资源描述：
《will the bad data please stand up》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。
1、CHAPTER7WilltheBadDataPleaseStandUp?PhilippK.JanertAmonghikersandclimbers,theysaythat“thereisnosuchthingasbadweather—onlyinappropriateclothing.”Andasanybodywhohasspentsometimeoutdoorscanattest,itisoftenpreciselytripsundertakenundermorechallengingcircumstancesthatleadtothemostnoteworthym
2、emories.Butonehastobewillingtoputoneselfoutthere.Inasimilarspirit,Idon’tthinkthereisreallysuchathingas“baddata”—onlyinap‐propriateapproaches.Tobesure,therearedatasetsthatrequiremorework(becauseofmissingdata,backgroundnoise,poorencoding,inconvenientfileformats,andsoon),buttheydon’tposefu
3、ndamentalchallenges.Givensufficienteffort,theseproblemscanbeovercome,andthereareusefultechniquesforhandlingsuchsituations(liketricksforstayingwarmduringalate-Novemberhike).Butbasically,that’sremainingwithinfamiliarterritory.Todiscovernewvistas,onehastobewillingtofollowanunmarkedtrailand
4、seewhereitleads.Orequivalently,whenworkingwithdata,onehastodaretohaveanopinionaboutwherethedataisleadingandthencheckwhetheronewasrightaboutit.Notethatthistakescourage:itisfarsafertomerelydescribewhatonesees,butdoingsoismissingawholelotofaction.Let’sevaluatesometrailreports.Later,we’llre
5、groupandseewhatlessonswehavelearned.Example1:DefectReductioninManufacturingAmanufacturingcompanyhaddevelopedarathercleverschemetoreducethenumberofdefectiveitemsshippedtotheircustomers—atnoadditionalcost.Thebasicideawastouseaquantitythatwasalreadybeingmeasuredforeachnewlymanufactureditem
6、(let’scallitthe“size”—itwasn’t,butitwon’tmatter)asindicatoroftheitemquality.If95thesizewas“off,”thentheitemprobablywasnotgoingtoworkright.Tothemanufac‐turer,thekeybenefitofthisindirectapproachwasthelowcost.Thesizewasalreadybeingmeasuredaspartofthemanufacturingprocess,soitdidnotimposeana
7、dditionaloverhead.(Themainproblemwithqualityassurancealwaysisthatithastobecheap.)Theyputasysteminplacethatflaggeditemsthatseemed“off”ascandidatesformanualinspection.Thequestionwas:howwellwasthissystemworking?Howgoodwasitatactuallydetectingdefectiveitems?Thiswasnotsoeasytotell,b
