1. 《Churn prediction in telecommunication using ML》 Abstract Setbacks (difficulties): enormous database; large feature space; ...
客户流失预测——相关论文学习笔记
RaymondLove~
于 2020-07-0309:28:32 发布
736
收藏
5
分类专栏:
机器学习
文章标签:
客户流失预测
客户流失预测相关论文
客户流失预测论文学习笔记
版权声明:本文为博主原创文章,遵循CC4.0BY-SA版权协议,转载请附上原文出处链接和本声明。
本文链接:https://blog.csdn.net/Emma_Love/article/details/107093390
版权
机器学习
专栏收录该内容
6篇文章
0订阅
订阅专栏
TableofContents
1.《ChurnpredictionintelecommunicationusingML》
2.《HandlingimbalanceddatainchurnpredictionusingADASYNandBack-propagationalgorithm》
3.《Customerchurnpredictionforretailbusiness》=>Useless
4.《Acomparativestudyofcustomerchurnpredictionintelecomindustryusingensemblebasedclassfiers》-Useless
5.《Customerchurnpredictioninaninternetserviceprovider》
6.《Areviewandanalysisofchurnpredictionmethodsforcustomerretentionintelecomindustries》
7.《Usingdeeplearningtopredictcustomerchurninamobiletelecommunicationnetwork》
8.《Churnanalysisandplanrecommendationfortelecomoperators》
9.《Adataminingprocessframeworkforchurnmanagementinmobiletelecommunicationindustry》
10.《Usingdeeplearningtopredictcustomerchurninamobiletelecommunicationnetwork》
1.《ChurnpredictionintelecommunicationusingML》Abstract
Setbacks(difficulties):
enormousdatabase;largefeaturespace;imbalancedclassdistribution:numberofchurner<pronetooverfittingRandomundersampling(RUS):randominstancesfromthemajorityclassarediscarded-->maydiscardsomeusefulinstancesModels:
Treeandrules-based: CART,PART,Ensembleoftrees:C5.0,baggedCART,RF,XGBoostLinear:LR,lineardiscriminantanalysisNon-linear:Neuralnetwork,SVM,KNN,NaïveBayesRelatedtechs
Models:KNN,RF,RotationForest,Adaboost,Modelfusion:orderedweightedaverage,voteFeatureextraction:PCA,F-score,Fisher'sratio,MinimumRedundancyMaximumRelevanceMethodsandmaterials
Dataset: 1:3,imbalancedDatapreprocessing:removinguselessfeatures+samplingusingSMOTE+featureselectionusingCo-relation,GainRatio,informationgainandoneRIntroSMOTE: https://www.cnblogs.com/Determined22/p/5772538.html Createnewsimilarinstancesinsteadofsameinstancesforminorityclassdata,sothatitcansoftendecisionboundary,andfurthertheclassificationcanbemoregeneralanddoesnotover-fit.Co-relationFeatureSelection:Pearson'sco-relationcoefficientandspearman'sco-relationcoefficient
Informationgainattributeselection: entropyGainRatioattributeselection: overcomethelimitationofIG(isusedtoselectattributesfortheterminalnodesofthedecisiontree)OneRAttributeSelection:shortforOneRule. Generatingoneruleforeachpredictorinthedata,thenselectstherulewiththesmallesttotalerrorasits"onerule"DecisiontreebasedclassificationPartialtreebasedclassificationPART: decisiontreesthatareprunethedecisiontreeontheirownBaggedtreeclassification: bootstrapaggregationorbaggingBoostedclassificationtrees:Conclusions
Adequatepreprocessinganddatabalancingincaseofimbalanceddatasetsareboundtoimprovetheclassificationperformancesoftheusedclassifiers.SMOTEbasedclassifier--->improveclassificationperformanceEnsembleapproachcanachieveperformanceCo-relationbasedfeatureextractedbetterthanotherselectionmethodsinthiscase2.《HandlingimbalanceddatainchurnpredictionusingADASYNandBack-propagationalgorithm》Abstract
Churnpredictiondifficulty:imbalanceddataMethodsinthepaper:Oversamplingalgorithm: ADASYN(Adaptivesynthetic sampling),anoversamplingmethods==>solveimbalancedproblem.ADASYNisanimprovement algorithmfromSMOTE(Syntheticminorityover-sampling)Classificationmethod:backpropagationalgorithmIntroduction
DATASIZE:tensofcolumns(attributes)&thousandsofrowsofdataModel:boosting,randomforestanditsmodificationMethodology
Inputdata:1yearTeldata,55features&200.387rows;in-balanceddata: churn:un-churned=0.04,0.96Featureselection:PearsoncorrelationequationResamplingusingADASYN:differencebetweenADASYN-SMOTE:ADASYNusesdensitybetadistributionasareferencefordeterminingthenumberofsyntheticdataConstructingchurnpredictionwithbackpropagationmethod
Forwardpropagationofoperatingsignal:Backpropagationoferrorsignal:Performancemeasurement
F1-Score&accuracy
Precision=TP/(TP+FP):体现了模型对负样本的识别能力,precision越高,说明模型对负样本的区分能力越强Recall(sensitivity)=TP/(TP+FN):体现了分类模型对正样本的识别能力,recall越高,说明模型识别正样本的能力越强F1-score=2TP/(2TP+FN+FP)=2*precision*recall/(precision+recall):是两者的结合,F1-score越高,说明分类模型越稳健ConfusionMatrixResults&analysis
Datasource:PTTelkomIndonesiarecordedfrom2014.10-2015.9; 200387rows,with55features,only %oftotaldataarechurned.Afterfeatureextraction,38featuresareused3.《Customerchurnpredictionforretailbusiness》=>UselessAbstract
Dataset:UCImachinelearningrepository;2010.1-2011.1transactionsrecordsMethod:preprocessingtoremoveNAs,validatingnumericalvalues,removingerroneousdatapoints;performaggregationsonthedatatogenerateinvoice+customerdatasets; MLalgorithms:SVM,RF,ExtremegradientboostingIntroduction
Customerchurn=customerattrition=customerdefectionObjectiveofthisproject:
PredictchurnvalueforallthecustomersofthecompanyforagivenperiodoftimeComputetheoverallchurnrateforthegiventimeProvidedeeperinsightintothesalesbyanalyzingcustomers'buyingpatternDetectcustomerswhoareabouttodropoutfromthebusinessinordertotakenecessarystepsProvideclearvisualizationsofthechurnpredictionstohelpbusinesscomeupwithbetterstrategiesHelpbusinessknowtherealvalueofapotentialchurncustomerandretainhim/herasaloyalcustomerbyestablishingpriorities,optimizingresources,puttingefficientbusinesseffortsandmaximizingthevalueoftheportfolioofthecustomerHelpbusinesscomeupwithpersonalizedcustomerretentionplanstoreducethechurnrateDeliverables
AsystemthatcanpredictifacustomerisachurnornotforaretailbusinessAsystemwhichcancomputechurnrateoftheretailbusinessAsystemwhichcanrunmultiplealgorithmsandcompareperformance amongthem.AlgorithmsincludeRF,SVM,GradientboostingtopredictthecustomerchurnforagivenperiodoftimeLiteraturesurvey
Sequentialpatterns
DEML:Geneticmodelling:NeuralNetworks:LogisticregressionandrandomforestsGametheory:博弈论Design
Architecturedesign:SequencediagramDataflowdiagram Implementation
Datasize:541909rowsPreprocess:cleaningdata+aggregationTraining:MLmodel:RF;SVM;Gradientboosting4.《Acomparativestudyofcustomerchurnpredictionintelecomindustryusingensemblebasedclassfiers》-UselessAbstract
Comparingensemblebasedclassifierswerecomparedwithwell-knownclassifiersnamely decisiontree,naïveBayesclassifier,andSVMIntroductionLiteraturesurveyWorkingmethodology
Decisiontree:C4.5---->accuracyishigh,butfailstorespondtonoiseNaïvebayes--->SVM --->notsuitedfordatawithnoiseBagging(bootstrapaggregation):
DividedatasetintoksubsetwithreplacementTrainthemodelbyusing(k-1)subsetandtestthemodelbyusingtherest1subsetBoosting
MaintainaweightforeachtrainingtupleRandomforest:
Disadvantage:cannothandleunbalanceddatasetbyusingrandomforest5.《Customerchurnpredictioninaninternetserviceprovider》Abstract
Methodology:Featureengineering:
SMOTEoversamplingmethod--reducetheimbalancebetweenthenumberofchurnersandnon-churneresMachinelearningmodel:adaboost,extratrees,knn,neuralnetwork,xgboostExperimentalResults
Xgboostisthebest,precision=45.71%; recall=42.06%Datasetcharacters:
Churn:un-churn=2:98Introduction
Problemstatement:predictwhethercustomerwillrenewtheirservice(monthlysubscriptions).Thatmeans,servicewillbeexpiredattheendofeachmonth,thevaliddurationtorenewtheservicesisfromtheexpiredtimetonext16days.
Churncustomers:don'trenewtheirsubscriptioninnext16days,beforetheendofcurrentservice & terminatecurrentserviceNoticed:thestatusofacustomer,churnornon-churn,isdeterminedattheendofeachmonth,regardlessofthepreviousstatusAfterthisperiod,userswhodonotrenewtheservicesareidentifiedaschurncustomers ----->we candefine,predictwhetherourcustomerwillrenewtheirsubscriptioninnexthalfofrenewterm.Relatedwork
Noticed: Metricsinchurnprediction:
Forfindingthemostpossiblechurningcustomers === precisionmeasureswouldbemoreeffectiveForpurposeofretainingmostcustomers===== recallofthemodelneedstobeimprovedReducingimbalanceddata:
SMOTEModel:
KNNAdaboost:sensitivetonosiedataandoutliersExtra-trees:NeuralNetwork:Dataandfeatureengineering
Featureswereseperated3maingroups ---> .1customerinformation;2.theirusagedata;3.servicedata
Customerinfo:registrationdate,terminationdate,location,servicetype,cabletype,bandwidth,paymenthistory,promotion,andsoonCustomerusagedata:theinitialdatetimeofconnection,disconnectiondate time,reasonforrejection,typeofmodem,user'sdailyusagesuchasamountofdatadownloaded&uploadedCustomerservicedata:customer'sinboundandoutboundcallphonehistory;customersatisfactionsurveys6.《Areviewandanalysisofchurnpredictionmethodsforcustomerretentionintelecomindustries》Abstract
Focusingonanalyzingthechurnpredictiontechniquestoidentifythechurnbehaviorandvalidatethereasonsforcustomerchurn
Summarizethechurnpredictiontechniques-->deeperunderstandofthecustomerchurnShowsthemostaccuratechurnprediction -->hybridmodelsratherthansinglealgorithmsAnalysisofcustomerchurnpredictionmethodologies
Preprocessing-imbalancedproblemandsamplingbaseonchurnpredictionEnsemblemethods:
Reference:http://scikit-learn.org/stable/modules/ensemble.htmlGoal:combinethepredictionsofseveralbaseestimatorsbutwithagivenlearningalgorithminordertoimprovegeneralizability/robustnessoverasingleestimator.Twofamiliesofensemblemethods:
Averagingmethods:thedrivingprincipleistobuildseveralestimatorsindependentlyandthentoaveragetheirpredictions.Onaverage,thecombinedestimatorisusuallybetterthananyofthesinglebaseestimatorbecauseitsvarianceisreduce
Examples:baggingmethods,forecastofrandomizedtrees
Baggingmeta-estimator
Baggingmethodformaclassofalgorithmswhichbuildseveralinstancesofablack-boxestimatoronrandomsubsetsoftheoriginaltrainingsetandthenaggregatetheirindividualpredictionstoformafinalprediction.
Forecastofrandomizedtrees
RF:eachtreeintheensembleExtra-tressBoostingmethods:baseestimatorsarebuiltsequentiallyandonetriestoreducethebiasofthecombinedestimator,.Themotivationistocombineseveralweakmodelstoproduceapowerfulensemble
Examples:Adaboost,gradienttreeboostingChurnpredictionfrombigtree7.《Usingdeeplearningtopredictcustomerchurninamobiletelecommunicationnetwork》Abstract
Auto-encoders -----> deepbeliefnetworks---->multi-layerfeedforwardnetworksFramework: four-layerfeedforwardarchitectureIntroduction
Motivation: usedeeplearningtoavoidtime-consumingfeatureengineeringeffortandideallytoincreasethepredictiveperformanceofpreviousmodels.Datasetintro:
Historicaldatafromatelecommunicationcompanywithnearly1.2millioncustomersandspanoversixteenmonthsChallengingCharacteristics:
ChurnrateisveryhighandallcustomersareprepaidusersChurnpredictioninprepaidmobiletelecommunicationnetwork
Goal:toinferwhenthislackofactivitymayhappeninthefutureforeachactivecustomerStatedefinition: Deeplearningmodelsforchurnprediction
Inputdata&preparation:8.《Churnanalysisandplanrecommendationfortelecomoperators》Abstract
Inthispaper,wedesignahybridMLclassifiertopredictifacustomerwillchurnbasedontheCDRparametersan wealsoproposearuleenginetosuggestbestplans9.《Adataminingprocessframeworkforchurnmanagementinmobiletelecommunicationindustry》Introduction
Aims:Byusingacombinationofexpertsystemsandmachinelearningtechniques,theprocessframeworkhandleschurnpredictionfrom3perspectives:
PredictionofwhichsubscribermaychurnDeterminationofreasonswhysubscribermaychurnRecommendationsofappropriatestrategyforcustomerretentionData
Arichchunkoftelecomsubscribers'demographicdataSubscribers'transactionsinformationSubscribers'complaintsinformationProcessframeworkExperiment
1.CollectRawDataset
SubscriberData:callernumber,callednumber,incomingroute,outgoingroute,amountb4call,amountaftercall,internationalmobilesubscriberidentity,exchangeid,recordtype,eventtype,dateofsubscription,typeofservicesubscribedandsubscribenumberComplaintData: request_complained_id,dateofcomplaint,timeofcomplaint,typefcomplaint,status(openandclose),imputer(internalstaffinitiator),handle_by_person2.Datapredictionmodel
RawData--->features
20featurers
Churn predictionwithartificialneuralnetworkGeneratingchurnreasonsandinterventionstrategy
ResultsobtainedfromchurnpredictionusingANN--->decisionsupportexpertsystem(DSES)togenerateprobablereasonsforchurn&recommendationsforcustomerretention.
DSES:-asetofif-thenrulesthatenabledthegenerationofrecommendationsofappropriateincentivesbasedonthecreditratingofasubscriber.
Classifysubscribers--->high-valued,medium-valuedandlow-valued,sowecanignorelowvaluedsubscribers,andputmoreeffortsonhigh-valued&mediumvalued.Generatechurnreasons
Basedontherulestodeterminethechurnreasons.
ThefollowingisasampleofJessRulesintheDSEM
10.《Usingdeeplearningtopredictcustomerchurninamobiletelecommunicationnetwork》
Understandingandcalculatingchurn
Highlevel
MeasureofhowmanycustomersleaveoverasettimeperiodMeasurehowmuchrevenueyouloosethroughcustomercancellationsHowchurncanimpactthebottomline
CalculateLifetimevalue(LTV)-tounderstandthevalueaCSMhas
BasicLTVCostofcustomeracquisition(COCA)CostofGoodssold(COGS)Calculatechurn
CustomerchurnRevenuechurnAnalyzingchurn
Reasons-
FindchurnreasonstofocusandprioritizeKnowwhetheractionstoretaincustomerisworkingMethods
Cohortreports列式报表
Type1
Churnbycustomerage-groupingyour customerbyageChurnbycustomerbehavior
Needtolookatcustomerswhouseacertainfeatureorcompleteacertainactionanddetermineit'simpactonchurn
RaymondLove~
关注
关注
1
点赞
踩
5
收藏
打赏
0
评论
客户流失预测——相关论文学习笔记
1.《ChurnpredictionintelecommunicationusingML》 Abstract Setbacks(difficulties): enormousdatabase; largefeaturespace; imbalancedclassdistribution:numberofchurner<<numberofnon-churners ..
复制链接
扫一扫
专栏目录
论文研究-客户流失预测的现状与发展研究.pdf
07-22
根据客户流失预测研究的发展历程和智能化程度的高低,将客户流失预测研究划分为三个阶段,包括基于传统统计学的预测方法、基于人工智能的预测方法和基于统计学习理论的预测方法,并通过分析每个阶段存在的问题提出了未来可研究的方向。
参与评论
您还未登录,请先
登录
后发表或查看评论
深度分析|《电信用户流失预测模型Telcocustomerchurn》(所有分类模型精度平均得分在0.8以上)
fulk6667g78o8的专栏
03-05
1308
1、研究背景
1、做好“用户流失预测”可以降低营销成本。
老生常谈,“新客户开发成本”是“老客户维护成本”的5倍。
2、获得更好的用户体验。
并不是所有的增值服务都可以有效留住客户。
3、获得更高的销售回报。
价格敏感型客户和非价格敏感性客户
2、提出问题
1、流失客户有哪些显著性特征?2、当客户在哪些特征下什么条件下比较容易发生流失?
3、数据集描述
该数据是datafountain上的《电信客户流失数据》点...
telco-customer-churn-prediction
WenYu的博客
03-12
1085
—##转载自:https://www.kaggle.com/liyingiris90/telco-customer-churn-prediction
title:"ChurnPrediction-
LogisticRegression,DecisionTreeandRandomForest"
output:
html_document:default
pdf_document...
卡狗项目学习记录-ChurningCustomersPrediction
weixin_43935883的博客
01-16
303
卡狗学习记录-ChurningCustomersPrediction1.数据预处理1.1导入需要的包1.2读取数据集1.3数据分析1.3.1Customer_Age1)matplotlib.pyplot柱状图2)plotly.graph_objs柱状图3)分析结果1.3.2Gender1)matplotlib.pyplot饼图2)plotly.express饼图3)分析结果1.3.2Dependent_count
原链接及数据集:BankChurnDataExplo
sklearn复合评估器的构建(电信客户流失模型)
最新发布
tangyi2008的专栏
05-04
321
Sklearn(全称Scikit-Learn)是基于Python语言的机器学习工具。
它建立在NumPy,SciPy,Pandas和Matplotlib之上,API的设计非常好,所有对象的接口简单,很适合新手上路。
CAT源码分析10-MVC实现原理
AKA毅成的博客
11-15
172
深度解析Cat源码系列专栏点击访问
持续更新中
文章目录CAT源码分析10-MVC实现原理1.概述2.MVC配置详解2.1JavaBean+Annocation配置2.2配置管理和解析2.3解析逻辑3.MVC流程简介3.1Servlet拦截入口3.2MVC类3.3Service方法
CAT源码分析10-MVC实现原理
1.概述
CAT没有使用SpringMVC作为MVC架构的实现框架,而使用了一套自研的MVC框架。
虽然与SpringMVC实现不同,但MVC的框架基本原理.
互联网服务客户流失分析(个人练习+源代码)
jidongdaoshi的博客
01-29
1355
互联网服务客户流失分析(个人练习+源代码)
机器学习客户流失_机器学习的客户流失预测
weixin_26750481的博客
08-27
959
机器学习客户流失Buildingupandkeepingaloyalclientelecanbechallengingforanycompany,especiallywhencustomersarefreetochoosefromavarietyofproviderswithinaproductcategory.Furthermore,...
【CAT魔改】为cat-home添加链路追踪查询
夫礼者的专栏
08-18
574
扩展cat-home,实现traces链路追踪。
数据洞察和数据分析_利用数据洞察力提高客户保留率
weixin_26642481的博客
09-16
822
数据洞察和数据分析首先,让我们知道客户保留率是多少?(Firstofall,letusknowwhatcustomerretentionis?)
Itrepresentsthenumberofcustomerswhocontinuepurchasingfromacompanyaftertheirfirstpurchaseinsimplela...
时间序列深度学习:状态LSTM模型预测太阳黑子(上)
R语言中文社区
06-15
7844
作者:徐瑞龙整理分享量化投资与固定收益相关的文章博客专栏: https://www.cnblogs.com/xuruilong100 本文翻译自《TimeSeries...
用户画像·用户流失预测
AIWorld
05-09
863
文章目录1、为什么预测流失2、需求分析
1、为什么预测流失
每个企业都渴望建立和保持一个忠实的客户群,而事实是由于各方面原因不可避免的会流失一些用户。
如果我们根据用户的活跃度及消费情况,判断用户的流失意向,及时对有流失趋向的用户进行营销召回,这对公司来讲是非常有必要的。
2、需求分析
模型标签(用户流失的概率)
...
神经网络在客户流失预测系统的应用
08-26
摘要:本文简述了BP神经网络的基本原理,提出了一种基于BP神经网络的客户流失
预测模型。
实验表明,该模型的辨识精度高,能正确的对客户的需求进行评估,以减少客户
流失来提高企业的利润。
关键词:神经网络;BP算法;客户流失;CRM
集成学习和随机森林方法
qq_38325803的博客
11-03
708
集成学习和随机森林方法
介绍
本次实验介绍了集成学习的概念及主要方法,包括Bootstraping、Bagging、随机森林,随后计算随机森林中各个特征的重要性,找出对模型贡献较大的特征。
知识点
集成
Bootstraping
Bagging
随机森林
特征重要性
集成
之前的几个实验中,介绍了不同的分类算法,以及验证、评估模型的技术。
现在,假设已经为某一特定问题选中了最佳的模型,想进...
机器学习笔记-GradientBoostedDecisionTree
XuShuai
11-29
8397
上一篇介绍了RandomForest,该算法利用Bagging中的bootstrapping机制得到不同的DecisionTree,然后将这些DecisionTree融合起来。
除了基本的Bagging和DecisionTree之外,RandomForest还在DecisionTree中加入了更多的randomness。
有了这些机制之后,我们发现这个算法可以利用OOB数据做self-V
scikit-learn(工程中用的相对较多的模型介绍):1.11.Ensemblemethods
热门推荐
mmc2015的专栏
08-04
1万+
参考:http://scikit-learn.org/stable/modules/ensemble.html
在实际项目中,我们真的很少用到那些简单的模型,比如LR、kNN、NB等,虽然经典,但在工程中确实不实用。
今天我们关注在工程中用的相对较多的Ensemblemethods。
Ensemblemethods(集成方法)主要是综合多个esti
从decisiontree到bagging、boosting
u010859324的博客
07-26
1683
本文主要讲解randomforest、GBDT、xgboost的原理,对三种算法进行比较
一、bagging和boosting
参考资料:
https://www.quora.com/What-are-the-differences-between-Random-Forest-and-Gradient-Tree-Boosting-algorithms...
数据分析案例——客户流失分析与预测
liushuang99的博客
10-04
3951
客户流失分析与预测
一、数据来源
https://www.kaggle.com/blastchar/telco-customer-churn
二、数据整理
1、导入函数包
importpandasaspd
importnumpyasnp
importmatplotlib.pyplotasplt
importseabornassns
2、导入数据并展示
data=pd.read_csv(r"D:\百度网盘\数据分析—实例\运营商客户流失分析与预测\WA_Fn-UseC_-Telco-C
对抗样本方向(AdversarialExamples)2018-2020年最新论文调研
huitailangyz的博客
06-15
1万+
调研范围
2018NIPS、2019NIPS、2018ECCV、2019ICCV、2019CVPR、2020CVPR、2019ICML、2019ICLR、2020ICLR
2018NIPS
ContaminationAttacksandMitigationinMulti-PartyMachineLearning(防御)
作者:JamieHayes(UniveristyCollegeLondon)OlgaOhrimenko(MicrosoftResearch)
摘要:Machine
cat服务端获取路由列表原理解析
star_apple的博客
03-24
454
Cat客户端上报消息时会先从服务端拉取路由列表。
请求URL如:http://127.0.0.1:2281/cat/s/router?domain=strict-kevin-dubbo-test&ip=10.10.135.118&op=json。
服务端由com.dianping.cat.system.page.router.Handler类来负责处理请求。
源码分析
请求处理入口,H...
“相关推荐”对你有帮助么?
非常没帮助
没帮助
一般
有帮助
非常有帮助
提交
©️2022CSDN
皮肤主题:大白
设计师:CSDN官方博客
返回首页
RaymondLove~
CSDN认证博客专家
CSDN认证企业博客
码龄10年
暂无认证
90
原创
7万+
周排名
95万+
总排名
11万+
访问
等级
1850
积分
74
粉丝
95
获赞
28
评论
481
收藏
私信
关注
热门文章
Mac完全卸载IDEA的方法(可重新安装,亲测有效)
15599
机器学习知识点总结-拉格朗日乘子法(LagrangeMultiplierMethod)详解
8047
目标检测(ObjectDetection)(一):评估标准(mAP,IOU,NMS,FPS)
7865
win10深度学习开发环境搭建:Python3.7+VS2019+Anaconda3+cuda10.1+cuDNN+tensorflow-gpu+keras-gpu
6616
大数据概论-学习笔记-01大数据导论
4920
分类专栏
CS294_158笔记:无监督学习和生成模型
1篇
Few-ShotSegmentation
4篇
元学习
1篇
图像处理
1篇
论文学习笔记
1篇
英语学习
CS231n学习笔记
18篇
《数据库系统概念》学习笔记
13篇
《程序员代码面试指南》学习笔记
11篇
机器学习
6篇
Python相关
6篇
问题总结
25篇
《Tensorflow实战》学习笔记
1篇
cs224n学习笔记
1篇
最新评论
从零开始,元学习入门
小手冰凉_:
写的很好,感谢
Mac完全卸载IDEA的方法(可重新安装,亲测有效)
m0_67707505:
~/.idea/.在哪里
《数据库系统概念》学习笔记——第二章关系模型介绍
Raocel:
你好,我想问下,section和time_slot为什么有根连线,而且还没有箭头呢?是因为time_slot_id属性不是外码,但是另一关系的主码的一部分吗?
【Few-ShotSegmentation论文笔记】CANet,IAAA,2019
邵市民:
对于k-shot问题,文章figure3是不是画的有问题?浅黄色、橙色、深黄色的立方体不应该都是Queryimge嘛应该是同一个特征表示呀?怎么图是三种颜色。
解决IDEAExternallibraries中不显示Maven中引入的repository
qq_47212572:
感谢感谢
您愿意向朋友推荐“博客详情页”吗?
强烈不推荐
不推荐
一般般
推荐
强烈推荐
提交
最新文章
从Autoencoder到VAE及其变体
【Few-ShotSegmentation论文笔记】CANet,IAAA,2019
【Few-ShotSegmentation论文阅读笔记】PANet:Few-ShotImageSemanticSegmentationwithPrototype,ICCV,2019
2021年5篇
2020年34篇
2019年32篇
2018年7篇
2016年12篇
目录
目录
分类专栏
CS294_158笔记:无监督学习和生成模型
1篇
Few-ShotSegmentation
4篇
元学习
1篇
图像处理
1篇
论文学习笔记
1篇
英语学习
CS231n学习笔记
18篇
《数据库系统概念》学习笔记
13篇
《程序员代码面试指南》学习笔记
11篇
机器学习
6篇
Python相关
6篇
问题总结
25篇
《Tensorflow实战》学习笔记
1篇
cs224n学习笔记
1篇
目录
评论
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
查看更多评论
打赏作者
RaymondLove~
你的鼓励将是我创作的最大动力
¥2
¥4
¥6
¥10
¥20
输入1-500的整数
余额支付
(余额:--)
扫码支付
扫码支付:¥2
获取中
扫码支付
您的余额不足,请更换扫码支付或充值
打赏作者
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额
0
抵扣说明:
1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、C币套餐、付费专栏及课程。
余额充值