datapath system data dcentric analyt process engin larg data warehous sketch sampl data stream decis tree classif decis tree system abstract technolog build knowledg base system induct infer exampl demonstr successfulli practic applic paper summar approach synthes decis tree varieti system describ system id detail result recent studi show wai methodolog modifi deal inform noisi incomplet report shortcom basic algorithm discuss mean overcom compar paper conclud illustr current research direct am sketch exact comput aggreg queri size join relat requir larg amount memori constrain data stream commun constrain distribut comput larg process time situat approxim techniqu provabl guarante sketch solut perform sketch depend crucial abil gener pseudo random number paper investig theoret empir problem gener wise independ pseudo random number gener wise independ pseudorandom number fast rang summabl sum linear time specif contribut provid comparison pseudorandom number gener scheme studi theoret empir fast rang summat properti wise independ gener scheme provid algorithm fast rang summat wise independ scheme bch extend ham show convinc theoret empir evid extend ham scheme perform wise independ scheme estim size join relat am sketch wise independ scheme gener estim significantli outperform state art solut problem size spatial join select estim analyz techniqu improv fault toler aggreg tree sensor network multi dqueri optim sketch dbase estim analyz techniqu improv fault toler aggreg tree sensor network fault toler aggreg heterogen sensor network turbo dcharg estim converg dbo dbo databas system util random algorithm give statist meaning estim final answer multi tabl disk base queri start finish queri execut dbo time til util ttu time dbo give estim overli larg case databas tabl join queri case join queri includ select predic tabl data skew paper describ turbo dbo prototyp databas system answer multi tabl join queri scalabl fashion dbo turbo dbo lower ttu dbo kei innov turbo dbo make algorithm rememb partial match tupl random fashion tupl satisfi boolean predic queri possibl grown tupl contribut final queri result time semi danalyt method analyz model model select measur base moment analysi articl propos moment base method studi model model select measur focus probabilist space classifi induc classif algorithm dataset obtain effici character comput moment visual result formula complic direct interpret assum data drawn independ ident distribut underli probabl distribut space dataset establish gener relationship gener error hold set error cross valid error leav error exemplifi method result studi behavior error naiv bay classifi confid bound sampl dbase group estim dbo databas system demonstr prototyp dbo databas system dbo design facilit scalabl analyt process larg data archiv dbo analyt process perform competit databas system unlik exist research industri system dbo maintain statist meaning guess final answer queri start finish queri process guess accur second minut answer queri hour result signific save user comput time user abort queri happi guess accuraci confid bound sampl dbase group estim scalabl approxim queri process dbo engin aggreg method larg dscale sensor network multipl dchoic random network server load balanc statist analysi sketch estim sketch techniqu provid approxim answer aggreg queri data stream distribut comput small space summari linear properti requir type applic preval method analyz sketch moment analysi distribut independ bound base moment method produc clean easi interpret theoret bound deriv asymptot result theoret bound obscur fine detail behavior sketch indic type sketch practic signific empir comparison sketch techniqu publish make choic harder paper close sketch techniqu propos literatur statist point view goal determin properti actual behavior produc tighter confid bound interestingli statist analysi reveal techniqu fast agm count min provid result case order magnitud theoret predict conduct extens empir studi compar sketch techniqu order corrobor statist analysi conclus draw studi expect perform sketch crucial techniqu practition conclus studi fast agm sketch full spectrum problem close sketch techniqu make fast agm sketch prefer choic irrespect situat scalabl approxim queri process dbo engin pseudo drandom number gener sketch dbase estim exact comput aggreg queri size join relat requir larg amount memori constrain data stream commun constrain distribut comput larg process time situat approxim techniqu provabl guarante sketch solut perform sketch depend crucial abil gener pseudo random number paper investig theoret empir problem gener wise independ pseudo random number gener wise independ pseudorandom number fast rang summabl sum linear time specif contribut provid comparison pseudorandom number gener scheme studi theoret empir fast rang summat properti wise independ gener scheme provid algorithm fast rang summat wise independ scheme bch extend ham show convinc theoret empir evid extend ham scheme perform wise independ scheme estim size join relat am sketch wise independ scheme gener estim significantli outperform state art solut problem size spatial join select estim fast rang dsummabl random variabl effici aggreg estim exact comput aggreg queri requir larg amount memori constrain data stream commun constrain distribut comput larg process time situat approxim techniqu provabl guarante sketch viabl solut perform sketch crucial depend abil effici gener pseudo random number paper investig theoret empir problem gener wise independ pseudo random number gener wise independ pseudorandom number fast rang summabl sum linear time specif contribut provid empir comparison pseudo random number gener scheme studi theoret empir fast rangesumm practic wise independ gener scheme provid effici implement wise independ scheme show convinc theoret empir evid extend ham scheme perform wise independ scheme estim size join am sketch wise independ gener scheme produc estim significantli outperform state art solut problem size spatial join select estim sort dmerg dshrink join common oper analyt queri process applic aggreg function result relat join describ algorithm call sort merg shrink sm join comput answer queri larg disk base input tabl kei innov sm join input data cluster statist random fashion disk time join onlin statist estim eventu answer queri probabilist confid bound user monitor progress join execut stop join satisfi estim accuraci run algorithm complet total time requir longer common join algorithm contrast onlin join algorithm offer statist guarante offer guarante long input data fit main memori histogram revisit histogram approxim method aggreg join tradit statist assumpt interpret histogram justifi approxim queri process method base element bucket frequenc call uniform distribut assumpt paper show significantli restrict statist assumpt element bucket randomli arrang frequenc lead ident formula approxim aggreg queri histogram observ identifi scenario histogram suit approxim method fact show situat sampl sketch significantli wors provid tight error guarante qualiti approxim time show averag histogram poor approxim scenario disk dbase join probabilist guarante common oper analyt queri process applic aggreg function result relat join describ algorithm comput answer queri larg disk base input tabl kei innov algorithm time onlin statist estim eventu answer queri probabilist confid bound user monitor progress join execut stop join satisfi estim accuraci run algorithm complet total time requir longer common join algorithm contrast onlin join algorithm offer statist guarante offer guarante long input data fit core memori onlin estim subset dbase sql queri largest databas todai larg answer queri minut hour dai address problem make approxim algorithm previou work onlin aggreg consid give onlin estim everincreas accuraci aggreg function relat join select queri exist work applic onlin estim subset base sql queri queri correl subqueri link outer queri exist exist orin claus queri asexcept intersect subset base queri paper develop algorithm onlin estim queri difficult problem provid probabilist accuraci guarante time queri execut sketch dbase multi dqueri process data stream abstract recent year wit increas interest design algorithm queri analyz stream data data fix order limit memori provid approxim answer queri continu data stream crucial requir applic environ exampl includ larg telecom ip network instal perform data part network continu collect analyz random techniqu base comput small sketch synops stream recent shown effect tool approxim result singl sql queri stream data tupl paper investig problem aris data stream sketch process multipl queri concurr demonstr presenc multipl queri express intellig share sketch concurr queri evalu result substanti improv util sketch space qualiti result approxim error guarante provid suffici condit multi queri sketch share guarante correct result estim process prove optim sketch share typic rise hard question propos heurist algorithm find good sketch share configur practic result experiment studi realist workload verifi effect approach demonstr benefit sketch share methodolog gossip dbase comput aggreg inform comput result paradigm shift central highli distribut system massiv scale massiv instabl node link failur norm except highli volatil system decentr gossip base protocol emerg approach maintain simplic scalabl achiev fault toler inform dissemin secret scalabl linear regress tree algorithm recent increas interest develop regress model larg dataset accur easi interpret regressor properti regress tree linear model leav algorithm propos construct scalabl paper propos regress tree construct algorithm accur scale larg dataset main idea intermedi node em algorithm gaussian mixtur find cluster data local transform regress problem classif problem base close cluster good split measur gini gain determin split variabl split point classif tree construct scalabl algorithm enhanc emploi scalabl version em classif tree construct algorithm test real artifici data show propos algorithm accuraci compar linear regress tree algorithm requir order magnitud comput time larg dataset process complex aggreg queri data stream recent year wit increas interest design algorithm queri analyz stream data data fix order limit memori provid approxim answer queri continu data stream crucial requir applic environ exampl includ larg telecom ip network instal perform data part network continu collect analyz bia correct classif tree construct 