Load Prediction Using Hybrid Model for Computational Grid

Size: px
Start display at page:

Download "Load Prediction Using Hybrid Model for Computational Grid"

Transcription

1 Load Predicion Using Hybrid Model for Compuaional Grid Yongwei Wu, Yulai Yuan, Guangwen Yang 3, Weimin Zheng 4 Deparmen of Compuer Science and Technology, Tsinghua Universiy, Beijing 00084, China, 3, 4 {wuyw, ygw, zwm-dcs}@singhua.edu.cn yuan-yl05@mails.singhua.edu.cn Absrac Due o he dynamic naure of grid environmens, schedule algorihms always need assisance of a long-ime-ahead load predicion o make decisions on how o use grid resources efficienly. In his paper, we presen and evaluae a new hybrid model, which predics he n-sep-ahead load saus by using inerval values. This model inegraes auoregressive (AR) model wih confidence inerval esimaions o forecas he fuure load of a sysem. Meanwhile, wo filering echnologies from signal processing field are also inroduced ino his model o eliminae daa noise and enhance predicion accuracy. The resuls of experimens conduced on a real grid environmen demonsrae ha his new model is more capable of predicing n-sep-ahead load in a compuaional grid han previous works. The proposed hybrid model performs well on predicion advance ime for up o 50 minues, wih significan less predicion errors han convenional AR model. I also achieves an inerval lengh accepable for ask scheduler. I. INTRODUCTION Grid compuing [] is he high-performance and inerne-based infrasrucure ha aggregaes geographically disribued and diverse resources o deliver compuaional power o users in a ransparen way, supporing large-scale, resource-inensive and disribued applicaions. Task scheduling is an imporan facor on improving resource usage efficiency in such a dynamic disribued environmen. Obviously, sysem load predicion can be used o forecas ask run ime [8] and guide scheduling sraegies, hus o achieve high performance and more efficien resource usage [] [3] [4]. In mos previous works, load predicion models, such as mean-based mehods, median-based mehods, auoregressive (AR) models, polynomial fiing, ec., use poin value o represen fuure load saus of worksaions, clusers and grids. Papers [5] [6] [0] describe such poin value predicion models and heir performances. Some oher works, like [], implemen inerval value predicion on srucural model o forecas sysem performance in a disribued producion environmen. Though poin value predicion sraegy has been widely adaped by recen sysem load predicion models, his kind of sraegy do have some drawbacks for highly disribued environmens. Because grid asks usually ake long run ime, ask scheduler in a compuaional grid needs a comparaively long-ime-ahead predicion or an n-sep-ahead predicion wih large sep inervals. Poin value can hardly cover load variabiliy in such a long ime frame (i.e., sep inerval), and herefore hese convenional poin value predicion models do no perform well on n-sep-ahead load predicion for large n. The inerval value predicion model proposed in [] addresses hese wo problems of poin value predicion and provides valuable addiional informaion ha can be used by a ask scheduler. Bu i does no achieve a reasonable predicion inerval lengh, which is imporan for he ask scheduler. Furhermore, mos of he previous predicion works also ignore some problems in he hisory load daa, which significanly affecs he predicion accuracy. One problem is he load measuremen error, which inherenly exiss in all measuremen mehods and can be amplified by a predicion model; anoher problem is he load daa noise inroduced by he load flucuaion in a compuaional grid. The endency of hisory daa is used by some endency-rack models, like AR model or polynomial fiing, o forecas fuure saus. Bu he fuure load endency migh be disored or concealed in he noisy daa, and herefore misleads he endency rack models, which will consequenly impair predicion accuracy. Our conribuion in his paper is ha we propose a new hybrid model, which can forecas far more n-sep-ahead load saus han previous models and mehods, wih comparaively less predicion errors and accepable predicion inerval lengh for ask scheduler and oher uiliies. To deal wih he dynamic naure of compuaional grids and he needs of ask schedulers, we inegrae AR model wih confidence inerval esimae in our hybrid model, where AR model is used as a basic n-sep-ahead poin value predicion mehod, and he confidence inerval of fuure load saus is esimaed based on boh hisorical and prediced poin values. In order o enhance predicion accuracy, Kalman Filer [] is used o minimize he measuremen errors of hisory load daa, and Savizky-Golay filer funcions [7] are used as he endency revealing ool for daa noise eliminaing. The resuls of our experimens on a real compuaional grid environmen demonsrae ha his hybrid model is of an excellen capabiliy on n-sep-ahead load predicion, and i also achieves much smaller predicion mean square error han convenional AR model. Furhermore, n-sep-ahead load predicion is achieved for large n in our hybrid model as well. The corresponding predicion advance ime can be up o 50 minues, wih low predicion mean square error (0.04 on average) and accepable confidence inerval lengh (less han 0%) for he ask scheduler. The res of his paper is organized as follows. Secion II inroduces he relaed work. Secion III analyses he causes of /07/$ IEEE 35 8 h Grid Compuing Conference

2 measuremen errors and describes an error minimize ool Kalman filer. Secion IV presens our hybrid model on load predicion, and also inroduces a signal smoohing algorihm, Savizky-Golay filer, which is used o eliminae daa noise and reveal he endency of he daa more clearly. Secion V describes he experimen resuls of our hybrid model on a real compuaional grid. Finally, we conclude our work and describe our fuure work in secion VI. II. RELATED WORK Previous works [7] [0] indicae ha load has such properies as self-similariy and epochal behaviour, and is srongly correlaed over ime, which implies ha sysem load is consisenly predicable from he pas behaviour. Therefore, correcly correlaing he hisory daa wih he fuure values is he kernel o make accurae predicions. Several load predicion sraegies in he disribued environmen have been proposed in he pas. [4] exends he predicion by using seasonal variaion and Markov model-based mea-predicor in addiion o seasonal variaion for -sep-ahead predicion. [5] proposes a muli-resource predicion model ha uses boh auocorrelaion and cross correlaion o achieve higher predicion accuracy. The Nework Weaher Service (NWS) provides a dynamically monioring and forecasing mehod o implemen -sep-ahead predicion on nework and compuaional resources [5]. NWS uses various predicion mehods, such as mean-based mehods, median-based mehods, and AR mehods o forecas he fuure sysem saus a he same ime. NWS racks he accuracy of all predicors, and selecs he one exhibiing he lowes cumulaive error measure a any given momen o generae a forecas. In his way, NWS auomaically idenifies he bes forecasing echnique for any given resource. In Dinda s paper [6], he predicion power of several linear models, including AR, MA, ARMA, ARIMA, and ARFIMA are evaluaed in deail. Their resuls show ha simple, pracical models such as AR are sufficien for load predicion and AR(6) models or beer are recommended for CPU load predicion. In [6], several homeosaic and endency-rack -sep-ahead predicion sraegies are presened and evaluaed. Homeosaic mehods assume ha he load of a sysem always remains seady in a given ime frame, while endency-rack sraegies are based on he assumpion ha he endency of he load variabiliy exiss in he load daa, and can be properly revealed by he endency-rack models. The predicion of fuure value is adjused according o he magniudes of he las load measuremen and he las predicion error, and he evaluaions of his echnique show ha i ouperforms NWS for CPU load predicion. The predicion sraegies menioned above in his secion are all using poin value o represen he fuure load saus. Convenional poin value predicion models are ofen inaccurae since hey can only represen one poin in a range of possible behaviours. In [], sochasic inerval values are inroduced in he area of sysem load predicion, which can address his problem. Auhors of he paper define sochasic values using inervals, and hen hey define he inerval arihmeic formulas and implemen inerval value on he poin value predicion model. The inerval value predicion model in [] performs more effecively han poin value predicion models in producion environmens. III. MEASUREMENT ERROR AND ERROR MINIMIZATION Measuremen error is unavoidable in every measuremen mehod. On one hand, he monioring sensors bring some degrees of disurbers ino arge sysem; on he oher hand, some measuremen mehods gain performance value by esimaion, for example, compuing CPU availabiliy from he average load of he sysem, or using he value of a ime poin in a period o represen he average performance of ha period. Because of hese, measuremen would ineviably be inaccurae. Even worse, a predicion model would amplify hese measuremen inaccuracies in he predicion resul. In his secion, we ignore he errors caused by he monioring sensors, considering ha hese errors are relaively small compared wih hose caused by he measuremen mehods. In his secion, he measuremen error repored by he Compuer Science and Engineering Deparmen a UCSD and measuremen error observed by us in Bioinformaics Applicaion Grid (BioGrid) of ChinaGrid [0] are discussed firs, followed by he inroducion of he soluion o he problem, which use he famous Kalman filer o minimize he error observed in he BioGrid. A. Measuremen mehods and errors observed by UCSD In [0], hree kinds of measuremen mehods are given and he resuls gahered by monioring a collecion of worksaions and compue servers in he Compuer Science and Engineering Deparmen a UCSD are repored. To deermine he accuracy of hese hree mehods, hey compared he measuremens of hese hree mehods wih he readings generaed by an independen en-second, CPU bound process which hey refer o as he es process. The mean absolue measuremen errors of hese hree mehods observed for differen hoss a UCSD vary from as small as 3.% o as big as 4.3%, which would ineviably deeriorae he predicion accuracy. B. Measuremen mehods and errors observed in he BioGrid of ChinaGrid The BioGrid of ChinaGrid (see secion V for deails) uses a simple mehod o generae he average performance value of every fixed ime frame. The monior cenre gahers various performance daa from he nodes of he Grid every 60 seconds, and uses he poin value observed a he beginning of he monior inerval o represen he average performance of his ime frame. Because he asks running on he BioGrid are ime-consuming and he CPU load of each node in he BioGrid is sable in a moderae period, we assume ha he load daa deeced by he node sensors a a cerain ime poin can be used o represen he average load of he ime frame o which his ime poin belongs o. We also use he es process mehod in [0] o deermine he measuremen accuracy. We se a es process o repor 36

3 CPU load every 0 seconds, whereas he original BioGrid sensors repor he monioring informaion every 60 seconds. Fig. deails he measuremen errors observed from a node of he BioGrid. Fig. CPU load measuremen errors observed on one node (in Tsinghua Universiy) of he BioGrid. Mean absolue Measuremen Error is , and Mean Square Measuremen Error is C. Using Kalman filer o minimize measuremen errors In 960, R.E. Kalman proposed a recursive soluion o he discree-daa linear filering problem in his famous paper []. Kalman filer has been exensively researched and applied, paricularly in he area of auonomous or assised navigaion [9]. I provides an efficien recursive mehod o esimae he sae of a process, and minimizes mean square error. Kalman filer uses feedback conrol o esimae a process: firsly he filer esimaes he process sae a cerain ime and secondly obains feedback from he measuremens. Thus we can divide he equaions of Kalman filer ino wo groups: ime updae equaions and measuremen updae equaions. The specific equaions for ime and measuremen updaes are presened in [] and [9]. Our work applies Kalman filer o minimize measuremen errors, and herefore enhance predicion accuracy, which is evaluaed in secion V. Kalman filer has wo iniial parameers, process noise covariance Q and measuremen noise covariance R. In pracice, hese wo parameers migh vary wih each ime sep or measuremen, bu we assume hey are consans here. Presuming a very small process variance, we le Q = e 5, and fix he measuremen variance R = (0.06) = Because his is he rue average measuremen error variance we observed from he pervious load daa in he BioGrid of ChinaGrid, we would expec he bes performance in erms of balancing responsiveness and esimae variance. IV. PREDICTION Task schedule and load balance sraegies can benefi a lo from accurae load predicion. Because he load variabiliy and resource consuming siuaion on a compuaional grid are srongly correlaed over ime and have he properies such as self-similariy, period sabiliy, long ime load predicion should be feasible. In his secion, we inroduce our hybrid model on load predicion for compuaional grid, which inegraes AR model wih confidence inerval esimae. We use AR model as he basic linear model o predic an n-sep-ahead load poin value, and hen hisory load daa and predicion load daa are used o esimae he possible load inerval of an n-sep-ahead fuure ime frame. Even wih Kalman filer applied o eliminae he measuremen errors, he noise sill exiss in he load daa, and he endency of load variabiliy is concealed in he flucuaing daa noise. We need a smoohing algorihm o expose he feaures of load daa, and provide a reasonable saring approach for parameric fiing. If he curve of load daa is smoohed, he endency of fuure sae would be more obvious and he predicion would be more accurae. A. Smooh hisory load daa using Savizky-Golay filer Savizky-Golay smoohing filer [7] is ypically used o "smooh ou" a noisy signal wih a large frequency span. In our work, we use Savizky-Golay smoohing filer o smooh load daa in several seps of our predicion model. Savizky-Golay filer can be regarded as a generalized moving average. The filer coefficiens are derived by implemening an unweighed linear leas square fi using a polynomial of a given degree. A higher degree polynomial makes i possible o achieve a high level of smoohing wihou aenuaing daa feaures. For frequency daa, his filer is effecive a preserving he high-frequency componens of he signal. In our work, we se he parameers of Savizky-Golay filer wih span=0, pd (polynomial degree) = 4. B. Predic fuure sae using AR models and inerval value In [5], he performance predicion minimum mean square error and mean percenage error values for each nework seing of differen predicion models are summarized. Alhough bes predicor of each performance characerisic is, in general, no obvious and varies from resource o resource, wih a series evaluaion in [6], AR models are recommended o predic compuing resource relaed meric, such as CPU load. ) Using AR models o predic a fuure poin value The main idea behind using a linear ime series model in load predicion is o rea he sequence of periodic samples of performance, <Z >, as a realizaion of a sochasic process ha can be modelled by linear saionary models, such as AR, MA, ARMA, ARIMA, ARFIMA, ec. [6]. The coefficiens of he model can be esimaed by observing pas daa sequence. If he acion of he model can cover mos of he daa sequence variabiliy, is coefficiens could be used o esimae fuure daa sequence values wih low mean squared error. AR model is a model used o find an esimaion of a daa based on previous inpus of he daa. The general form of a ph-order AR(p) model is as follows: p X = ϕ X + ε i= i i (4.) 37

4 AR model consiss of wo pars: an error or noise parε, and an auoregressive summaion ϕ i X. The summaion i represens a fac ha curren inpu value depends only on he previous inpu values. The variable p is he order of AR model. The higher he order of AR model, he more accurae a represenaion will be. As he order of he model approaches infiniy, we ge almos an exac represenaion of he inpu daa; however, he compuaional expense on calculaing he AR(p) coefficiens increases wih he p order. Therefore, we should balance he predicion accuracy and he compuaional expense. In [6], he auhors colleced a large number of Hz benchmark load races, which capure all he dynamics of load signal, and subjeced hem o a deailed saisical analysis, drew a conclusion ha AR(6) models or beer are recommended for hos load predicion. The problem in AR(p) analysis is o derive he "bes" values for φi, given a series X -i. The majoriy of mehods assume ha he series X -i is linear and saionary. By convenion he series X -i is assumed o be zero mean, if no his is simply anoher erm ε in he summaion of he equaion (4.). Even for relaively large values of p, wih he Burg algorihm [3] ha we used o compue he AR(p) coefficiensϕ i, his can be done almos insananeously. Then he -sep-ahead value X can be easily compued. However, in highly disribued compuaional grid environmens, -sep-ahead load forecasing can no saisfy he demands of ask scheduler. The furher he load predicion of a compuaional grid can reach, he more appropriae he ask scheduler can make scheduling sraegies and balance he load. So wha he scheduler in hese large disribued environmen needs is he n-sep-ahead predicion (n =,3,4 or even 30,40,50..), which can forecas he load variabiliy far more ahead before i really happens. AR(p) model aemps o predic an oupu X of a sysem based on he previous inpus (X -, X -, X -3...) and he coefficiens ( ϕ, ϕ,... ϕ p ). Before we use AR(p) model o implemen he n-sep-ahead predicion, he following wo imporan assumpions have o be addressed: The coefficiens of AR(p) model represen he variabiliy of he hisory load daa X -, X -,, and hey are also suiable o represen he endency of he load in a fuure period, and he variabiliy of fuure unknown daa X, X +, X +n-, and X +n-. The predicion of he X +n- is based on he daa X +n-, X +n-3,, X, X -, X -,,and only daa X -, X -, is he rue measured daa; X, X +,, X +n- mus be compued sep by sep before predicing he X +n-, and hen X, X +,, X +n- can be assumed as he rue daa when we predic he X +n-. Based on hese wo assumpions, he n-sep-ahead performance poin value predicion is generaed using he following algorihm in Table I. p i= TABLE I N-STEP-AHEAD AR(P) MODEL PREDICTION ALGORITHM ϕ ϕ,... ϕ, Compue he AR coefficiens p based on he rue monioring daa X -, X -, ; for i = o n use he AR coefficiensϕ, ϕ,... ϕ p and X -+i-, X -+i-,, o compue he X -+I ; end ) Confidence inerval value predicion AR(p) model predics he fuure load saus by using poin value. In pracice, poin value is ofen an ideal esimae, or a value ha is accurae only for a given ime frame []. I ofen represens he average load in a given ime frame, and can no reflec he variabiliy of a sysem in his ime frame. In he highly disribued sysems like grids, poin values can no convey enough dynamic load informaion for he ask scheduler o make scheduling sraegies. Anoher ype of value wih he form of X±Y, or X~X, is referred as inerval value. The load of a compuaional grid varies over ime, and he inerval value can provide an esimae of his variabiliy in a given ime frame. Fig. CPU load variabiliy of THU node cluser of he BioGrid Unforunaely, i is difficul o cover all he daa of a given ime frame in an inerval value. Someimes he load can be exremely abnormal and flucuan in a wide range, as shown in Fig.. If we wan o use an inerval value o represen all he load variabiliy in a ime frame, i would lenghen he inerval and be meaningless for grid ask scheduler. In [], many large-sample-size real phenomena generae disribuions which are close o normal disribuion. Therefore we can use confidence inerval o esimae his large sample size normal disribuion of load, and for he small sample size daa, -disribuion is more suiable. This approach can eliminae some abnormal behaviour of load variabiliy, and dramaically shoren he inerval lengh in hese siuaions. Confidence inervals represen a means of providing a range of values in which he flucuaion value can be expeced o lie. This esimaed range is calculaed from a given se of sample daa. Here, we se he sample size o be CIwindow+N, where CIwindow (Confidence inerval window) is he basic sample windows size, and N is he addiional windows size, which 38

5 includes he n-sep-ahead predicion values generaed by AR(p) model. We se he confidence level o be 95% in our experimens. The confidence inerval value here is described as a pair of wo poin values (X lower (), X upper ()), where X lower () denoes he lower limi of he load predicion of ime frame, while X upper () means he upper limi of he load predicion of ime frame. In pracice, suddenly abnormal load flucuaion in a compuaional grid does no happen frequenly and ofen lass ransiorily, which makes his phenomenon hard o be prediced. In his paper, we assume ha load is sable in mos of he ime, and herefore he load confidence inerval predicion values should also display as wo smoohness curves which reflec he sabiliy of he load. So afer we compue he n-sep-ahead load confidence inerval, we use Savizky-Golay filer inroduced in formula 4. o smooh he inerval daa, he lower limi daa and he upper limi daa, o do some correcions on he original load confidence inerval predicion values. TABLE II LOAD PREDICTION PROCESS OF HYBRID MODEL. Using Kalman filer o minimize he measuremen errors of he hisory daa X measure (), generae he more accuracy daa X filer ().. Using Savizky-Golay filer o smooh he daa X filer () wih he former daa X filer (-), X filer (-),., generae he daa X smooh (). ϕ ϕ,... ϕ, 3. Compue he AR(p) coefficiens p based on he X smooh (), X smooh (-) 4. For he given n-sep-ahead parameer N, using he X smooh (), X smooh (-) and AR(p) coefficiens ϕ, ϕ,... ϕ p o compue he predic fuure value of X predic (+), X predic (+) X predic (+n). 5. Compue he confidence inerval (X lower (+n), X upper (+n)) wih sample size CIwindow+N, a confidence level 95%, using he daa X filer (-CIwindow+),X filer (-), X filer (-), X predic (), X predic (+) X predic (+n). 6.Using Savizky-Golay filer o smooh he predicion inerval value (X lower (+n), X upper (+n)) wih he former daa (X lower (+n-), X upper (+n-)), (X lower (+n-), X upper (+n-)),, generae he smoohed predicion inerval value(x smoohed_lower (+n), X smoohed_upper (+n)) 3) The whole predicion process The whole process of our hybrid model on n-sep-ahead load predicion is described in Table II. V. EVALUATION We evaluaed our hybrid model in an real grid environmen, BioGrid. This compuaional grid provides bioinformaics supercompuing services for bioinformaics researchers hrough Web inerface ransparenly. I is par of he China Educaion and Research Grid Projec ChinaGrid, an imporan projec funded by Chinese Minisry of Educaion. ChinaGrid aims o inegrae heerogeneous mass resources disribued in he China Educaion and Research Nework (CERNET) ino a public plaform for research and educaion in China. The BioGrid has 7 nodes disribued in 6 disan ciies of China (Beijing, Shanghai, Wuhan, Ji nan, Lanzhou, Xi an), and each of hem is a cluser wih processor number from 64 o 56. Wihou he loss of generaliy, we use a se of 7 series of CPU load daa of he BioGrid from he ChinaGrid Super Vision (CGSV [9]) o evaluae our hybrid model, he oal sample size is These CPU load daa is gahered from he fron-end machines of 7 nodes in he BioGrid every 60 seconds, and he CPU load of each node(cluser) is denoed as 0% o 00%, using poin value o represen he average CPU load of every ime frame. All he following evaluaion resuls are generaed from his se of CPU load daa. A. The opimal parameers of he AR(p) model There is no doub ha more accurae poin value predicion in our work will lead o a nicer confidence inerval value forecasing. In he conclusion of [6], AR(6) model or beer are appropriae for he hos load predicion. We measured he accuracy of he AR n-sep-ahead predicion wih differen n, where he resul of his measuremen reveals he capabiliy of AR(p) model in he furher n-sep-ahead predicion. Define he mean square predicion error as (5.) MSEpo in_ value_ predicion( ) = ( errorpo in_ value_ predicion( i) ) (5.) + i= 0 TABLE III MEAN SQUARE ERROR OF AR(P) N-STEP-AHEAD POINT VALUE PREDICTION, CPU LOAD DATA FROM THE BIOGRID, TOTAL 00% P sep ahead 0 sep ahead 30 sep ahead 50 sep ahead From Table III we noice ha AR(p) model performs well a p 6 and sep ahead beween o 0, where he mean square error could be smaller han The resuls sugges ha AR(p) model can no predic much furher fuure CPU load of he node in a compuaional grid. Considering he predicion accuracy and compuaional expense, we use p=3 in our evaluaions below. B. The accuracy of he confidence inerval value predicion using hybrid model Unlike he poin value predicion using AR(p) model which esimaes a possible poin value of a fuure ime poin, confidence inerval value predicion gives a view of possible values in a fuure ime frame. If he rue CPU load measured by he monioring sysem in a fuure ime frame falls in he prediced confidence inerval value, we define he predicion accuracy of his siuaion as 00%, in oher words, he 39

6 predicion error is 0. The mean square error of he confidence inerval value predicion is defined as (5.), and Fig. 3 shows he evaluaion resuls. MSEin erval _ value _ predicion( ) = ( errorin erval _ value _ predicion( i) ) (5.) + i= 0 The errorin erval _ value _ predicion ( i) here is slighly differen from he error in he poin value predicion. We compue he errorin erval _ value _ predicion( i) using he definiion of (5.3). if X lower (i) <= X rue (i) <= X upper (i) errorin erval _ value _ predicion ( i) = 0 ; Else errorin erval_ value_ predicion( i) (5.3) = Min{ abs( X ( i) X ( i)), abs( X ( i) X ( i))} rue lower rue upper confidence inerval lengh of he predicion, which is ofen ignored in previous inerval predicion models. If he confidence inerval is oo wide, i may be meaningless for he ask scheduler. For example, he range of CPU load is described as 0% o 00%, bu he predicion load confidence inerval lengh of a fuure ime is as wide as 60%. This predicion is so fain ha he ask scheduler can hardly arrange he ask dispaching based on i. On he conrary, if his inerval lengh of predicion can be limied as small as 5% or 0%, i is much easier for he scheduler o recognize he fuure saus of he sysem CPU load. (a) (a) (b) (b) Fig. 3 Mean square error of hybrid model predicion C. The predicion confidence inerval lengh using hybrid model As shown in Fig. 3, our hybrid model is of an ousanding forecasing capabiliy. I illusraes ha we should use as small as possible he CIwindow o achieve small predicion mean square error. Bu here is anoher problem abou he Fig. 4 Mean predicing confidence inerval lengh Fig. 4 demonsraes he influence of he parameers, CIwindow and N, on he predicion inerval lengh. Noiceably, he mean confidence inerval lengh is almos below 0% when he CIwindow is bigger han 5. As a compromise beween he predicion accuracy and inerval lengh, we se he CIwindow as 0 when using our model o forecas he CPU load of he BioGrid. D. Comparison beween hybrid model and AR model We compared he predicion accuracy of our hybrid model and AR model on a se of 7 series of CPU load daa colleced from he BioGrid. The predicion mean square error of our 40

7 Fig. 5 Predicion errors compare on 7 series of CPU load daa colleced from he BioGrid hybrid model (using equaion (5.)) and AR model (using equaion (5.)) on his se of load daa are shown in Fig. 5. The experimenal resuls show ha he hybrid model ouperforms AR model on he CPU load predicion. The predicion mean square error of hybrid model is 77.07% less on average han ha of AR model. We also evaluaed he predicion mean square error of he inerval median predicion by using he hybrid model (using equaion (5.)). This variaion from our hybrid model also shows beer predicion accuracy ha he predicion mean square error is 5.3% less on average han ha of AR(3) model. In hese evaluaions, we se he parameer of he hybrid model as P=3, CIwindow =0. E. The power of Kalman filer in minimizing he measuremen errors In secion III, we analyse he causes of he measuremen errors. Measuremen error is unavoidable and may be amplified by predicion models, hus impairing he predicion accuracy. In our work, we use Kalman filer o minimize hese measuremen errors. We compare he mean square error of our hybrid model using Kalman filer wih he one wihou using Kalman filer. The resuls are shown in Fig. 6, which demonsraes ha Kalman filer works very well on minimizing he measuremen errors and he predicion accuracy is improved consequenly. BioGrid. Sample ime is from :7 o :08, oal 30 sample poin, wih sample inerval of 60 seconds, in Tsinghua Universiy node. Fig. 7(b), (c), (d) are he 0, 30 and 50 sep-ahead CPU load confidence inerval predicion by hybrid model, using he same parameer p=3 and CIwindow=0. Table IV summarizes he mean square error of hese predicions a differen n-sep-ahead and he mean inerval lengh. (a) Real CPU load (b) 0-sep-ahead predicion (c) 30-sep-ahead predicion Fig. 6 Mean square error amplificaion wihou using Kalman filer, where Z-label = MSE no _ kalman _ filer MSEu sin g _ kalman _ filer F. A sample of CPU load predicion using hybrid model Fig. 7 is a sample of he CPU load predicion using our hybrid model. Fig. 7(a) is he CPU load measuremen daa of he (d) 50-sep-ahead predicion Fig. 7 CPU load confidence inerval predicion using hybrid model 4

8 TABLE IV CONFIDENCE INTERVAL PREDICTION ERRORS AND INTERVAL LENGTH STATISTICS OF FIGURE 7 Evaluaion merics 0-sep -ahead 30-sep -ahead 50-sep -ahead Confidence inerval predicion MSE Mean inerval lengh of predicion (confidence 7.34% ±0.% 8.05% ±0.9% 6.7% ±0.8% level 95%) Fig. 7 and Table IV show ha our hybrid model works excellenly in CPU load predicion, which is always he boleneck for a compuaional grid. Using he parameer p=3 and CIwindow=0, he inerval lengh is limied o below 0% and he mean square error is smaller han 0.05 even for 50-sep-ahead (50 minues ahead) CPU load. VI. CONCLUSION AND FUTURE WORK To predic he load of a compuaional grid, we have developed a hybrid model which inegraes AR model wih he confidence inerval esimae. Whereas he poin value predicion is always he ideal esimae of he load in he fuure ime poin, he confidence inerval predicion adoped in his hybrid model can reflec he load variabiliy in a fuure ime frame and convey more informaion o he ask scheduler of a compuaional grid. In order o enhance he predicion accuracy, we use Kalman filer o minimize he load measuremen errors, and Savizky-Golay filer o smooh he hisory daa. The evaluaion resuls demonsrae ha hese noise eliminaing ools perform very well and lead o a significan improvemen on predicion accuracy. Considering he rade-off beween he predicion accuracy and he confidence inerval lengh, we use parameers p=3 and CIwindow = 0 o predic he CPU load of he BioGrid. The predicion advance ime can be even 50-sep-ahead long, wih significan less predicion mean square error han he convenional AR model and has accepable inerval lengh for he schedule algorihm. The opimal parameer values may be slighly differen according o he differen compuaional applicaions and programs running on a compuaional grid, and we will evaluae he predicion performance variabiliy of our hybrid model under differen applicaions and parameers in he fuure. Some machine learning mechanism and parameer auo-adapaion funcion could also be added in our model o fi in differen condiions. ACKNOWLEDGEMENT This Work is suppored by Naural Science Foundaion of China (904006, 9040, , 90606, ), Naional Key Basic Research Projec of China (004CB38000, 003CB37007), Naional High Technology Developmen Program of China (006AA0A08, 006AA0A, 006AA0A0), and IST programme of he European Commission (DG Informaion Sociey and Media, projec n 03444). The auhors acknowledge Ms. Jing Zhu of UCSD and Dr. Ying Zhao of Tsinghua Universiy for heir revision o his paper. REFERENCES [] I. Foser and C. Kesselman, The Grid: Blueprin for a New Compuing Infrasrucure, Morgan Kaufmann Publishers, San Francisco, CA, 999. [] C. Liu, L. Yang, I. Foser, and D. Angulo, Design and Evaluaion of a Resource Selecion Framework for Grid Applicaions, Proceedings of he h IEEE Inernaional Symposium on High-Performance Disribued Compuing (HPDC 00), Edinburgh, Scoland, 00. [3] D. Lu, H. Sheng, and P. Dinda, Size-based scheduling policies wih inaccurae scheduling informaion, h IEEE In l Symp, on Modeling, Analysis, and Simulaion of Compuer and Telecommunicaions Sysems (MASCOTS 004), pp. 3-38, 004. [4] S. Jang, X. Wu, and V. Taylor, Using performance predicion o allocae grid resources, Technical repor, GriPhyN 004-5, pp. -, 004. [5] R. Wolski. Dynamically Forecasing Nework Performance Using he Nework Weaher Service, Journal of Cluser Compuing, :9 3, January 998. [6] P. A. Dinda and D. R. O'Hallaron, Hos load predicion using linear models, Cluser Compuing 3, 4 (000). [7] P. A. Dinda and D. R. O'Hallaron, The Saisical Properies of Hos Load, Fourh Workshop on Languages, Compilers, and Run-ime Sysems for Scalable Compuers (LCR 998), Pisburgh, PA, 998. [8] Y. Zhang, W. Sun, and Y. Inoguchi, CPU Load Predicions on he Compuaional Grid, IEEE, Proc. in 6h Inernaional Conference on Cluser Compuing and he Grid (CCGrid 006), pp. 3-36, May 006. [9] Greg Welch, Gary Bishop, An Inroducion o he Kalman Filer, Universiy of Norh Carolina a Chapel Hill, Chapel Hill, NC, 995 [0] R. Wolski, N. Spring, and J. Hayes, Predicing he CPU availabiliy of Time-shared Unix Sysems, Proceedings of 8h IEEE High Performance Disribued Compuing Conference (HPDC 999), 999. [] J. Schopf and F. Berman, Performance predicion in producion environmens, in: h Inernaional Parallel Processing Symposium, Orlando, FL (April 998) pp [] Kalman, R. E A New Approach o Linear Filering and Predicion Problems, Transacion of he ASME Journal of Basic Engineering, pp (March 960). [3] Marple, S. L., Jr., Digial Specral Analysis wih Applicaions. Englewood Cliffs, NJ: Prenice-Hall, 987 [4] S. Akioka and Y. Muraoka, Exended forecas of CPU and nework load on compuaional grid, 004 IEEE In l Symp, on Cluser Compuing and he Grid (CCGrid 004), pp , 004. [5] J. Liang, K. Nahrsed, and Y. Zhou, Adapive Muli-Resource Predicion in Disribued Resource Sharing Environmen, 004 IEEE In l Symp, on Cluser Compuing and he Grid (CCGrid 004), pp. -8, 004. [6] L. Yang, I. Foser, and J.M. Schopf, Homeosaic and endency-based CPU load predicions, In l Parallel and Disribued Processing Symp. (IPDPS 003), pp. 4-50, 003. [7] Orfanidis, S.J., Inroducion o Signal Processing, Prenice-Hall, Englewood Cliffs, NJ, 996. [8] Y. Zhang, W. Sun, and Y. Inoguchi, Predicing Running Time of Grid Tasks based on CPU Load Predicions, 7h IEEE/ACM Inernaional Conference on Grid Compuing(Grid 006),Sep. 006 [9] (006) The ChinaGrid Super Vision websie. [Online]. Available: hp:// [0] (006) The ChinaGrid websie. [Online]. Available: hp://chinagrid.hus.edu.cn or hp:// 4

INTRODUCTION TO FORECASTING

INTRODUCTION TO FORECASTING INTRODUCTION TO FORECASTING INTRODUCTION: Wha is a forecas? Why do managers need o forecas? A forecas is an esimae of uncerain fuure evens (lierally, o "cas forward" by exrapolaing from pas and curren

More information

Measuring macroeconomic volatility Applications to export revenue data, 1970-2005

Measuring macroeconomic volatility Applications to export revenue data, 1970-2005 FONDATION POUR LES ETUDES ET RERS LE DEVELOPPEMENT INTERNATIONAL Measuring macroeconomic volailiy Applicaions o expor revenue daa, 1970-005 by Joël Cariolle Policy brief no. 47 March 01 The FERDI is a

More information

USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES

USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES Mehme Nuri GÖMLEKSİZ Absrac Using educaion echnology in classes helps eachers realize a beer and more effecive learning. In his sudy 150 English eachers were

More information

MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR

MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR The firs experimenal publicaion, which summarised pas and expeced fuure developmen of basic economic indicaors, was published by he Minisry

More information

The naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1

The naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1 Business Condiions & Forecasing Exponenial Smoohing LECTURE 2 MOVING AVERAGES AND EXPONENTIAL SMOOTHING OVERVIEW This lecure inroduces ime-series smoohing forecasing mehods. Various models are discussed,

More information

Chapter 8: Regression with Lagged Explanatory Variables

Chapter 8: Regression with Lagged Explanatory Variables Chaper 8: Regression wih Lagged Explanaory Variables Time series daa: Y for =1,..,T End goal: Regression model relaing a dependen variable o explanaory variables. Wih ime series new issues arise: 1. One

More information

SELF-EVALUATION FOR VIDEO TRACKING SYSTEMS

SELF-EVALUATION FOR VIDEO TRACKING SYSTEMS SELF-EVALUATION FOR VIDEO TRACKING SYSTEMS Hao Wu and Qinfen Zheng Cenre for Auomaion Research Dep. of Elecrical and Compuer Engineering Universiy of Maryland, College Park, MD-20742 {wh2003, qinfen}@cfar.umd.edu

More information

SPEC model selection algorithm for ARCH models: an options pricing evaluation framework

SPEC model selection algorithm for ARCH models: an options pricing evaluation framework Applied Financial Economics Leers, 2008, 4, 419 423 SEC model selecion algorihm for ARCH models: an opions pricing evaluaion framework Savros Degiannakis a, * and Evdokia Xekalaki a,b a Deparmen of Saisics,

More information

TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS

TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS RICHARD J. POVINELLI AND XIN FENG Deparmen of Elecrical and Compuer Engineering Marquee Universiy, P.O.

More information

Forecasting, Ordering and Stock- Holding for Erratic Demand

Forecasting, Ordering and Stock- Holding for Erratic Demand ISF 2002 23 rd o 26 h June 2002 Forecasing, Ordering and Sock- Holding for Erraic Demand Andrew Eaves Lancaser Universiy / Andalus Soluions Limied Inroducion Erraic and slow-moving demand Demand classificaion

More information

A New Type of Combination Forecasting Method Based on PLS

A New Type of Combination Forecasting Method Based on PLS American Journal of Operaions Research, 2012, 2, 408-416 hp://dx.doi.org/10.4236/ajor.2012.23049 Published Online Sepember 2012 (hp://www.scirp.org/journal/ajor) A New Type of Combinaion Forecasing Mehod

More information

DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS

DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS Hong Mao, Shanghai Second Polyechnic Universiy Krzyszof M. Osaszewski, Illinois Sae Universiy Youyu Zhang, Fudan Universiy ABSTRACT Liigaion, exper

More information

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613.

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613. Graduae School of Business Adminisraion Universiy of Virginia UVA-F-38 Duraion and Convexiy he price of a bond is a funcion of he promised paymens and he marke required rae of reurn. Since he promised

More information

Real-time Particle Filters

Real-time Particle Filters Real-ime Paricle Filers Cody Kwok Dieer Fox Marina Meilă Dep. of Compuer Science & Engineering, Dep. of Saisics Universiy of Washingon Seale, WA 9895 ckwok,fox @cs.washingon.edu, mmp@sa.washingon.edu Absrac

More information

Task is a schedulable entity, i.e., a thread

Task is a schedulable entity, i.e., a thread Real-Time Scheduling Sysem Model Task is a schedulable eniy, i.e., a hread Time consrains of periodic ask T: - s: saring poin - e: processing ime of T - d: deadline of T - p: period of T Periodic ask T

More information

Vector Autoregressions (VARs): Operational Perspectives

Vector Autoregressions (VARs): Operational Perspectives Vecor Auoregressions (VARs): Operaional Perspecives Primary Source: Sock, James H., and Mark W. Wason, Vecor Auoregressions, Journal of Economic Perspecives, Vol. 15 No. 4 (Fall 2001), 101-115. Macroeconomericians

More information

Automatic measurement and detection of GSM interferences

Automatic measurement and detection of GSM interferences Auomaic measuremen and deecion of GSM inerferences Poor speech qualiy and dropped calls in GSM neworks may be caused by inerferences as a resul of high raffic load. The radio nework analyzers from Rohde

More information

Journal Of Business & Economics Research September 2005 Volume 3, Number 9

Journal Of Business & Economics Research September 2005 Volume 3, Number 9 Opion Pricing And Mone Carlo Simulaions George M. Jabbour, (Email: jabbour@gwu.edu), George Washingon Universiy Yi-Kang Liu, (yikang@gwu.edu), George Washingon Universiy ABSTRACT The advanage of Mone Carlo

More information

Time-Expanded Sampling (TES) For Ensemble-based Data Assimilation Applied To Conventional And Satellite Observations

Time-Expanded Sampling (TES) For Ensemble-based Data Assimilation Applied To Conventional And Satellite Observations 27 h WAF/23 rd NWP, 29 June 3 July 2015, Chicago IL. 1 Time-Expanded Sampling (TES) For Ensemble-based Daa Assimilaion Applied To Convenional And Saellie Observaions Allen Zhao 1, Qin Xu 2, Yi Jin 1, Jusin

More information

DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR

DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR Invesmen Managemen and Financial Innovaions, Volume 4, Issue 3, 7 33 DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR Ahanasios

More information

Usefulness of the Forward Curve in Forecasting Oil Prices

Usefulness of the Forward Curve in Forecasting Oil Prices Usefulness of he Forward Curve in Forecasing Oil Prices Akira Yanagisawa Leader Energy Demand, Supply and Forecas Analysis Group The Energy Daa and Modelling Cener Summary When people analyse oil prices,

More information

How Useful are the Various Volatility Estimators for Improving GARCH-based Volatility Forecasts? Evidence from the Nasdaq-100 Stock Index

How Useful are the Various Volatility Estimators for Improving GARCH-based Volatility Forecasts? Evidence from the Nasdaq-100 Stock Index Inernaional Journal of Economics and Financial Issues Vol. 4, No. 3, 04, pp.65-656 ISSN: 46-438 www.econjournals.com How Useful are he Various Volailiy Esimaors for Improving GARCH-based Volailiy Forecass?

More information

Information Theoretic Evaluation of Change Prediction Models for Large-Scale Software

Information Theoretic Evaluation of Change Prediction Models for Large-Scale Software Informaion Theoreic Evaluaion of Change Predicion Models for Large-Scale Sofware Mina Askari School of Compuer Science Universiy of Waerloo Waerloo, Canada maskari@uwaerloo.ca Ric Hol School of Compuer

More information

Performance Center Overview. Performance Center Overview 1

Performance Center Overview. Performance Center Overview 1 Performance Cener Overview Performance Cener Overview 1 ODJFS Performance Cener ce Cener New Performance Cener Model Performance Cener Projec Meeings Performance Cener Execuive Meeings Performance Cener

More information

An Agent-based Bayesian Forecasting Model for Enhanced Network Security

An Agent-based Bayesian Forecasting Model for Enhanced Network Security An Agen-based Forecasing Model for Enhanced Nework Securiy J. PIKOULAS, W.J. BUCHANAN, Napier Universiy, Edinburgh, UK. M. MANNION, Glasgow Caledonian Universiy, Glasgow, UK. K. TRIANTAFYLLOPOULOS, Universiy

More information

Distributing Human Resources among Software Development Projects 1

Distributing Human Resources among Software Development Projects 1 Disribuing Human Resources among Sofware Developmen Proecs Macario Polo, María Dolores Maeos, Mario Piaini and rancisco Ruiz Summary This paper presens a mehod for esimaing he disribuion of human resources

More information

Market Liquidity and the Impacts of the Computerized Trading System: Evidence from the Stock Exchange of Thailand

Market Liquidity and the Impacts of the Computerized Trading System: Evidence from the Stock Exchange of Thailand 36 Invesmen Managemen and Financial Innovaions, 4/4 Marke Liquidiy and he Impacs of he Compuerized Trading Sysem: Evidence from he Sock Exchange of Thailand Sorasar Sukcharoensin 1, Pariyada Srisopisawa,

More information

Supply chain management of consumer goods based on linear forecasting models

Supply chain management of consumer goods based on linear forecasting models Supply chain managemen of consumer goods based on linear forecasing models Parícia Ramos (paricia.ramos@inescporo.p) INESC TEC, ISCAP, Insiuo Poliécnico do Poro Rua Dr. Robero Frias, 378 4200-465, Poro,

More information

Individual Health Insurance April 30, 2008 Pages 167-170

Individual Health Insurance April 30, 2008 Pages 167-170 Individual Healh Insurance April 30, 2008 Pages 167-170 We have received feedback ha his secion of he e is confusing because some of he defined noaion is inconsisen wih comparable life insurance reserve

More information

II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal

II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal Quarerly Repor on he Euro Area 3/202 II.. Deb reducion and fiscal mulipliers The deerioraion of public finances in he firs years of he crisis has led mos Member Saes o adop sizeable consolidaion packages.

More information

DDoS Attacks Detection Model and its Application

DDoS Attacks Detection Model and its Application DDoS Aacks Deecion Model and is Applicaion 1, MUHAI LI, 1 MING LI, XIUYING JIANG 1 School of Informaion Science & Technology Eas China Normal Universiy No. 500, Dong-Chuan Road, Shanghai 0041, PR. China

More information

Morningstar Investor Return

Morningstar Investor Return Morningsar Invesor Reurn Morningsar Mehodology Paper Augus 31, 2010 2010 Morningsar, Inc. All righs reserved. The informaion in his documen is he propery of Morningsar, Inc. Reproducion or ranscripion

More information

Multiprocessor Systems-on-Chips

Multiprocessor Systems-on-Chips Par of: Muliprocessor Sysems-on-Chips Edied by: Ahmed Amine Jerraya and Wayne Wolf Morgan Kaufmann Publishers, 2005 2 Modeling Shared Resources Conex swiching implies overhead. On a processing elemen,

More information

Predicting Stock Market Index Trading Signals Using Neural Networks

Predicting Stock Market Index Trading Signals Using Neural Networks Predicing Sock Marke Index Trading Using Neural Neworks C. D. Tilakarane, S. A. Morris, M. A. Mammadov, C. P. Hurs Cenre for Informaics and Applied Opimizaion School of Informaion Technology and Mahemaical

More information

The Application of Multi Shifts and Break Windows in Employees Scheduling

The Application of Multi Shifts and Break Windows in Employees Scheduling The Applicaion of Muli Shifs and Brea Windows in Employees Scheduling Evy Herowai Indusrial Engineering Deparmen, Universiy of Surabaya, Indonesia Absrac. One mehod for increasing company s performance

More information

The Transport Equation

The Transport Equation The Transpor Equaion Consider a fluid, flowing wih velociy, V, in a hin sraigh ube whose cross secion will be denoed by A. Suppose he fluid conains a conaminan whose concenraion a posiion a ime will be

More information

The Kinetics of the Stock Markets

The Kinetics of the Stock Markets Asia Pacific Managemen Review (00) 7(1), 1-4 The Kineics of he Sock Markes Hsinan Hsu * and Bin-Juin Lin ** (received July 001; revision received Ocober 001;acceped November 001) This paper applies he

More information

SEASONAL ADJUSTMENT. 1 Introduction. 2 Methodology. 3 X-11-ARIMA and X-12-ARIMA Methods

SEASONAL ADJUSTMENT. 1 Introduction. 2 Methodology. 3 X-11-ARIMA and X-12-ARIMA Methods SEASONAL ADJUSTMENT 1 Inroducion 2 Mehodology 2.1 Time Series and Is Componens 2.1.1 Seasonaliy 2.1.2 Trend-Cycle 2.1.3 Irregulariy 2.1.4 Trading Day and Fesival Effecs 3 X-11-ARIMA and X-12-ARIMA Mehods

More information

Hotel Room Demand Forecasting via Observed Reservation Information

Hotel Room Demand Forecasting via Observed Reservation Information Proceedings of he Asia Pacific Indusrial Engineering & Managemen Sysems Conference 0 V. Kachivichyanuul, H.T. Luong, and R. Piaaso Eds. Hoel Room Demand Forecasing via Observed Reservaion Informaion aragain

More information

A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation

A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation A Noe on Using he Svensson procedure o esimae he risk free rae in corporae valuaion By Sven Arnold, Alexander Lahmann and Bernhard Schwezler Ocober 2011 1. The risk free ineres rae in corporae valuaion

More information

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Profi Tes Modelling in Life Assurance Using Spreadshees PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Erik Alm Peer Millingon 2004 Profi Tes Modelling in Life Assurance Using Spreadshees

More information

Stock Trading with Recurrent Reinforcement Learning (RRL) CS229 Application Project Gabriel Molina, SUID 5055783

Stock Trading with Recurrent Reinforcement Learning (RRL) CS229 Application Project Gabriel Molina, SUID 5055783 Sock raing wih Recurren Reinforcemen Learning (RRL) CS9 Applicaion Projec Gabriel Molina, SUID 555783 I. INRODUCION One relaively new approach o financial raing is o use machine learning algorihms o preic

More information

Analogue and Digital Signal Processing. First Term Third Year CS Engineering By Dr Mukhtiar Ali Unar

Analogue and Digital Signal Processing. First Term Third Year CS Engineering By Dr Mukhtiar Ali Unar Analogue and Digial Signal Processing Firs Term Third Year CS Engineering By Dr Mukhiar Ali Unar Recommended Books Haykin S. and Van Veen B.; Signals and Sysems, John Wiley& Sons Inc. ISBN: 0-7-380-7 Ifeachor

More information

ANALYSIS FOR FINDING AN EFFICIENT SALES FORECASTING METHOD IN THE PROCESS OF PRODUCTION PLANNING, OPERATION AND OTHER AREAS OF DECISION MAKING

ANALYSIS FOR FINDING AN EFFICIENT SALES FORECASTING METHOD IN THE PROCESS OF PRODUCTION PLANNING, OPERATION AND OTHER AREAS OF DECISION MAKING Inernaional Journal of Mechanical and Producion Engineering Research and Developmen (IJMPERD ) Vol.1, Issue 2 Dec 2011 1-36 TJPRC Pv. Ld., ANALYSIS FOR FINDING AN EFFICIENT SALES FORECASTING METHOD IN

More information

Mathematics in Pharmacokinetics What and Why (A second attempt to make it clearer)

Mathematics in Pharmacokinetics What and Why (A second attempt to make it clearer) Mahemaics in Pharmacokineics Wha and Why (A second aemp o make i clearer) We have used equaions for concenraion () as a funcion of ime (). We will coninue o use hese equaions since he plasma concenraions

More information

Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya.

Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya. Principal componens of sock marke dynamics Mehodology and applicaions in brief o be updaed Andrei Bouzaev, bouzaev@ya.ru Why principal componens are needed Objecives undersand he evidence of more han one

More information

Why Did the Demand for Cash Decrease Recently in Korea?

Why Did the Demand for Cash Decrease Recently in Korea? Why Did he Demand for Cash Decrease Recenly in Korea? Byoung Hark Yoo Bank of Korea 26. 5 Absrac We explores why cash demand have decreased recenly in Korea. The raio of cash o consumpion fell o 4.7% in

More information

Real Time Bid Optimization with Smooth Budget Delivery in Online Advertising

Real Time Bid Optimization with Smooth Budget Delivery in Online Advertising Real Time Bid Opimizaion wih Smooh Budge Delivery in Online Adverising Kuang-Chih Lee Ali Jalali Ali Dasdan Turn Inc. 835 Main Sree, Redwood Ciy, CA 94063 {klee,ajalali,adasdan}@urn.com ABSTRACT Today,

More information

The Real Business Cycle paradigm. The RBC model emphasizes supply (technology) disturbances as the main source of

The Real Business Cycle paradigm. The RBC model emphasizes supply (technology) disturbances as the main source of Prof. Harris Dellas Advanced Macroeconomics Winer 2001/01 The Real Business Cycle paradigm The RBC model emphasizes supply (echnology) disurbances as he main source of macroeconomic flucuaions in a world

More information

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS R. Caballero, E. Cerdá, M. M. Muñoz and L. Rey () Deparmen of Applied Economics (Mahemaics), Universiy of Málaga,

More information

UPDATE OF QUARTERLY NATIONAL ACCOUNTS MANUAL: CONCEPTS, DATA SOURCES AND COMPILATION 1 CHAPTER 7. SEASONAL ADJUSTMENT 2

UPDATE OF QUARTERLY NATIONAL ACCOUNTS MANUAL: CONCEPTS, DATA SOURCES AND COMPILATION 1 CHAPTER 7. SEASONAL ADJUSTMENT 2 UPDATE OF QUARTERLY NATIONAL ACCOUNTS MANUAL: CONCEPTS, DATA SOURCES AND COMPILATION 1 CHAPTER 7. SEASONAL ADJUSTMENT 2 Table of Conens 1. Inroducion... 3 2. Main Principles of Seasonal Adjusmen... 6 3.

More information

Module 3 Design for Strength. Version 2 ME, IIT Kharagpur

Module 3 Design for Strength. Version 2 ME, IIT Kharagpur Module 3 Design for Srengh Lesson 2 Sress Concenraion Insrucional Objecives A he end of his lesson, he sudens should be able o undersand Sress concenraion and he facors responsible. Deerminaion of sress

More information

Maintaining Multi-Modality through Mixture Tracking

Maintaining Multi-Modality through Mixture Tracking Mainaining Muli-Modaliy hrough Mixure Tracking Jaco Vermaak, Arnaud Douce Cambridge Universiy Engineering Deparmen Cambridge, CB2 1PZ, UK Parick Pérez Microsof Research Cambridge, CB3 0FB, UK Absrac In

More information

Stock Price Prediction Using the ARIMA Model

Stock Price Prediction Using the ARIMA Model 2014 UKSim-AMSS 16h Inernaional Conference on Compuer Modelling and Simulaion Sock Price Predicion Using he ARIMA Model 1 Ayodele A. Adebiyi., 2 Aderemi O. Adewumi 1,2 School of Mahemaic, Saisics & Compuer

More information

Option Put-Call Parity Relations When the Underlying Security Pays Dividends

Option Put-Call Parity Relations When the Underlying Security Pays Dividends Inernaional Journal of Business and conomics, 26, Vol. 5, No. 3, 225-23 Opion Pu-all Pariy Relaions When he Underlying Securiy Pays Dividends Weiyu Guo Deparmen of Finance, Universiy of Nebraska Omaha,

More information

Chapter 8 Student Lecture Notes 8-1

Chapter 8 Student Lecture Notes 8-1 Chaper Suden Lecure Noes - Chaper Goals QM: Business Saisics Chaper Analyzing and Forecasing -Series Daa Afer compleing his chaper, you should be able o: Idenify he componens presen in a ime series Develop

More information

Appendix D Flexibility Factor/Margin of Choice Desktop Research

Appendix D Flexibility Factor/Margin of Choice Desktop Research Appendix D Flexibiliy Facor/Margin of Choice Deskop Research Cheshire Eas Council Cheshire Eas Employmen Land Review Conens D1 Flexibiliy Facor/Margin of Choice Deskop Research 2 Final Ocober 2012 \\GLOBAL.ARUP.COM\EUROPE\MANCHESTER\JOBS\200000\223489-00\4

More information

GoRA. For more information on genetics and on Rheumatoid Arthritis: Genetics of Rheumatoid Arthritis. Published work referred to in the results:

GoRA. For more information on genetics and on Rheumatoid Arthritis: Genetics of Rheumatoid Arthritis. Published work referred to in the results: For more informaion on geneics and on Rheumaoid Arhriis: Published work referred o in he resuls: The geneics revoluion and he assaul on rheumaoid arhriis. A review by Michael Seldin, Crisopher Amos, Ryk

More information

Forecasting. Including an Introduction to Forecasting using the SAP R/3 System

Forecasting. Including an Introduction to Forecasting using the SAP R/3 System Forecasing Including an Inroducion o Forecasing using he SAP R/3 Sysem by James D. Blocher Vincen A. Maber Ashok K. Soni Munirpallam A. Venkaaramanan Indiana Universiy Kelley School of Business February

More information

Making a Faster Cryptanalytic Time-Memory Trade-Off

Making a Faster Cryptanalytic Time-Memory Trade-Off Making a Faser Crypanalyic Time-Memory Trade-Off Philippe Oechslin Laboraoire de Securié e de Crypographie (LASEC) Ecole Polyechnique Fédérale de Lausanne Faculé I&C, 1015 Lausanne, Swizerland philippe.oechslin@epfl.ch

More information

A Probability Density Function for Google s stocks

A Probability Density Function for Google s stocks A Probabiliy Densiy Funcion for Google s socks V.Dorobanu Physics Deparmen, Poliehnica Universiy of Timisoara, Romania Absrac. I is an approach o inroduce he Fokker Planck equaion as an ineresing naural

More information

How To Calculate Price Elasiciy Per Capia Per Capi

How To Calculate Price Elasiciy Per Capia Per Capi Price elasiciy of demand for crude oil: esimaes for 23 counries John C.B. Cooper Absrac This paper uses a muliple regression model derived from an adapaion of Nerlove s parial adjusmen model o esimae boh

More information

COMPARISON OF AIR TRAVEL DEMAND FORECASTING METHODS

COMPARISON OF AIR TRAVEL DEMAND FORECASTING METHODS COMPARISON OF AIR RAVE DEMAND FORECASING MEHODS Ružica Škurla Babić, M.Sc. Ivan Grgurević, B.Eng. Universiy of Zagreb Faculy of ranspor and raffic Sciences Vukelićeva 4, HR- Zagreb, Croaia skurla@fpz.hr,

More information

Improvement in Forecasting Accuracy Using the Hybrid Model of ARFIMA and Feed Forward Neural Network

Improvement in Forecasting Accuracy Using the Hybrid Model of ARFIMA and Feed Forward Neural Network American Journal of Inelligen Sysems 2012, 2(2): 12-17 DOI: 10.5923/j.ajis.20120202.02 Improvemen in Forecasing Accuracy Using he Hybrid Model of ARFIMA and Feed Forward Neural Nework Cagdas Hakan Aladag

More information

Chapter 7. Response of First-Order RL and RC Circuits

Chapter 7. Response of First-Order RL and RC Circuits Chaper 7. esponse of Firs-Order L and C Circuis 7.1. The Naural esponse of an L Circui 7.2. The Naural esponse of an C Circui 7.3. The ep esponse of L and C Circuis 7.4. A General oluion for ep and Naural

More information

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1 Absrac number: 05-0407 Single-machine Scheduling wih Periodic Mainenance and boh Preempive and Non-preempive jobs in Remanufacuring Sysem Liu Biyu hen Weida (School of Economics and Managemen Souheas Universiy

More information

Diane K. Michelson, SAS Institute Inc, Cary, NC Annie Dudley Zangi, SAS Institute Inc, Cary, NC

Diane K. Michelson, SAS Institute Inc, Cary, NC Annie Dudley Zangi, SAS Institute Inc, Cary, NC ABSTRACT Paper DK-02 SPC Daa Visualizaion of Seasonal and Financial Daa Using JMP Diane K. Michelson, SAS Insiue Inc, Cary, NC Annie Dudley Zangi, SAS Insiue Inc, Cary, NC JMP Sofware offers many ypes

More information

Time Series Analysis Using SAS R Part I The Augmented Dickey-Fuller (ADF) Test

Time Series Analysis Using SAS R Part I The Augmented Dickey-Fuller (ADF) Test ABSTRACT Time Series Analysis Using SAS R Par I The Augmened Dickey-Fuller (ADF) Tes By Ismail E. Mohamed The purpose of his series of aricles is o discuss SAS programming echniques specifically designed

More information

Forecasting and Forecast Combination in Airline Revenue Management Applications

Forecasting and Forecast Combination in Airline Revenue Management Applications Forecasing and Forecas Combinaion in Airline Revenue Managemen Applicaions Chrisiane Lemke 1, Bogdan Gabrys 1 1 School of Design, Engineering & Compuing, Bournemouh Universiy, Unied Kingdom. E-mail: {clemke,

More information

The Grantor Retained Annuity Trust (GRAT)

The Grantor Retained Annuity Trust (GRAT) WEALTH ADVISORY Esae Planning Sraegies for closely-held, family businesses The Granor Reained Annuiy Trus (GRAT) An efficien wealh ransfer sraegy, paricularly in a low ineres rae environmen Family business

More information

PRACTICES AND ISSUES IN OPERATIONAL RISK MODELING UNDER BASEL II

PRACTICES AND ISSUES IN OPERATIONAL RISK MODELING UNDER BASEL II Lihuanian Mahemaical Journal, Vol. 51, No. 2, April, 2011, pp. 180 193 PRACTICES AND ISSUES IN OPERATIONAL RISK MODELING UNDER BASEL II Paul Embrechs and Marius Hofer 1 RiskLab, Deparmen of Mahemaics,

More information

How To Predict A Person'S Behavior

How To Predict A Person'S Behavior Informaion Theoreic Approaches for Predicive Models: Resuls and Analysis Monica Dinculescu Supervised by Doina Precup Absrac Learning he inernal represenaion of parially observable environmens has proven

More information

System Performance Improvement By Server Virtualization

System Performance Improvement By Server Virtualization Sysem Performance Improvemen By Server Virualizaion Hioshi Ueno, Tomohide Hasegawa, and Keiichi Yoshihama Absrac Wih he advance of semiconducor echnology, microprocessors become highly inegraed and herefore

More information

Using Weather Ensemble Predictions in Electricity Demand Forecasting

Using Weather Ensemble Predictions in Electricity Demand Forecasting Using Weaher Ensemble Predicions in Elecriciy Demand Forecasing James W. Taylor Saïd Business School Universiy of Oxford 59 George Sree Oxford OX1 2BE, UK Tel: +44 (0)1865 288678 Fax: +44 (0)1865 288651

More information

Nikkei Stock Average Volatility Index Real-time Version Index Guidebook

Nikkei Stock Average Volatility Index Real-time Version Index Guidebook Nikkei Sock Average Volailiy Index Real-ime Version Index Guidebook Nikkei Inc. Wih he modificaion of he mehodology of he Nikkei Sock Average Volailiy Index as Nikkei Inc. (Nikkei) sars calculaing and

More information

cooking trajectory boiling water B (t) microwave 0 2 4 6 8 101214161820 time t (mins)

cooking trajectory boiling water B (t) microwave 0 2 4 6 8 101214161820 time t (mins) Alligaor egg wih calculus We have a large alligaor egg jus ou of he fridge (1 ) which we need o hea o 9. Now here are wo accepable mehods for heaing alligaor eggs, one is o immerse hem in boiling waer

More information

A Re-examination of the Joint Mortality Functions

A Re-examination of the Joint Mortality Functions Norh merican cuarial Journal Volume 6, Number 1, p.166-170 (2002) Re-eaminaion of he Join Morali Funcions bsrac. Heekung Youn, rkad Shemakin, Edwin Herman Universi of S. Thomas, Sain Paul, MN, US Morali

More information

4 Convolution. Recommended Problems. x2[n] 1 2[n]

4 Convolution. Recommended Problems. x2[n] 1 2[n] 4 Convoluion Recommended Problems P4.1 This problem is a simple example of he use of superposiion. Suppose ha a discree-ime linear sysem has oupus y[n] for he given inpus x[n] as shown in Figure P4.1-1.

More information

The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas

The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas The Greek financial crisis: growing imbalances and sovereign spreads Heaher D. Gibson, Sephan G. Hall and George S. Tavlas The enry The enry of Greece ino he Eurozone in 2001 produced a dividend in he

More information

How To Write A Demand And Price Model For A Supply Chain

How To Write A Demand And Price Model For A Supply Chain Proc. Schl. ITE Tokai Univ. vol.3,no,,pp.37-4 Vol.,No.,,pp. - Paper Demand and Price Forecasing Models for Sraegic and Planning Decisions in a Supply Chain by Vichuda WATTANARAT *, Phounsakda PHIMPHAVONG

More information

Forecasting Sales: A Model and Some Evidence from the Retail Industry. Russell Lundholm Sarah McVay Taylor Randall

Forecasting Sales: A Model and Some Evidence from the Retail Industry. Russell Lundholm Sarah McVay Taylor Randall Forecasing Sales: A odel and Some Evidence from he eail Indusry ussell Lundholm Sarah cvay aylor andall Why forecas financial saemens? Seems obvious, bu wo common criicisms: Who cares, can we can look

More information

The Relationship between Stock Return Volatility and. Trading Volume: The case of The Philippines*

The Relationship between Stock Return Volatility and. Trading Volume: The case of The Philippines* The Relaionship beween Sock Reurn Volailiy and Trading Volume: The case of The Philippines* Manabu Asai Faculy of Economics Soka Universiy Angelo Unie Economics Deparmen De La Salle Universiy Manila May

More information

Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith**

Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith** Relaionships beween Sock Prices and Accouning Informaion: A Review of he Residual Income and Ohlson Models Sco Pirie* and Malcolm Smih** * Inernaional Graduae School of Managemen, Universiy of Souh Ausralia

More information

Research on Inventory Sharing and Pricing Strategy of Multichannel Retailer with Channel Preference in Internet Environment

Research on Inventory Sharing and Pricing Strategy of Multichannel Retailer with Channel Preference in Internet Environment Vol. 7, No. 6 (04), pp. 365-374 hp://dx.doi.org/0.457/ijhi.04.7.6.3 Research on Invenory Sharing and Pricing Sraegy of Mulichannel Reailer wih Channel Preference in Inerne Environmen Hanzong Li College

More information

Term Structure of Prices of Asian Options

Term Structure of Prices of Asian Options Term Srucure of Prices of Asian Opions Jirô Akahori, Tsuomu Mikami, Kenji Yasuomi and Teruo Yokoa Dep. of Mahemaical Sciences, Risumeikan Universiy 1-1-1 Nojihigashi, Kusasu, Shiga 525-8577, Japan E-mail:

More information

Towards Intrusion Detection in Wireless Sensor Networks

Towards Intrusion Detection in Wireless Sensor Networks Towards Inrusion Deecion in Wireless Sensor Neworks Kroniris Ioannis, Tassos Dimiriou and Felix C. Freiling Ahens Informaion Technology, 19002 Peania, Ahens, Greece Email: {ikro,dim}@ai.edu.gr Deparmen

More information

Contrarian insider trading and earnings management around seasoned equity offerings; SEOs

Contrarian insider trading and earnings management around seasoned equity offerings; SEOs Journal of Finance and Accounancy Conrarian insider rading and earnings managemen around seasoned equiy offerings; SEOs ABSTRACT Lorea Baryeh Towson Universiy This sudy aemps o resolve he differences in

More information

Model-Based Monitoring in Large-Scale Distributed Systems

Model-Based Monitoring in Large-Scale Distributed Systems Model-Based Monioring in Large-Scale Disribued Sysems Diploma Thesis Carsen Reimann Chemniz Universiy of Technology Faculy of Compuer Science Operaing Sysem Group Advisors: Prof. Dr. Winfried Kalfa Dr.

More information

ARCH 2013.1 Proceedings

ARCH 2013.1 Proceedings Aricle from: ARCH 213.1 Proceedings Augus 1-4, 212 Ghislain Leveille, Emmanuel Hamel A renewal model for medical malpracice Ghislain Léveillé École d acuaria Universié Laval, Québec, Canada 47h ARC Conference

More information

Forecasting and Information Sharing in Supply Chains Under Quasi-ARMA Demand

Forecasting and Information Sharing in Supply Chains Under Quasi-ARMA Demand Forecasing and Informaion Sharing in Supply Chains Under Quasi-ARMA Demand Avi Giloni, Clifford Hurvich, Sridhar Seshadri July 9, 2009 Absrac In his paper, we revisi he problem of demand propagaion in

More information

A New Schedule Estimation Technique for Construction Projects

A New Schedule Estimation Technique for Construction Projects A New Schedule Esimaion Technique for Consrucion Projecs Roger D. H. Warburon Deparmen of Adminisraive Sciences, Meropolian College Boson, MA 02215 hp://people.bu.edu/rwarb DOI 10.5592/omcj.2014.3.1 Research

More information

Cointegration: The Engle and Granger approach

Cointegration: The Engle and Granger approach Coinegraion: The Engle and Granger approach Inroducion Generally one would find mos of he economic variables o be non-saionary I(1) variables. Hence, any equilibrium heories ha involve hese variables require

More information

Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?

Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking? Supplemenary Appendix for Depression Babies: Do Macroeconomic Experiences Affec Risk-Taking? Ulrike Malmendier UC Berkeley and NBER Sefan Nagel Sanford Universiy and NBER Sepember 2009 A. Deails on SCF

More information

CHARGE AND DISCHARGE OF A CAPACITOR

CHARGE AND DISCHARGE OF A CAPACITOR REFERENCES RC Circuis: Elecrical Insrumens: Mos Inroducory Physics exs (e.g. A. Halliday and Resnick, Physics ; M. Sernheim and J. Kane, General Physics.) This Laboraory Manual: Commonly Used Insrumens:

More information

An Online Learning-based Framework for Tracking

An Online Learning-based Framework for Tracking An Online Learning-based Framework for Tracking Kamalika Chaudhuri Compuer Science and Engineering Universiy of California, San Diego La Jolla, CA 9293 Yoav Freund Compuer Science and Engineering Universiy

More information

Statistical Analysis with Little s Law. Supplementary Material: More on the Call Center Data. by Song-Hee Kim and Ward Whitt

Statistical Analysis with Little s Law. Supplementary Material: More on the Call Center Data. by Song-Hee Kim and Ward Whitt Saisical Analysis wih Lile s Law Supplemenary Maerial: More on he Call Cener Daa by Song-Hee Kim and Ward Whi Deparmen of Indusrial Engineering and Operaions Research Columbia Universiy, New York, NY 17-99

More information

A Natural Feature-Based 3D Object Tracking Method for Wearable Augmented Reality

A Natural Feature-Based 3D Object Tracking Method for Wearable Augmented Reality A Naural Feaure-Based 3D Objec Tracking Mehod for Wearable Augmened Realiy Takashi Okuma Columbia Universiy / AIST Email: okuma@cs.columbia.edu Takeshi Kuraa Universiy of Washingon / AIST Email: kuraa@ieee.org

More information

Q-SAC: Toward QoS Optimized Service Automatic Composition *

Q-SAC: Toward QoS Optimized Service Automatic Composition * Q-SAC: Toward QoS Opimized Service Auomaic Composiion * Hanhua Chen, Hai Jin, Xiaoming Ning, Zhipeng Lü Cluser and Grid Compuing Lab Huazhong Universiy of Science and Technology, Wuhan, 4374, China Email:

More information

FORECASTING NETWORK TRAFFIC: A COMPARISON OF NEURAL NETWORKS AND LINEAR MODELS

FORECASTING NETWORK TRAFFIC: A COMPARISON OF NEURAL NETWORKS AND LINEAR MODELS Session 2. Saisical Mehods and Their Applicaions Proceedings of he 9h Inernaional Conference Reliabiliy and Saisics in Transporaion and Communicaion (RelSa 09), 21 24 Ocober 2009, Riga, Lavia, p. 170-177.

More information

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C.

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Finance and Economics Discussion Series Divisions of Research & Saisics and Moneary Affairs Federal Reserve Board, Washingon, D.C. The Effecs of Unemploymen Benefis on Unemploymen and Labor Force Paricipaion:

More information