深度学习—训练集、验证集和测试集概念.doc

深度学习—训练集、验证集和测试集概念.doc

ID:55265098

大小:91.50 KB

页数:3页

时间:2020-05-08

深度学习—训练集、验证集和测试集概念.doc_第1页
深度学习—训练集、验证集和测试集概念.doc_第2页
深度学习—训练集、验证集和测试集概念.doc_第3页
资源描述:

《深度学习—训练集、验证集和测试集概念.doc》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、ØTraining,ValidationandTestDataExample:(A)Wehavedataon16dataitems,theirattributesandclasslabels.RANDOMLYdividetheminto8fortraining,4forvalidationand4fortesting.TrainingItemNo.d–AttributesClass1.02.03.KNOWNFORALL14.15.DATAITEMS16.17.08.0Validation9.010.011.112.0Test13

2、.014.015.116.1(B).Next,supposewedevelop,threeclassificationmodelsA,B,Cfromthetrainingdata.Letthetrainingerrorsonthesemodelsbeasshownbelow(recallthatthemodelsdonotnecessarilyprovideperfectresultsontrainingdata—neithertheyarerequiredto).ClassificationresultsfromItemNo.

3、d-AttributesTrueClassModelAModelBModelC1.00112.ALLKNOWN00003.10104.11015.10006.11117.00008.0000ClassificationError2/83/83/8(C).Next,usethethreemodelsA,B,Ctoclassifyeachiteminthevalidationsetbasedonitsattributevales.Recallthatwedoknowtheirtruelabelsaswell.Supposeweget

4、thefollowingresults:ClassificationresultsfromItemNo.d-AttributesTrueClassModelAModelBModelC9.010010.001011.101012.0010ClassificationError2/42/41/4Ifweuseminimumvalidationerrorasmodelselectioncriterion,wewouldselectmodelC.(D).NowusemodelCtodetermineclassvaluesforeachd

5、atapointinthetestset.Wedosobysubstitutingthe(known)attributevalueintotheclassificationmodelC.Again,recallthatweknowthetruelabelofeachofthesedataitemssothatwecancomparethevaluesobtainedfromtheclassificationmodelwiththetruelabelstodetermineclassificationerroronthetests

6、et.Supposewegetthefollowingresults.ClassificationresultsfromItemNo.d-AttributesTrueClassModelC13.0014.ALLKNOWN0015.1016.11ClassificationError1/4(E).Basedontheabove,anestimateofgeneralizationerroris25%.WhatthismeansisthatifweuseModelCtoclassifyfutureitemsforwhichonlyt

7、heattributeswillbeknown,nottheclasslabels,wearelikelytomakeincorrectclassificationsabout25%ofthetime.(F).Asummaryoftheaboveisasfollows:ModelTrainingValidationTestA2550----B37.550----C37.52525ØCrossValidationIfavailabledataarelimited,weemployCrossValidation(CV).Inthis

8、approach,dataarerandomlydividedintoalmostkequalsets.Trainingisdonebasedon(k-1)setsandthek-thsetisusedfortest.Thisprocessisrepeatedk

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。