8)20161007 Classification Probabilistic Generative Model

8)20161007 Classification Probabilistic Generative Model

ID:40254284

大小:1.37 MB

页数:35页

时间:2019-07-29

8)20161007 Classification Probabilistic Generative Model_第1页
8)20161007 Classification Probabilistic Generative Model_第2页
8)20161007 Classification Probabilistic Generative Model_第3页
8)20161007 Classification Probabilistic Generative Model_第4页
8)20161007 Classification Probabilistic Generative Model_第5页
资源描述:

《8)20161007 Classification Probabilistic Generative Model》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、Classification:ProbabilisticGenerativeModelClassification?FunctionClassn•CreditScoring•Input:income,savings,profession,age,pastfinancialhistory……•Output:acceptorrefuse•MedicalDiagnosis•Input:currentsymptoms,age,gender,pastmedicalhistory……•Output:whichkindofdiseases•Handwritten

2、characterrecognitionoutput:Input:金•Facerecognition•Input:imageofaface,output:personExampleApplication?=?=?=pokemongames(NOTpokemoncardsorPokemonGo)ExampleApplication•Total:sumofallstatsthatcomeafterthis,ageneralguidetohowstrongapokemonis320•HP:hitpoints,orhealth,defineshowmuch

3、damageapokemoncanwithstandbeforefainting35•Attack:thebasemodifierfornormalattacks(eg.Scratch,Punch)55•Defense:thebasedamageresistanceagainstnormalattacks•SPAtk:specialattack,thebasemodifierforspecialattacks(e.g.fire40blast,bubblebeam)•SPDef:thebasedamageresistanceagainstspecia

4、lattacks5050•Speed:determineswhichpokemonattacksfirsteachround90Canwepredictthe“type”ofpokemonbasedontheinformation?ExampleApplicationHowtodoClassification•TrainingdataforClassification?1,?ො1?2,?ො2……??,?ො?ClassificationasRegression?BinaryclassificationasexampleTraining:Class1m

5、eansthetargetis1;Class2meansthetargetis-1Testing:closerto1→class1;closerto-1→class2b+wx+wx=0todecreaseerror1111Class2Class2-1-111x2x2Class1Class1>>1y=b+wx+wxerror1122xx11Penalizetotheexamplesthatare“toocorrect”…(Bishop,P186)•Multipleclass:Class1meansthetargetis1;Class2meansthe

6、targetis2;Class3meansthetargetis3……problematicIdealAlternatives??•Function(Model):??>0Output=class1?????Output=class2•Lossfunction:Thenumberoftimesf??=෍????≠?ො?getincorrectresultson?trainingdata.•Findthebestfunction:•Example:Perceptron,SVMNotTodayTwoBoxesBox1Box2P(B1)=2/3P(B2)

7、=1/3P(Blue

8、B1)=4/5P(Blue

9、B1)=2/5P(Green

10、B1)=1/5P(Green

11、B1)=3/5fromoneoftheboxesWheredoesitcomefrom??Blue

12、?1??1P(B

13、Blue)=1?Blue

14、?1??1+?Blue

15、?2??2EstimatingtheProbabilitiesTwoClassesFromtrainingdataClass1Class2P(C1)P(C2)P(x

16、C1)P(x

17、C2)Givenanx,whichclassdoesitbelongto??

18、?1??1??1

19、

20、?=??

21、?1??1+??

22、?2??2GenerativeModel??=??

23、?1??1+??

24、?2??2PriorCl

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。