classification,机器学习中分类问题的探讨

《classification,机器学习中分类问题的探讨》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、ClassificationMichaelI.JordanUniversityofCalifornia,BerkeleyClassificationInclassificationproblems,eachentityinsomedomaincanbeplacedinoneofadiscretesetofcategories:yes/no,friend/foe,good/bad/indifferent,blue/red/green,etc.Givenatrainingsetoflabeledentities,developaruleforassigninglabelsto

2、entitiesinatestsetManyvariationsonthistheme:binaryclassificationmulti-categoryclassificationnon-exclusivecategoriesrankingManycriteriatoassessrulesandtheirpredictionsoverallerrorscostsassociatedwithdifferentkindsoferrorsoperatingpointsRepresentationofObjectsEachobjecttobeclassifiedisrepre

3、sentedasapair(x,y):wherexisadescriptionoftheobject(seeexamplesofdatatypesinthefollowingslides)whereyisalabel(assumedbinaryfornow)Successorfailureofamachinelearningclassifieroftendependsonchoosinggooddescriptionsofobjectsthechoiceofdescriptioncanalsobeviewedasalearningproblem,andindeedwe’l

4、ldiscussautomatedproceduresforchoosingdescriptionsinalaterlecturebutgoodhumanintuitionsareoftenneededhereDataTypesVectorialdata:physicalattributesbehavioralattributescontexthistoryetcWe’llassumefornowthatsuchvectorsareexplicitlyrepresentedinatable,butlater(cf.kernelmethods)we’llrelaxthata

5、sumptionDataTypestextandhypertext.stdtext{font-family:Verdana,Arial,Helvetica,sans-serif;font

6、-size:11px;color:#1F3D4E;}.stdtext_wh{font-family:Verdana,Arial,Helvetica,sans-serif;font-size:11px;color:WHITE;}

7、FN/en/CDA/Images/common/labels/decorative_2px_blk.gif">