyacknowledgedasaclassicinsoftwareengineering,showsyou,viacompellingexamplesdistilledfromrealworldprojects,howspecificOOPpatternscanvastlyimproveyourcode'sreusability