bestmachinelearningresourcesforgettingstartedmachinelearningmastery.docx

bestmachinelearningresourcesforgettingstartedmachinelearningmastery.docx

《bestmachinelearningresourcesforgettingstartedmachinelearningmastery.docx》由会员分享，可在线阅读，更多相关《bestmachinelearningresourcesforgettingstartedmachinelearningmastery.docx（6页珍藏版）》请在冰豆网上搜索。

bestmachinelearningresourcesforgettingstartedmachinelearningmastery

BestMachineLearningResourcesforGettingStartedMachineLearningMastery

ThiswasareallyhardposttowritebecauseIwantittobereallyvaluable.Isatdownwithablankpageandaskedthereallyhardquestionofwhataretheverybestlibraries,courses,papersandbooksIwouldrecommendtoanabsolutebeginnerinthefieldofMachineLearning.

Ireallyagonisedoverwhattoincludeandwhattoexclude.Ihadtoworkhardtoputmyselfintheshoesofaprogrammerandbeginneratmachinelearningandthinkaboutwhatresourceswouldbestbenefitthem.

Ipickedthebestforeachtypeofresource.Ifyouareatruebeginnerandexcitedtogetstartedinthefieldofmachinelearning,Ihopeyoufindsomethinguseful.Mysuggestionwouldbetopickonething,onebookoronelibraryandreaditcovertocoverorworkthroughallofthetutorials.Pickoneandsticktoit,thenonceyoumasterit,pickanotherandrepeat.Let’sgetintoit.

ProgrammingLibraries

Iamanadvocateof“learnjustenoughtobedangerousandstarttryingthings”.ThisishowIlearnedtoprogramandI’msuremanyotherpeoplelearnedthatwaytoo.Knowyourlimitationsandexploityourstrengths.Ifyouknowhowtoprogram,leveragethattogetdeepintomachinelearningfast.Thenhavethedisciplinetogoandlearnthemathforthetechniquebeforeyouimplementitaproductionsystem.

Findalibraryandreadthedocumentation,followthetutorialsandstarttryingthingsout.Thefollowingarethebestopensourcemachinelearningprogramminglibrariesoutthere.Idon’tthinktheyareallsuitableforusinginyourproductionsystem,buttheyareidealforlearning,exploringandprototyping.

Startwithalibraryinalanguageyouknowwellthenmoveontoothermorepowerfullibraries.Ifyou’reagoodprogrammer,youknowyoucanmovefromlanguagetolanguagereasonablyeasily.It’sallthesamelogic,justdifferingsyntaxandAPIs.

RProjectforStatisticalComputing:

Thisisanenvironmentandalisp-likescriptinglanguage.AllthestatsstuffyoucouldeverwanttodowillbeprovidedintoR,includingamazingplotting.TheMachineLearningcategoryonCRAN（think:

third-partyMachineLearningpackages）hascodewrittenbyleadersinthefieldwithstateoftheartmethods,aswellasanythingelseyoucanthinkof.LearningRisamustifyouwanttoprototypeandexplorequickly.Itjustmightnotbethefirstplaceyoustart.

WEKA:

ThisisaDataMiningworkbenchprovidingAPI,andanumberofcommandlineandgraphicaluserinterfacesforthewholedatamininglifecycle.Youcanpreparedata,visualizeexplore,buildclassification,regressionandclusteringmodelsandmanyalgorithmsareprovidedbuiltinaswellasprovidedinthirdpartyplugins.NotrelatedtoWEKA,MahoutisagoodJavaframeworkforMachineLearningonHadoopinfrastructureifthatismoreyourthing.Ifyou’renewtobigdataandmachinelearning,stickwithWEKAandlearnonethingatatime.

ScikitLearn:

MachineLearninginPythonbuiltontopofNumPyandSciPy.IfyouareaPythonoraRubyprogrammer,thisisthelibraryforyou.It’sfriendly,powerfulandcomeswithexcellentdocumentation.Orangewouldbeagoodalternativeifyou’dliketotrysomethingelse.

Octave:

IfyouarefamiliarwithMatLaboryou’reaNumPyprogrammerlookingforsomethingdifferent,considerOctave.ItisanenvironmentfornumericalcomputingjustlikeMatlabandmakesiteasytowriteprogramstosolvelinearandnon-linearproblems,suchasthosethatunderliemostmachinelearningalgorithms.Ifyouhaveanengineeringbackground,thismightbeagoodplaceforyoutostart.

BigML:

Maybeyoudon’twanttodoanyprogramming.YoucandrivetoolslikeWEKAcompletelywithoutprogramming.YoucangoonestepfurtheranduseserviceslikeBigMLthatoffermachinelearninginterfacesonthewebwhereyoucanexplorebuildingmodelsallinthebrowser.

Pickaplatformanduseittodoyourpracticalmachinelearningeducation.Don’tjustread,do.

VideoCourses

Videoisaverypopularwaytogetstartedinmachinelearning.IwatchalotofmachinelearningvideosonYouTubeandVideoLectures.Net.Theriskisthatallyouwilldoisconsumeandfailtotakeaction.Irecommendyoushouldalwaystakenoteswhenwatchingavideo,evenifyoudiscardthenoteslater.Ialsorecommendtryingoutwhateveritisyou’relearninginthelecture.

Frankly,noneofthevideocoursesIhaveseenarereallysuitableforabeginner,foratruebeginner.Theyallpresupposeaworkingknowledgeofatleastlinearalgebraandprobabilitytheory,andmore.AndrewNg’sStanfordlecturesareprobablythebestplacetostartforacourse,otherwisethereareone-offvideosIrecommend.

StanfordMachineLearning:

AvailableviaCourseraandtaughtbyAndrewNg.Inadditiontoenrolling,youcanwatchallthelecturesanytimeandgetthehandoutsandlecturenotesfromtheactualStanfordCS229course.ThecourseincludeshomeworkandquizzesandfocusesonlinearalgebraandusingOctave.

CaltechLearningfromData:

AvailableviaedXandtaughtbyYaserAbu-Mostafa.AllthelecturesandmaterialsareavailableontheCalTechsite.Again,liketheStanfordclass,youcantakeitatyourownpaceandcompletethehomeworkandassignments.Itcoverssimilarsubjectsandgoesintoalittlebitmoredetailsandismoremathematical.Thehomeworkisprobablytoochallengingforabeginner.

MachineLearningCategoryonVideoLectures.Net:

Thisisaneasyplacetodrownintheoverloadofcontent.Lookforvideosthatseeminterestingandtrythemout.Bailifit’satthewronglevelortakenotesifyou’reenjoyingit.IfindIkeepcomingbacktorefreshmyselfontopicsandtopickupentirelynewtopics.Also,it’sgreattoseewhatthemastersofthefieldactuallylooklike.

“GettingInShapeForTheSportOfDataScience”–TalkbyJeremyHoward:

AtalktoalocalRusersgrouponthepracticalprocessfordoingwellincompetitivemachinelearning.Thisisveryvaluablebecausesofewpeopletalkaboutwhatit’sactuallyliketoworkonaproblemandhowtodoit.Inot-so-secretlyfantasiseaboutfundingawebrealityTVshowthatfollowsparticipantsinmachineleaningcompetitions.That’showintoitIam!

OverviewPapers

Ifyouarenotusedtoreadingresearchpapers,youwillfindthelanguageverystiff.Apaperislikeasnippetofatextbook,butdescribesanexperimentorsomeotherfrontierofthefield.Nevertheless,therearesomepapersthatyoumightfindinterestingifyouarelookingtogetstartedinmachinelearning.

TheDisciplineofMachineLearning:

AwhitepaperdefiningthedisciplineofMachineLearningbyTomMitchell.ThiswasapieceoftheargumentMitchellusedtoconvincethePresidentofCMUtocreateastandaloneMachineLearningdepartmentforasubjectthatwillstillbearoundin100years（alsoseethisshortinterviewwithTomMitchell）.

AFewUsefulThingstoKnowaboutMachineLearning:

Thisisagreatpaperbecauseitpullsbackfromspecificalgorithmsandmotivatesanumberofimportantissuessuchasfeatureselectiongeneralizabilityandmodelsimplicity.Thisisallgoodstufftogetrightandthinkclearlyaboutfromthebeginning.

I’veonlylistedtwoimportantpapers,becausereadingpaperscanreallybogyoudown.

BeginnerMachineLearningBooks

Therearealotofmachinelearningbooksandveryfewarewrittenforbeginners.Whatisabeginnerreally?

Mostlikelyyou’recomingtomachinelearningfromanotherfield,mostlikelycomputerscience,programmingorstatistics.Eventhen,mostbooksexpectyoutohaveagroundinginatleastlinearalgebraandprobabilitytheory.

Nevertheless,thereareafewbooksouttherethatencourageeagerprogrammerstogetstartedbyteachingtheminimumintuitionforanalgorithmandpointtotoolsandlibrariessothatyoucanrunofftoandtrythingsout.MostnotablyProgrammingCollectiveIntelligence,MachineLearningforHackersandDataMining:

PracticalMachineLearningToolsandTechniquesforPython,R,andJavarespectively.Ifindoubt,graboneofthesethreebooks!

BooksforMachineLearningBeginners

ProgrammingCollectiveIntelligence:

BuildingSmartWeb2.0Applications（AffiliateLink）:

Thisbookwaswrittenforyoudearprogrammer.It’sliteontheory,heavyoncodeexamplesandpracticalwebproblemsandsolutions.Buyit,readit,dotheexercises.

MachineLearningforHackers（AffiliateLink）:

I’drecommendthisbookafterreadingProgrammingCollectiveIntelligence（above）.Itagainprovidesworkedexamplesthatarepractical,butithasamoreofadataanalysisflavorandusesR.Ireallylikethisbook!

MachineLearning:

AnAlgorithmicPerspective（AffiliateLink）.ThisbookislikeamoreadvancedversionofProgrammingCollectiveIntelligence（above）.Ithassimilaraims（getprogrammersstartedinMachineLearning）,butitincludesmathsandreferencesaswellasexamplesandsnippetsinpython.I’drecommendreadingthisafterreadingProgrammingCollectiveIntelligenceifyou’restillinterested.

DataMining:

PracticalMachineLearningToolsandTechniques,ThirdEdition（AffiliateLink）:

Iactuallystartedwiththisbook,actuallyitwasthefirsteditionanditwasabouttheyear2000.IwasaJavaprogrammerandthisbookandthecompanionlibraryWEKAprovidedaperfectenvironmentformetotrythingsout,implementmyownalgorithmsasplug-insandgenerallypracticeMachineLearningandthebroaderprocessofDataMining.Ihighlyrecommendthisbookandthispath.

MachineLearning（AffiliateLink）:

Thisisanoldbookanddoesincludeformulasandlotsofreferences.It’satextbookbutisalsoveryaccessiblewithgroundedmotivationsforeach