会】TalkingData全球算法大赛盘点.pdf

《会】TalkingData全球算法大赛盘点.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、TalkingData全球算法大赛盘点路瑶数据科学部Whatisthe…•PowerofDataScience•MysteryofDataScientist•MagicofDataScientist•TheArenaofDataScientistTheArenaofDataScientistKaggle:leadingplatformforcrowdsourcingdatachallenges.•Acommunitytobuildthebestsolutiononproblemsposedbyindustry,go

2、vernmentandacademia.•Over1,200datasciencechallenges.•Morethan600,000registeredusersover194countriesfromaroundtheworld,fromawidevarietyofeducationalbackgroundsandareoftenexpertsintheirfields.•PlatformofKDDCUPFamouscompetitionandscientistofKaggleKaggleDrugDiscov

3、eryCompetition,2012GeoffreyHintonTalkingDatahostthecompetitioninKagglefor•Provideopendataandopenplatformfor•Smartestscientistandsmartestmethodology•TosolvemostchallengingtopicsTalkingDataCompetitionwasannouncedinDataScienceSummitinSanFranciscoonJuly13thAcknowl

4、edgeTuri•AfamousMachineLearningcompanywhocreatedGraphLab.•AcquiredbyAppleInc.inAugustTalkingDataMobileUserDemographicsGoalofthecompetition:•Toknowyourusers’profileby•Learningfromtheirbehavior.Given•Applicationusageandtracewithtimestamps•Mobilebrandanddevicemod

5、eTooptimize•TheestimationoftheirageandgroupEvaluatedbyParticipationofthecompetition1689teams,1961playerssubmitted,24,629entries,2729kernels•LargestcompetitionhostedbyaChinesecompany.•AmongthehighestnumberofKernelsofall.•Greatproportionoftopkagglersparticipated

6、.•70+countriesandregionsrepresentedbytheparticipantpool.Whatisthe…•PowerofDataScience•MysteryofDataScientist•MagicofDataScientist•TheArenaofDataScientistMagicofDataScientist•Interprettherawdata•Executefeatureengineering•Finetuneindividualandensemblemodels•Avoi

7、doverfitting•FetchthebesttoolsInterpretyourdataFeatureengineeringActivetimeslotsAge/GenderasfeatureActiveareaAppLabelPatternsDistrictspecificfeaturesBinary/WeightedFeatureInstallbutnotactiveFinetuneindividualandensemblemodelsRandomForestGBDTSVMLogisticNeuralNe

8、tworkRegressionAdaboostStackAvoidOverfittingTrainSetTestSetAlwaysdoPublicPrivateBoardCrossValidationBoardFetchthebesttoolsKeras•highlymodularneuralnetworkslibrary,writteninPython•c