Home

User Guide

1. the Descriptive Statistics which provide information on the variables used to generate the model such as the variables types and categories the data set size the cross statistics the Performance Indicators which provide information on the performance of the model thanks to various indicators such as the forecasts error bars and efficiency the U2 the standard deviation the Cyclic Variables which provide you with an analysis of the seasonal and cyclic variables displayed as graphs CUSTOMER SAP Infinitelnsight 6 5 SP5 46 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie vV To Display the Statistical Reports 1 Onthe screen Using the Model click the option Statistical Reports The screen Model Reporting is displayed oily E ir Descriptive Statistics H E Variables oF Category Frequencies z E Continuous Variables z E Continuous Targets Number i Data Set Size i Cross Statistics with the Target s H E Performance Indicators zi Outliers E KTS Advanced Settings Model Reporting gE aSsua Vanable Missing Missing Role Values Count Values Weight Oinput Otarget skip Value Storage continuous date continuous number continuous integer Previous 3 Double click the desired report to display it in the right frame T Bl Descriptive Statistics H E Variables ori Category Frequende
2. Model Autosave J Enable Model Autosave Description Data Type Text Files Folder Samples Census J Browse File Table MyModel txt J Browse CUSTOMER SAP Infinitelnsight 6 5 SP5 24 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 Set the parameters listed in the following table Parameter Description Model This field allows you to associate a name with the model This name will then appear in the list of models to be Name offered when you open an existing model Description This field allows you to enter the information you want such as the name of the training data set used the polynomial degree or the KI and KR performance indicators obtained This information could be useful to you later for identifying your model Note This description will be used instead of the one entered in the panel Summary of Modeling Parameters Data Type this list allows you to select the type of storage in which you want to save your model The following options are available Text files to save the model in a text file Database to save the model in a database Flat Memory to save the model in the active memory SAS Files to save the model in a SAS compatible file for a specified version of SAS and a specified platform SAS v6 or 7 8 for Windows or UNIX SAS Transport to save the model in a generic SAS compati
3. SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series41 2014 SAP AG or an SAP affiliate company All rights reserved Al Signal vs Trend For this Scenario Time Series has recognized a descending linear trend it appears in blue on the plot Signal vs Trend ally View Signal Components ASHA View Graph Apply Settings Signal vs Trend 1958 1859 1960 196 1 1962 1963 1964 1965 1866 1967 1968 1969 18970 1871 18972 18973 1874 TIME R_ozone la kts_1Trend The detailed explanation of the trend can be found in Understanding the Model Debriefing gt Signal Components gt The Trend see The Trend on page 32 Signal vs Periodics For this scenario The model has not found any periodic A detailed explanation of the periodics can be found in section Understanding the Model Debriefing gt Signal Components gt The Cycles CUSTOMER SAP Infinitelnsight 6 5 SP5 42 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie Signal vs Fluctuation For this scenario Time Series has detected fluctuations that are represented in blue in the plot Signal Trend vs Fluctuation l Note You can see that the trend has been removed from the signal to allow better visualization of the fluctuations ally View Signal Components Asda View Graph Apply Settings Disp
4. 14 Maximum Forecast No Maximum Autosave Export KxShell Script Advanced The name of the model is filled automatically It corresponds to the name of the target variable R_ozone la for this scenario followed by the underscore sign _ and the name of the data source without its file extension R_ozone la in this case The field Number of Forecast s allows you to select the number of forecasts to generate The time unit used is determined by the data analyzed For example if the data set observations are recorded monthly the time unit will be one month See section Defining the Number of Forecasts on page 20 The maximum number of forecasts allowed is indicated in the field Maximum Forecast This number depends on the number of extra predictable variables available If there is no extra predictable variables the number of forecasts is unlimited The Autosave button allows you to activate the feature that will automatically save the model once it has been generated When the autosave option is activated a green check mark is displayed on the Autosave button p Autosave Defining the Number of Forecasts For this Scenario Define the number of forecasts to 24 that is two years CUSTOMER SAP Infinitelnsight 6 5 SP5 20 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie vV To Define the Forecasts Number
5. CUSTOMER End User Documentation Document Version 2 0 2014 04 25 1 2 2 1 Ze 2 3 2 4 25 Jd Ie 3 3 3 4 3 5 4 1 4 2 4 3 Welcome t this Guide iirinn aaa a aE 4 ADOOS DOCUMEN larran e T T RE 4 1 1 1 WhO shuld Read This DOCUMEN siontan ste tacenat decal t orien cst deocenth eyuec ate re sme oeetd teases 4 1 1 2 Prerequisites tor Use of this DOCUIMIOME sreresssnoeai iea E E E A A 4 1 1 3 Warts OC UIIMe eC OV Cl S soaa N A EA 5 1 1 4 FIOW TO WISE Ths DOCUMEN eena E E E A anek ae ane Seared 5 STOR Su SUNN apenas had ae aseunt then sue get allt cane sect ata toca ea dab bat uant aubatren a aaa 6 1 2 1 Files and Documentation Provided with this Guide s esssnnuussrrnnussrrrrrnrrrrrsrrrrrerrrrrrerrrrererrrree 6 General Introduction to SC MALiOS ssseccssnnsseeeeennsseeeeennsseceeennsseeeeonasseeeconsssseeeonssseeeeonssseceeonsssseesonsess 7 SCENO ets rt tel crcl eee cele card EEA ce taste EA 7 5 areca pct cn hate A aetna Saat cs aia ees ca OAS Andie aca teas ap ort cata ss 7 WAEFOGUICTION TO Ser DIe ISS act tes Attest sore Aaa ee a alts ta oes Hen ete aN aes DA a ite a te ee ale an ee a 8 Zail Additional Sample gl cis eae eee ti nena ert ee a eee Seemann eee 9 FileFormat SCC I Gal ONS va ieceto seca certo acpic sata entseueg an secc E camtese ama pet gaat eae acenauene gees aree ae nese 9 2 4 1 FOr Extra Predictable Variables 4a sci i san aa el hee casa ates Secunda eal eee eae al
6. TIME L The final model maximal horizon is 24 its complexity is 3 The final model Minimum Pearson P2 over the horizon is 0 627822 its cumulative mean square error ig 15 1116 Other performance indicators for the final model are L1 11 594659 and MAPE 4 78155 Fit performance on validation P2 0 704 L2 0 59117 Last forecast 24 performance on validation P2 0 631928 L2 0 659983 Chosen model is DetrendedAR Chosen trend is Linear TIME No Periodicity chosen vV To display the Progression Bar Click the Show Progression button The progression bar screen appears vV To Stop the Learning Process um 1 Click the Stop Learning Process button 2 Click Previous The screen Summary of Modeling Parameters appears 3 Go back to the section Checking Modeling Parameters SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series27 2014 SAP AG or an SAP affiliate company All rights reserved 27 3 3 2 Visualizing the Model Results At the end of the generation process asummary of the model results appears Training the Model amp FR AS Current Report All Reports Building Date 2014 03 10 13 31 10 Learning Time 4s Engine Name Kxen Time Series Author natacha yam Time Series First Date 1955 01 28 Time Series LastDate 1971 12 28 Time Series Horizon 24 Modeling Warnings Monotonic Variables Detected Yes Continuous Targets Numbe
7. insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert insert extral ReverseWorkingDaysIndices dummyPar_1l MondayMonthiInd dummyPar_2 TuesdayMonthInd dummyPar_3 WednesdayMonthInd dummyPar_4 ThursdayMonthind dummyPar_5 FridayMonthInd dummyPar_6 BeforeLastMonday dummyPar_ 7 LastMonday dummyPar_8 amp BeforeLastTuesday dummyPar_9 LastTuesday dummyPar_10 BeforeLastWednesday dummyPar_1l LastWednesday dummyPar_12 BeforeLastThursday dummyPar_13 LastThursday dummyPar_14 BeforeLastFriday dummyPar_15 LastFriday dummyPar_16 LastSwDaysiInd dummyPar_1 7 LastowDays dummyPar_18 Last4wDaysInd ummyPar_19 Last4wDays dummyPar_20 LastWMonth dummyPar_21 BeforeLastWMonth dummyPar_22 myKTS validateParameter Learning the model Once all the parameters have been set the learning phase is launched For this scenario learning the model model sendMode learn SAP Infinitelnsight 6 5 SP5 Scenario 2 Modeling with Extra Predictable Variables71 2014 SAP AG or an SAP affiliate company All rights reserved CUSTOMER 71 Saving the model Once a model has been generated you can save it Saving it preserves all the information that pertains to that model that is the modeling parameters its profit curves and so on Note that the directory in which the model is saved must exist For this scenario Saving the m
8. All rights reserved 59 vV To View the Corresponding Forecasts 1 On the menu Using a Model click the option View Forecasts 2 Click the Next button the screen View Forecasts appears ally View Forecasts As el ea View Graph Apply Settings Forecasts vs Signal Apr2001 May 2001 Jun 2001 Jul 2001 Aug 2001 Sep 2001 Cet2001 Now 20014 Dec 2001 Jan 2002 Feb 2002 Date E Cash Cash_Forecasts m Cash_Trend Outliers CUSTOMER SAP Infinitelnsight 6 5 SP5 60 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables 4 3 Modeling with Extra Predictable Inputs The following section will show how extra predictable variables can increase the performances on the current data set IN THIS CHAPTER Summary of the Modeling Settings to USC ccccccccseeseecceeeeeeeeeeeeceeeeeeaeeeeeceeeeesseeseeeeeeeessaeeeeeeeesssseaeeeeeeeesaaas 61 Selecting a Cutting Strategy and a Data Source 0 0 ceecccccccccccaesseeceeeeeeseeeseeeeeeeseeeeeeeeeeeeeessaeeeeeeeeessssaeeeeseeeeesaaas 61 Descnbing the Dala serinin EE Ea RREA A Rin EEAO an Enn 62 eee eR A dks 0 21 AEE E E E EE A E eee OEA SE TEE E EO OEE E T E TT 62 Defining the Forecasts NUMDCE icuceccssecctecccexcsernscrwescceiceseceevsces sedcuame dcteocebceececiesscaneceesucdsoedeuasatloceaenasdahadabescecsatecetuss 62 Viewing the Generated Forecasts cccccccssscccccececseeseseeccceeeseeeeeeecceeeeeaeeseeceeeeeeeeee
9. ReverseWorkingDaysIndices 4YS MondayMonthlnd Indices of the week days in the month An integer value TuesdayMonth nd WednesdayMonth nd ThursdayMonthlind FridayMonthInd Last5WDayslnd Indices of the 5 or 4 last working days of An integer value Last4WDaysInd the month CUSTOMER SAP Infinitelnsight 6 5 SP5 8 2014 SAP AG or an SAP affiliate company All rights reserved General Introduction to Scenarios The file KxDesc_CashFlows txt is the description file corresponding to the data file CashFlows txt 2 3 1 Additional Sample Files Additional sample files are provided to further test Time Series LaglAndCycles txt LaglAndCyclesAndWn txt TrendAndCyclic txt TrendAndCyclicAnd_4Wn txt TrendAndCyclicAndWn txt These files are located in the folder Samples KTS 2 4 File Format Specifications A training data file for Time Series must contain at least two columns The date column The signal column Three formats are supported for the Date column datetime ISO format yyyy mm dd hh mm which has an hour precision date ISO format yyyy mm dd which has a day precision number number for example seconds The signal column that is the target variable must be continuous An optional weight column can be used to tweak the modeling procedure Setting the weight of some rows to O allows ignoring these rows during the modeling process By default when a weight variable is not provided
10. all rows have a weight of 1 2 4 1For Extra Predictable Variables Future values of the Extra Predictable Inputs have to be filled in order to use them in a modeling session These variables have to be filled at least in the same range as the wanted forecasts The file will have the appearance described by the table below Time Series allows missing values in the extra predictable variables Line Index Date Monday Tuesday Wednesday Thursday Friday VWworkingDayindices Flows Ernitted 177 1996 09 22 176 1598 09 23 179 1598 09 24 180 1996 09 25 151 1598 09 26 152 1996 09 29 153 1598 09 30 154 1998 10 01 155 1996 10 02 156 1996 10 05 242 1920724 395 9563714 361 9100825 769 5131664 1679 640043 1124 200375 601 915403 aaa ono Aa A aa ooo a aA co SF OOOOH ooo Oooo HKH ooo SAP Infinitelnsight 6 5 SP5 CUSTOMER General Introduction to Scenarios9 2014 SAP AG or an SAP affiliate company All rights reserved 9 The last known signal value is at the line 183 with the corresponding date 1998 09 30 This line corresponds to the end of the training data set The figures present after this line are the future values of the extras predictable variables these figures are considered as predictive information Please note that the date variable has a special status This variable is not considered as an extra predictable input nevertheless it is possible to fill its future values If you fill this variable in
11. A E E 49 IOl ADDYS MEME iessen a a a aN CO MR eet ae ee 49 3952 SVINE TE MOGE loaner a a a a one ee come tem ae 53 595 OSS ct VOCS see ate cre ete se tet ach cote a a omen nono cena ane tt 54 Scenario 2 Modeling with Extra Predictable Variables ccsssssseccssnssseceeenssseceennssseeeeenssseeesens 56 RO EO eeeneeen rere et rt nee mer ene nt eae ner enceneet emt re iter htt tae eer rer hnet anrere Cheer Mee sen Miment nent a sehr et ereer erence Mey verteerreate 56 Anda MOEITE ariaa A Pheer nn Rs A nena SUE ASO ei A re Pe eon ere 57 4 2 1 Summary othe Modeline Settings to USE sarena ennaa N 57 4 2 2 Selecting a Data Source and a Cutting Strategy vicccnekshsenekis cide 57 4 2 3 DeSCrIDNE TIED TA aea aE E E 58 4 2 4 SIS CHINO GRASS a n rE AE EE AEE AEE AA E 58 4 2 5 Dennne tme FORECASTS NUMDO asss a ag uaetae anion ie shahenem 58 4 2 6 Viewimne the Generated POreCastS 5 eee E S AA 59 Modeling with Extra Fredictable MPU acesi A ee 61 4 3 1 Summary of the Modeling Settings to USC skh ich aude ivadbccdex avs duepotsladedte ce docile le antes onctnsealeran 61 4 3 2 Selecting a Cutting Strategy and a Data SOUrCe sa sti iid ate een ee ee 61 4 3 3 DesernDINETNE D oara a ee nee REO E E fe erm ee eee 62 CUSTOMER SAP Infinitelnsight 6 5 SP5 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide 4 4 SAP Infinitelnsight 6 5 SP5 Welcome to this Guideiii 4 3 4 SOLS CUS Vat OOS pectic eee 62 4 3 5
12. Defining the Forecasts NUMDEL ccccccccccseccceeeceeceeceeeaeeceeceeeceeeeecceeseeceeseeecesauaseeeaeeeessueeeesaeeeess 62 4 3 6 Viewing the Generated F Ore Casts sxieic cient iets eaters tendencies 63 Comparing the Forecasts With and Without Extra Predictable Variables cccccccccccecceseeeeeseeseesneeenenees 65 4 4 1 Understanding Forecasts without Extra Predictable Variables cccccccccccccseecceeeeeeeeaeeeeeeaneeees 65 4 4 2 Understanding Forecasts with Extra Predictable Varlables cccccccccceccceseecceeseeeceeeeeeeeeneeseeanees 66 CUSTOMER 2014 SAP AG or an SAP affiliate company All rights reserved lil 1 Welcome to this Guide IN THIS CHAPTER About this DOCUMEent ven ecsscincie cece sicscemdectandeidernaniueramindimnbddeicibecndn dine sdetesineiatadunsdhdciboma diet ellie pmtehinoaamantinnnn nedaabann deed Stiancdamcoenankecaneedewde Before BCCI sien siecencecmiesesensanceetondenkaees carheitdenasneeesdnd nbesneeasonesdiecebagnaeanocdesdonbereeaianeadieaeessaecesedsp qabesbaracesasecnoerbaecses 1 1 About this Document 1 1 1 Who Should Read this Document This document is addressed to people who want to evaluate or use SAP Infinitelnsight and in particular the Infinitelnsight Modeler Time Series feature 1 1 2 Prerequisites for Use of this Document Before reading this guide you should read chapters 2 and 3 of the Infinitelnsight User Guide that present respectively An introd
13. TIME R_ozone la m R_ozone la_Forecasts m R_ozone la_Trend Outliers CUSTOMER SAP Infinitelnsight 6 5 SP5 38 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie Understanding the Forecast Plot The Forecast plot allows you to visualize five types of information The signal The predicted signal The trend The error bars The outlier The following table describes each graphical element and its signification Element Symbolized by Represents Signal The green curve The information contained in the training data set Predicted The blue curve The signal predicted by the generated model Signal Trend The red curve The signal trend This curve is only displayed if the trend is a polynomial or a linear For more information on the trend see section The signal components on page 31 Error Bars The blue area around the The zone of possible error where the predicted signal could be The error bars end of the blue curve are only displayed for the forecasts Note The error bars are equal to twice the standard deviation computed on the Validation data set Outlier A red square A point where the predictive curve is very distant from the real curve Note An outlier is detected when the absolute value of the residuals is over twice the standard deviation computed on the Estimation data set As long as the original signal is
14. model has been Date and time in the format yyyy mm dd hh mm ss saved Commen Optional user defined comment Character string t that can be used to identify the model 5 Selecta model from the list 6 Click the Open button The Using the Model menu appears SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series55 2014 SAP AG or an SAP affiliate company All rights reserved 55 4 Scenario 2 Modeling with Extra Predictable Variables This section details _whatare the extra inputs variables for Time Series how to use these extra variables what is the impact of this feature on the modeling IN THIS CHAPTER PES AON aore E E noasemeacdencceceeentes neceuat E S 56 Denada Od eoe E E A S 57 Modeling with Extra Predictable INDUts cccccsssssccccceeeeeeecceeeseeeeeeeueceeeeeeueceeesseaeeeesseeueeeeeseaaeceesssageeesssaneeesenas 61 Comparing the Forecasts With and Without Extra Predictable Variables cccccccccccseceeseseceeeeeeeeseeeeeseeeeees 65 4 1 Presentation In Forecasting modeling extra variables are exogenous factors that may have an influence on the modeling These variables can be ordinal binary or continuous Time Series makes a distinction between two categories of extra variables the predictable and the unpredictable inputs The Predictable variables are variables which future values are known like the first Friday of the
15. month the first working day of the month and so on This type of variable can contain additional information which can be very useful for the trend and or the cyclic analysis The predictable variable is the subject of this section CUSTOMER SAP Infinitelnsight 6 5 SP5 56 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables 4 2 Standard Modeling IN THIS CHAPTER Summary of the Modeling Settings to USE 2 0 ceeccccccccceceesseeeeeeeeeeeeeeeceeeeeeeeeeseeceeeeeseeeeeeeeeeeessseeaeeeeeeesssaaaseeeeees 57 Selecting a Data Source and a Cutting Strategy cccccccesccccccceeceeeseeeeeeeseeeeeseeeeeeesseeseeeeeeeesseaeeeeeeeeessaaaseeeeeess 57 Describing the DAE wane cictcntnsineacoecnedendannsiecenedbaneninsleneiawashdendusbceetaxdnedeskwncnnenciedsnbantsiensexeandendocsecoeeanitanaachiceetexeendeedeatenes 58 SISTING Ve OS eea a EEE E E E E E cnet es uaevance 58 Defining the Forecasts NUMber saaannnenennnnesnnnnnnrrnrrnnnsrnrrrnsnrrresrnrrrnnnrrrnrrnrrrnnnnrrnsnnrrrnnntrrnrnrrrnnrnrrrennnrenn nn erenn 58 Viewing the Generated FOreCasts seacte A E E E EE 59 4 2 1Summary of the Modeling Settings to Use In this step you will follow the default scenario without using any extras predictable variables see section Standard Modeling with nfinite nsight Modeler Time Series on page 11 The table below summarizes the modeling settings that you must use It s
16. on page 15 4 2 4 Selecting Variables For this Scenario Keep Date as the time variable Keep Cash as the target variable Exclude all extra predictable variables Donot select a weight variable Check that the last training line is set at 251 On the screen displaying the signal Select the option Date in the Time list For detailed procedures refer to section 3 2 4 Selecting Variables on page 17 4 2 5 Defining the Forecasts Number For this Scenario Define the number of forecasts to 21 This number corresponds to the average number of days worked in one month For the detailed procedure refer to section Defining the Number of Forecasts on page 20 CUSTOMER SAP Infinitelnsight 6 5 SP5 58 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables 4 2 6 Viewing the Generated Forecasts For this Scenario The model obtained has the following form Model AR 37 polynom Date R 2014 03 10 14 11 54 4s Kxen TimeSenes Natacha yam 1955 01 28 1971 12 28 24 Modeling Warnings Monotonic Variables Detected Continuous Targets Number 1 33 7 54 3 662 129 Model Components Kxen TimeSeries Trend Cycles Fluctuations Model Performance KTS Model Performance Cancel Previous SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables59 2014 SAP AG or an SAP affiliate company
17. the variable you want to use as the time variable 2 Click the button gt located on the left of the Time field in section Required Variable upper right hand side The variable moves to the Time field Time l Note To remove the time variable select the variable in the Time field and click the button lt to move the variables back to the screen section Predictable Variables Kept SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series170 2014 SAP AG or an SAP affiliate company All rights reserved 17 vV To Select a Target Variable 1 Onthe screen Selecting Variables in the section Predictable Variables Kept left hand side select the variable you want to use as the Target Variable 2 Click the button gt located on the left of the Target field in section Required Variable upper right hand side The variable moves to the Target field Target lt R_ozoneta l Note To remove the target variable select the variable in the Target field and click the button lt to move the variables back to the screen section Predictable Variables Kept vV To Select a Weight Variable 1 Inthe field Predictable Variables Kept select the weight variable 2 Click the gt button located on the right of the Weight field Weight gt bs l Note To remove the weight variable select the Weight field and click the lt button vV To Exclude Variables 1 Inthe field Predictable
18. the P Print button A dialog box appears asking you to select the printer to use 2 Select the printer to use and set other print properties if need be 3 Click OK The screen content is printed Vv To Save the Screen Content 1 Click the L Save button A dialog box appears asking you to select the file properties 2 ype a name for your file 3 Select the destination folder Click OK If the screen displays A graph it is saved as a PNG image A report itis saved as an HTML file SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series11 2014 SAP AG or an SAP affiliate company All rights reserved Il vV To Copy the Screen Content Click the CA Copy button If the screen displays A report the application copies its HTML code You can paste it into a word processing program or into a spreadsheet program such as Excel and use it to generate your own graph A graph the application copies its parameters You can paste it into a spreadsheet program such as Excel and use it to generate your own graph vV To Display the Contextual Help 1 Click the help pe button at the bottom left of the screen The help screen appears 2 Click the Previous button to go back to the modeling screen Note that when there is no contextual help available for the current screen the help button is disabled CUSTOMER SAP Infinitelnsight 6 5 SP5 12 2014 SAP
19. the predicted range the Infinitelnsight Modeler Time Series engine will use the values for forecasting If you don t it will generate the future dates This is true whether extra predictable variables exist or not This feature can be very useful if you are not satisfied by the dates automatically generated by the nfinitelnsight Modeler Time Series engine The same file format has to be used for the training data set and the application data set l Note If you want to use your own dates instead of the automatic date generation please follow the same steps as the addition of extras predictable inputs 2 5 Infinitelnsight Modeling Assistant To accomplish the scenario you will use the Java based graphical interface of SAP Infinitelnsight Infinitelnsight modeling assistant allows you to select the SAP Infinitelnsight features with which you will work and help you at all stages of the modeling process VI To Start the Infinitelnsight Modeling Assistant 1 Select Start gt Programs gt SAP Business Intelligence gt SAP SAP Infinitelnsight gt SAP Infinitelnsight Infinitelnsight modeling assistant appears SAP SAP InfiniteInsight Version X Y Z Create a Clustering Model Create a Time Series Ana Create Association Rules Association Rules Time Series Analyze and Forecast Time Series Load a Model bo a ae Social Recommendation a Toolkit 2 Select the Create a Time Series Analysis feature CU
20. the variables in the regressions used in the model The variables and the target variable used in the regression depends on the component being modeled Two components can be modeled using a regression the trend and the fluctuation The following table details which generated variables and which target can be used for the regressions Modeled Component Possible Target Variable Variables Used for the Regression Trend signal functions of the date that is Time square Time sqrt Time and the extra predictable variables Fluctuation signal lag variables on the target signal trend signal trend cycles The following four types of plots allow you to visualize contributions by variables Variable Contributions Variable Weights Smart Variable Contributions Maximum Smart Variable Contributions The plot Presents Variable Contributions The relative importance of each variable in the built model Variable Weights The weights in the final polynomial of the normalized variables Smart Variable Contributions The variables internal contributions Maximum Smart Variable The variables internal contributions including only the maximum of Contributions similar variables For example only binned encoding of the continuous variable age will be displayed vV To Display the Variable Contributions 1 On the screen Using the Model click the option Regressions Contributions by Variables The screen Contributions by Vari
21. 14 SAP AG or an SAP affiliate company All rights reserved 5 1 2 Before Beginning 1 2 1 Files and Documentation Provided with this Guide Sample Data Files Both the evaluation version and the registered version of SAP Infinitelnsight are supplied with sample data files These files allow you to take your first steps using various features of SAP Infinitelnsight and evaluate them During installation of SAP Infinitelnsight the following sample files for nfinite insight Modeler Time Series are saved under the folder lt installation directory gt Samples KTS m R_ozone la txt m CashFlows txt m KxDesc_CashFlows txt To obtain a detailed description of these files see Introduction to Sample Files on page 8 Documentation Full Documentation Complete documentation is included with SAP Infinitelnsight This documentation covers The operational use of SAP Infinitelnsight features The architecture and integration of the SAP Infinitelnsight API The SAP Infinitelnsight Java graphical user interface Contextual Help Each screen in the Infinitelnsight modeling assistant is accompanied by contextual help that describes the options presented to you and the concepts required for their application vV To Display the Contextual Help 1 Click the Help button located on the screen lower left corner 2 Click the Previous button to go back to the original screen CUSTOMER SAP Infinitelnsight 6 5 SP5 6
22. 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide 2 General Introduction to Scenarios IN THIS CHAPTER BS IO eea aca eoeeuestednntheaeese seen eoeneedasaeaneainesatese E A 7 SCENO eax sign E E E poe eaeu ar eaioa sas veetondes useae seammeisaboesacvisptanoes E E onsesteeesne eitecersenesnc 7 Prodao OMFS SOS OS pcs access cette E T E 8 File Format SCC MC QUOI S a ecaciscectetes sess etre a a E 9 Infinitelnsight modeling ASSISTANL ccccceeeeeccceeeceeeseeseeeeceeecaeeeeeeeeeesesaeeeeeeeeeeeseeeeseeeeeeseseeeeeeeeesseaeaeaaeeeeeeessaaas 10 2 1 Scenario 1 This scenario demonstrates how to use the nfinite nsight Modeler Time Series feature for creating a standard model The data used in this scenario are monthly averages of hourly ozone O3 readings in downtown Los Angeles from 1955 to 1972 Ozone is a gas providing a protective shield against the ultraviolet radiation When found in the lower atmosphere it is a major component of smog Thus ozone rate is a common measure for smog intensity Los Angeles municipality took three measures in order to reduce this level and so decrease the smog downtown in 1960 the Golden State Freeway which sails round downtown opened inthe same year the rule 63 came into effect lowering the amount of allowable reactive hydrocarbons in gasoline in 1966 emission regulations for new car engines were introduced The purpose of this scenari
23. 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie vV To Display the Apply Results Graph 1 Select the tab Apply Setting The following panel is displayed ally View Forecasts EEE View Graph Apply Settings Application Data Set Data Type frextries J Browse Data Rozonedat O O O OOO F a Browse Forecast Parameters Number of Forecasts Used to Build the Model 24 Number of Forecasts for the Next Application 24 Select the data source type in the Data Type drop down list Text Files Data Base Flat Memory In the Folder field select the folder or data base where the apply data is located In the Data field select the file or table containing the apply data oF GW N In the section Forecasts Parameters set the number of forecasts to use for the application As an indication the number of forecasts used to build the model is displayed in a non modifiable field SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series37 2014 SAP AG or an SAP affiliate company All rights reserved 37 6 Click the Display button to visualize the graph resulting from the application of the model on the new data ally View Forecasts sid View Graph Apply Settings Forecasts vs Signal 1958 18959 1960 1961 1862 1963 1964 1965 1866 18967 1968 1969 1871 1872 1873 1874
24. 8 l 4 320 4 250 5 229 1959 02 20 1959 03 26 1959 04 25 Legend is used to compute gt are in the same vector of past values predicted value KTS_2is computed using date t Z known values for extra predictable variables at this date and past values of the signal However since the signal value at date tis unknown Time Series uses the last prediction that is KTS_1 to compute KTS_2 The following table shows how KTS_2 is computed using KTS_1 TIME Kxlndex R_ozone la kts 1 kts 2 kts 3 1959 01 20 1959 02 26 1959 05 20 1959 04 26 1959 02 26 1559 03 28 1559 04 28 Legend is used to compute gt are in the same vector of past values predicted value CUSTOMER SAP Infinitelnsight 6 5 SP5 52 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie KTS_3 is computed using date t Z2 known values for extra predictable variables at this date and past values of the signal However since the signal value at date tand t are unknown Time Series uses the last two predictions that is KTS_1 and KTS_2 to compute KTS_3 The following table shows how KTS_3 is computed using KTS_1 and KTS_2 TIME Kxlndex R_ozone la kts 1 kts 2 kts 3 1959 01 28 1959 02 28 1959 03 28 1959 04 28 4320 4 250 9 229 1959 02 28 1959 03 28 1959 04 28 Legend is used to compute gt are in the same vector of past values predicted value 3 5 2 Sa
25. AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 2 Step 1 Defining the Modeling Parameters IN THIS CHAPTER SS TING all OE Sato sede eit ccsnsceee caer ot cisscsasiceeisa sek oat cicates oheinemsauceesnes ek oawlusgulad ona se ueceoaied basset ecuaceaese eedes SEIECIING Cutting SU ALC OY 2iss ceca denise oteccdccadondactcacesdesdacteisacceeslondactoacaddeasianddatciaaddeumiendactasdaddeenseeddcateaddeendendestesie Describing the Data Selected Selec y 1k gt ene ane EE E E E ne ee eee Checking Modeling Parameters 3 2 1Selecting a Data Source For this scenario Use the file R_Lozone la txt as a training data set Vv To Select a Data Source 1 On the screen Select a Data Source select the option Data Type to select the data source format to be used 2 Click the Browse button The following dialog box appears Data Source Selection xj Select Source Folder for Data wl El gal Aga aff Samples o Eil Census Hil JapaneseData o E KelData E l KTC adk x ef Samples KTS Text Files dat data csv txt User Password Cancel 3 Double click the Samples folder then the KTS folder l Note Depending on your environment the samples folder may or may not appear directly at the root of the list of folders If you selected the default settings during the installation process you will find t
26. Data Source on page 13 and 3 2 2 Selecting a Cutting Strategy on page 14 4 3 3 Describing the Data For this Scenario Use the description file KxDesc_CashFlows txt For the detailed procedure refer to section 3 2 3 Describing the Data Selected on page 15 4 3 4 Selecting Variables The panel Selecting Variables allows you to select the time variable select the target variable select a weight variable optional set the last date to use for training the model select which variables should be kept for the modeling For this Scenario Keep Date as the time variable Keep Cash as the target variable Keepall the extra predictable variables Donot select a weight variable Check that the last training line is set at 251 On the screen Displaying the Signal Select the option Date in the Time list For the detailed procedures refer to section 3 2 4 Selecting Variables on page 17 4 3 5 Defining the Forecasts Number For this Scenario Define the number of forecasts to 21 This number corresponds to the average number of days worked in one month For the detailed procedure refer to section Defining the Number of Forecasts on page 20 CUSTOMER SAP Infinitelnsight 6 5 SP5 62 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables 4 3 6 Viewing the Generated Forecasts For this Scenario You get a model w
27. Length of Analyzed Cycles and the lagged variables created by changing the parameter Maximum Order of the Autoregressive Model Specific Parameters of the Model General Modify the Modeling Procedure Limitations on Variables Variable Selection a PE IA E PE PE S pao Percentage of Variable Contributions to Keep me J Activate for All Extrapredictable based Trends Maximum Order of the Autoregressive Model _ Activate for All Autoregressive Models J Ignore Outliers when Estimating the Trend Force Positive Forecasts C Cancel The Maximum Length of Analyzed Cycles controls the way Time Series analyzes the periodicities in the signal This is the length of the longest cycle Time Series will try to detect The default value is 450 It is also limited by the size of the estimation data set You can disable the cyclics analysis by setting this parameter to zero By reducing the default number of variables generated by Time Series you are able to reduce the computation time However it is strongly recommended to use the default settings otherwise the quality of modeling could be compromised The Maximum Order of the Autoregressive Model controls the way Time Series analyzes the random fluctuations in the signal This parameter defines the maximum dependency of the signal on its own past values You can set this parameter to zero to disable the fluctuations analysis Defining the Other Modeling Options The Ignore Outli
28. On the screen Summary of Modeling Parameters in the field Number of Forecast s enter the number of forecasts you want to obtain d Previous Generate Defining the Advanced Parameters The advanced parameters allow you to limit the number of analyzed variables define the modeling procedure vV To Define the Advanced Parameters 1 Click the Advanced button The panel Specific Parameters of the Model is displayed Specific Parameters of the Model General Modify the Modeling Procedure Limitations on Variables Variable Selection PENEN E E E VERE P AE O pao Percentage of Variable Contributions to Keep mE J Activate for All Extrapredictable based Trends Maximum Order of the Autoregressive Model 7 Activate for All Autoregressive Models J Ignore Outliers when Estimating the Trend Force Positive Forecasts 2 Setthe parameters as explained in the following sections 3 Click the OK button to save the new parameters The panel Summary of Modeling Parameters is displayed SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series21 2014 SAP AG or an SAP affiliate company All rights reserved 21 LL Defining the Number of Analyzed Variables optional Time Series automatically generates the variables that necessary to the modeling Among these itis possible to reduce the number of the cyclic variables by changing the parameter Maximum
29. SAP Infinitelnsight read the Introductory Guide to SAP Infinitelnsight 1 1 4 How to Use this Document Organization of this Document This document is subdivided into three chapters This chapter Welcome to this Guide serves as an introduction to the rest of the guide This is where you will find information pertaining to the reading of this guide and information that will allow you to contact us The Chapter 2 General Introduction to Scenario provides a summary to the nfinite nsight Modeler Time Series application scenario It also introduces the user interface and the data files used in this scenario The Chapter 3 Using the Infinitelnsight Modeler Time Series feature presents the nfinite nsight Modeler Time Serres feature This chapter is organized in two parts The first part presents the standard use of the nfinite nsight Modeler Time Series feature The second part presents the use of the nfinitelnsight Modeler Time Series with extra predictable inputs A summary and detailed table of contents located at the beginning of the guide and cross references throughout the document allow you to find the information that you need quickly and easily If you want more information on SAP Infinitelnsight and on the essential concepts of modeling data read the Infinitelnsight User Guide provided with SAP Infinitelnsight software SAP Infinitelnsight 6 5 SP5 CUSTOMER Welcome to this Guided 20
30. STOMER SAP Infinitelnsight 6 5 SP5 10 2014 SAP AG or an SAP affiliate company All rights reserved General Introduction to Scenarios 3 Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series Data modeling with nfinite nsight Modeler Time Seriesis subdivided into four broadly defined stages Defining the Modeling Parameters Generating and Validating the Model Analyzing and Understanding the Analytical Results Using a Generated Model A OND IN THIS CHAPTER Options of the Infinitelnsight modeling ASSISTANL ccccceeeeecceeeeeseceeeceeseceecceeseceeecaeseceeeeeeaeceeeseaseceeeseeeceeesaeaes 11 Step 1 Defining the Modeling Parameters ccccccssssccceceesscceeceeseceeeceesseeeecaeseceeeseeseeeeeseeseceeeseeseceeesaegeceeseaanes 13 Step 2 Generating the Model ccasesecsscast wenateseumpanccectnntaceauctendnadaseatnatdsenueanedeedbptesescnadasieadeandsansnebaeacboudnincracseroies 26 Step 3 Analyzing and Understanding the Generated MOdel cccccsssecccecesseceeeceeseceeecaeeeceeeseeeeceeeseeaeceeeeeanes 29 otep 4 Using the Model s wonsiosieansnnyantaasisnanetsmeantsoneasunparaaiinien AE a SARE AAE E EEEN EEE NE ENAN RN 49 3 1 Options of the Infinitelnsight modeling assistant On every screen of the Infinitelnsight modeling assistant one or more of the following options may be available in a toolbar located under the screen title vV To Print the Screen 1 Click
31. Sample Data View is displayed if Sample Data View i j X Date Selection Choose last training date Data Set Training tive roe or jomos l jonora mojom oeoa mna mas F 204 1971 12 28 1 21 204 First Row Index 144 Last Row Index 244 Refresh Current Selection tite romen or m a 2 Uses the First Row Index and Last Row Index to display the line containing the date you want to select as the last training date 3 Click the Refresh button to update the list of rows displayed 4 Click the row corresponding to the date you want to select The selected row is displayed in the section Current Selection at the bottom of the panel 5 Click the OK button to validate your selection The window closes and the Last Training Date information are updated in the main panel SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series19 2014 SAP AG or an SAP affiliate company All rights reserved 19 3 2 5 Checking Modeling Parameters The screen Summary of Modeling Parameters allows you to check the modeling parameters just before generating the model Summary of Modeling Parameters Model Name R_ozone ta_R_ozoneda Description Model Type fen TimeSeries Data to be Modeled Cutting Strategy Sequential withouttest DataDescription Nome Target Variable Rezne Weight Variable Optional None Number of Forecast s
32. The model performance CUSTOMER SAP Infinitelnsight 6 5 SP5 30 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie The Model Overview This section details the following information Name Significance For this scenario Model Name Name of the model R_ozone la_R_ozone la It is generated by using the target variable name and the data set name Data Set Name of the data source used for the model R_ozone la txt Initial Number of Input Total number of variables in the data set 3 Variables Number of Selected Variables Number of variables used to generate the model 1 Number of Records Number of observation in the data source file 204 Building Date Date and time when the model was build 2014 03 10 10 51 26 Learning Time Duration of the learning process 3s Engine Name Name of the component used to build the model Kxen TimeSeries The Targets Statistics For continuous targets Name Significance For this scenario lt target name gt name of the target variable for which the statistics are stated R_ozone la Min Minimum value found in the data set for the target variable L33 Max Maximum value found in the data set for the target variable 7 54 Mean Mean of the target variable 3 662 Standard Measure of the extent to which the target values are spread around their 1 29 Deviation Aera The Signal Components This section details the model compone
33. Variables Kept left hand side select the variables you want to exclude l Note To select all the variables of a field click inside the field and push the keys Ctrl and A at the same time 2 Click the gt button located near the top right corner of the field Excluded Variable Excduded Variables gt lt A bel D Alphabetic Sort S SAP Infinitelnsight 6 5 SP5 18 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie vV To Display the Signal 1 On the screen Selecting Variables click the button Plot Data located in the section Required Variables The screen Display Signal is displayed 2 Click the Time list and select the variable containing the time information l Note By default Time Series uses the first column of the data set as time variable 3 Click the Signal list and select the variable containing the signal information 4 Click the Previous button to go back to the screen Selecting Variables Reducing the Training Data Set The last date found in the data set is automatically selected as the last training date The second field which indicates the number of the line in the data set is automatically updated depending on the date you have selected Last Training Date 1971 12 28 line 204 Select Date vV To Reduce the Training Data Set 1 On the screen Selecting Variables click the button Select Date The panel
34. a set model openNewStore Kxen FileStore model newDataSet Training CashFlows txt model readSpaceDescription Training KxDesc_CashFlows txt set the line index of the end of training bind model DataSet Training myDataset myDataset getParameter myDataset changeParameter Parameters LastRow 251 myDataset validateParameter delete myDataset SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables69 2014 SAP AG or an SAP affiliate company All rights reserved 69 setting the time series transform parameter The Forecasting transform parameters are the following AutoFeedCount saves the number of forecasts asked by the user default 1 MaxCyclics indicates the maximum number of cyclicalities that will be analyzed by Time Series default 450 DateColumnName saves the name of the date variable required parameter ForecastsConnection gives the format of the forecasts in the output of Time Series default 1 If its value is 1 then the forecasts will be transposed at the end of the KTS_1 variable with the corresponding dates If its value is O then the forecasts stay in the last line of the file LastRowWithForecastingInformation saves the index of the last line of the file This parameter is required if you want to use extras predictable inputs PredictableExtras saves the names of the extras predictable variable _UnPredictableExtras saves the names of the extr
35. able is displayed 2 Select the type of plot you want to display in the drop down list Chart Type The plot Maximum Smart Variables is displayed by default l Note In case of a regression on one variable only the plot Maximum Smart Variables is not available Use the drop down list Chart Type to select another plot SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series45 2014 SAP AG or an SAP affiliate company All rights reserved A5 Contributions by Variables A Ss h A Chart Type Maximum Smart Variable Contributions Models Regression R_ozone ta TIME Maximum Smart Variable Contributions 0 00 0 05 0 10 0 15 0 20 0 25 0 30 0 35 0 40 0 45 0 50 0 55 0 60 0 65 0 70 0 75 0 50 0 65 0 80 0 85 1 00 in 2 m m gle Cancel 7 Previous 3 Select the regression you want to analyze in the Models drop down list Note that if there is only one regression in the model the Models drop down list is not displayed Understanding the Contributions by Variables In this scenario the model contains only the regression linear TIME which defines the trend 3 4 5 Statistical Reports To help you analyze your modeling results and to enable you to possibly share these results with your colleagues managers partners or customers Time Series provides you with a set of statistical reports in various formats There are three categories of reports
36. as unpredictable variables This parameter is not activated today ExtraMode gives the the format of the output of Time Series default No Extra No Extra value is the default format with the KTS_ variables Signal Components value is the format which includes with the previous cited variables each component of each variables trend cycles seasonality fluctuations Component Residues value is the format which includes with the previous format the residues after each variable component For this scenario setting KTS basic parameters bind model TransformInProtocol Default 0 myKTS myKTS getParameter myKTS changeParameter Parameters DateColumnName Date myKTS changeParameter Parameters AutoFeedCount 20 myKTS changeParameter Parameters LastRowWithForecastingInformation 271 myKTS validateParameter CUSTOMER SAP Infinitelnsight 6 5 SP5 70 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables Setting the PredictableExtras parameter All the variables except Date and FlowsEmitted are added to the PredictableExtras parameter For this scenario setting KTS PredictableExtras parameter myKTS bindParameter Parameters PredictableExtras extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral extral delete insert
37. ata Set Size o B Cross Statistics with the Target s H 4 Performance Indicators El Cydic Variables i fi Extra predictable Variables Analys r Outiers E KTS Advanced Settings Extra predictable Variables Analysis 2 500 6 000 7 500 10 000 12 500 E i iw pal p ba a a lt iw m i m i a E PE T ba m Em E E D jable PeriodicExtrasPred Manda E Smoothed Target Mean M Target Mean Understanding Cyclic Details This screen presents the cyclic PeriodicExtrasPred_MondayMonthInd as shown in the list Extra Variable The numbers 1 2 3 4 and 5 represent the index of the Mondays in a month This plot shows a pick on the index 3 that is on the third Monday of the month CUSTOMER SAP Infinitelnsight 6 5 SP5 74 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables www sap com contactsap 2014 SAP AG or an SAP affiliate company All rights reserved No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP AG The information contained herein may be changed without prior notice Some software products marketed by SAP AG and its distributors contain proprietary software components of other software vendors National product specifications may vary These materials are provided by SAP AG and its affiliated companies SAP G
38. ble file Folder Depending upon which option you selected this field allows you to specify the ODBC source the memory store or the folder in which you want to save the model File Table This field allows you to enter the name of the file or table that is to contain the model The name of the file must contain one of the following format extensions txt text file in which the data is separated by tabs or csv text file in which the data is separated by commas 4 Click OK SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series25 2014 SAP AG or an SAP affiliate company All rights reserved 25 3 3 Step 2 Generating the Model Once the modeling parameters are defined you can generate the model vV To Generate the Model On the screen Summary of Modeling Parameters click the Generate button The screen Training the Model appears The model is being generated A progress bar allows you to follow the process 1 Training the Model a 2 BSG Computing statistics Stop Current Task 2 fthe Autosave option has been activated in the panel Summary of Modeling Parameters a warning message is displayed at the end of the learning process confirming that the model has been saved fe SAP InfiniteInsight Messages X A Warning Wew The model R_ozoneta_R_ozone a has been saved 3 10 14 10 48 08 AM 3 Click Close IN THIS CHAPTER Following
39. ceeesaeseceeesaaeeceessuaeeeessaaeees 45 Statistical FREDO S 2s lt c ccc cctacesestcorecostutcnecsieaccemssnauesunacasas tote aainsctoncaasnatineadeniadanchotntanawnsneesseeabsannaclaeaaseateaesendensoneenaanid 46 SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series29 2014 SAP AG or an SAP affiliate company All rights reserved 29 3 4 1Model Debriefing vV To Display the Model Debriefing 1 On the Using the Model menu select the Model Overview option The screen Model Overview appears oil Model Overview As a Report Type Model Overview Sunray are LE T a L E E T Learning Time 4s Engine Name Kxen TimeSeres Author natacha yam Time Series First Date 1955 01 28 Time Series Last Date 1971 12 28 Time Series Horizon 24 Monotonic Variables Detected Yes Continuous Targets Number Min 1 33 Max T M Mean 3 662 Standard Deviation 1 29 Model Components Trend Linear TIME Cycles Fluctuations AR 37 Model Performance KTS Model Performance Horizon wide MAPE 0 178 ile Cancel 7 Previous l Note If you have built more than one model in the same session all model debriefing will be displayed on this screen sorted by Date of Build Understanding the Model Debriefing The Model Debriefing screen is composed of four sections detailing information on The overview of the model The targets statistics The model components
40. del select the option View Signal Components 2 Inthe list Display Options select the option Signal Trend vs Periodics The following screen appears ally View Signal Components sda View Graph Apply Settings Display Options Signal Trend vs Periodics y we i Il i i A i Pal ye WN WN ai WA wN i Y y VI yN WYN V Jan 2001 Feb 2001 Mar2001 Apr2001 May 2001 Jun 2001 Jul 2001 Aug 2001 Sep 2001 Oct2001 Nov2001 Dec 2001 Jan 2002 Feb 2002 Date ekts_1ResiduesTrend w kts_1PeriodicExtrasPred_MondayMonthind i B SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables67 2014 SAP AG or an SAP affiliate company All rights reserved 67 Modeling using KxShell scripts This section details the KxShell script corresponding to Scenario 2 IN THIS CHAPTER Creating the time series MOE ccccccccccccseeeeceeeeeceeeeseeeceeeeseaeeeseceeeeeeseeesseeeeeeeseaesseeeeeeeesaueaeeeeeeeessaaaaeeeeeeensaaas 68 Setting EOS a ANS ES cerere ir Ra EE RERE R NRE FEE ERSE rA RORA NAER 68 Opening the Tamning Data Set rsen isnie ae EERE ceadsaupaaennitedsomderaidonsaliaucdiocweanmanceieane 69 Setting the time series transform ParaMetel cccccccccssecccceeececeeseeeceeeceeseeeeeseeeeecseeceeseaecesseeeeessaeeeesaueeesseeeesaaes 70 Setting the PredictableExtras parameter swissswinsinnsdenanancsnenniesnseinamsnnenrsiinvdrentocdnensin niednemadiinasiidwsasvatisnoadirtnnnsudeniteninsn
41. displayed you can measure the accuracy of the predicted signal against the original one When the original signal stops the error bars allow you to measure the level of confidence of the predicted signal The error bars are not displayed further than the first forecast for which the model cannot guarantee the accuracy of the predicted signal In other words when the model cannot evaluate if its prediction is correct it stops displaying the error bars l Note If you have selected the Sequential cutting strategy the error bars are displayed on the test part of the Signal SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series39 2014 SAP AG or an SAP affiliate company All rights reserved 39 Options vV To Zoom the Plot In Out 1 Right click the plot area you want to zoom in or out A contextual menu appears Properties Save as Print Zoom Out i kr wA r Sa e Both Axes Domain Axis Auto Range Range Axis 2 Select the type of zoom you want to apply 3 Select on which axes you want to zoom Note that the point where you click is the central point of the Zoom v To Display the Value of a Specific Element of a Signal Place your cursor on a selected point of a signal curve A pop up displays the information for this point of the signal 3 4 3 Signal Components v To Display the Signal Components On the menu Using the Model click the option V
42. e file description is displayed Wi Guessed Description US SES ee ee ee ee TIM oe ntinuous D e a E w SiKxdndex integer continuous o do o O o Automaa A 0 Add Filter in Data Set Grane G openoesopton G sveoesoson 2 Validate the description storage type and value 3 Click the Next button CUSTOMER SAP Infinitelnsight 6 5 SP5 16 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 2 4 Selecting Variables The panel Selecting Variables allows you to select the time variable 1 select the target variable 2 select a weight variable optional 3 set the last date to use for training the model 4 select which variables should be kept for the modeling 5 and which should be excluded 6 Selecting Variables Predictable Variables Kept Required Variables 5 lt a Target Fee 2 Plot Data Weight gt 3 Excluded Variables gt KxIndex lt 6 E A sO Alphabetic Sort 4 Last Traning vate paaa ine BOAT Selecta E bel FP Alphabetic Sort Ho es For this Scenario Keep TIME as the time variable Keep R_ozone la as the target variable Donot select a weight variable Keep the last training date selected by default vV To Select a Time Variable 1 On the screen Selecting Variables in the section Predictable Variables Kept left hand side select
43. ed one step forward This is the basic forecast where the predicted observation equals the latest signal observation 2 the signal moved two step forward For this scenario The trend found in the signal is a Linear TIME The following graph shows the trend compared with the signal io ANE AA Pre pv VA yn am F oo am co _ Cd ca tt Lr oo F oo T co _ Lr p Lr p Lr oo iam am a am oo iam oo iam oo F F a a a a an am a a an a a a an a a an a ao ao ao ao ao ao a a a o o ao ao ao ao ao ao Le We Le Le We Le Le Le Le Le Le Le Le Le Le Le LL CUSTOMER SAP Infinitelnsight 6 5 SP5 32 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie The Cycles The cycles are periodic elements that can be found at least twice in the Estimation data set The two types of cycles are detailed in the following table Type of cycle Description Periodic a cycle not depending on the date A periodic is defined by the number of time units it covers Seasonal a cycle depending on the date or time For example the day of the month the hour of the day and so on Note Time Series can automatically detect the following seasonal variables dayOfWeek dayOfMonth dayOfYear weekOfYear monthOfYear hourOfDay For this scenario No periodic has been detected for this scenario The following graph presents an example of two periodics found in
44. erface 5 Inthe section Results Generated by the Model select the file format for the output file Text files Database Flat Files 6 Click the Browse button to select In the Folder field the folder in which the result file will be saved In the Data field the name of the result file 7 Click the Apply button 8 Click the Next button The screen Using the Model appears Once application of the model has been completed the results files of the application is automatically saved in the location that you had defined from the screen Applying the Model Application Data Set Requirements The data set used to apply a Time Series model is generally the same used for training the model Applying a Time Series model produces a similar output data set with extra columns and or rows containing the requested forecasts It is also possible to apply a model to a different data set provided that the following conditions are fulfilled the application data set must contain the target variable all the input variables from the training data set that is all the variables that have not been excluded during the variables selection step all the key variables from the training data set except for the key variables automatically generated by SAP Infinitelnsight such as Kx Index the date column must be sorted strictly increasing Order Level 1 for ODBC sources the first date of the application data set must be present
45. erized by 2 data items These data or variables are described in the following table Variable Description Example of Values Time Month and year of the readings A date in the format yyyy mm dd such as 1953 Uls2s R_ozone la Average of the hourly readings for A numerical value with two decimals the month The file CashFlows txt is the sample data file used to follow Scenario 2 of the Time Series feature and use the extras predictable inputs This file presents daily measures of cash flows from January 2 1998 to September 30 1998 Each observation is characterized by 25 data items The data or variables are described in the following table Variable Description Example of values Date Day month and year of the readings A date in the format yyyy mm dd such as 1998 01 02 Cash Cash flow A numerical value with n decimals BeforeLastMonday Boolean variables that indicates if the 1 if the information is true LastMonday information is true or false BeforeLast Tuesday LastTuesday BeforeLastWednesaay Ll astWednesday BeforeLast Thursday LastThursday BeforeLastFriday LastFriday Last5WDays Last4WDays Boolean variables that indicate if the date 1 if the information is true is in the 5 or 4 last working days of the month LastWMonth Boolean variables that indicates if the 1 if the information Is true Beforel astWMonth information is true or false WorkingDayslIndices Indices or reverse indices of the working An integer value
46. ers when Estimating the Trend checkbox uses a strategy for reducing the effect of outliers when estimating the regressions in deterministic trends This leads to an improvement in trend estimation The Force Positive Forecasts checkbox allows users to force Time Series to generate a positive model with positive forecasts only CUSTOMER SAP Infinitelnsight 6 5 SP5 22 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie Defining the Variables This parameter groups some controls for the variable selection feature When a variable selection is used an automatic selection process is performed on trends or AR models during the competition and the result is kept only if it improves the final model The Percentage of Variable Contributions to Keep is the percentage of contributions that are kept in the automatic selection process The default value is 95 The Activate for All Extra predictable Based Trends option performs a variable selection on all extrapredictable based trends User variables are kept only if they have sufficient contributions in the trend regression The checkbox is enabled by default The Activate for All Autoregressive Models option performs an automatic variable selection on the past values of the signal for all autoregressive models This leads to a more parsimonious AR model that is asimpler model and a lower order The box is no
47. es Si eee ane 9 animritelasisnit Modeling ASSIS ta ies anni at ihe a te lt gil aa a ol tat te ial ae oh Nee 10 Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series sccssssssseeeees 11 Options of the Infinitelnsight Modeling ASSISTANT eee cc cece eeececeec eee eceeeceeeeeeeeeaeeeeeeeeaaeeeeeeeaaeeeeeenaneeeeeaaas 11 Step 1 Defining the Modeling Rar ainieterS inci ees es hs ae ia uo hee aa ee oe 13 324 Selene a Dwa O E a N oats an Looe a esi t oe fake 13 S22 SCI CUMS aC SU aC OY ae th r eee alt ial clon OAA 14 3 2 3 DESCHIDING TE Data o Clete saa e A 15 3 2 4 SelecUNe Val AOS Sres E EA O 17 3 2 9 Checking Modeling Paramete Sciieornerie r inian EEN E EA OEA ESEE eaten ang ees 20 Step 2 Generating the Model smri aa eia E EEE A AE ARAE ae e Era Ea AAE Ba EE AANE 26 3 3 1 FOUOWIRG the Generating Froce SSe e EA E EE E RA 27 532 VISUANIZIN GUMS WIOGE IG SUIUS merisier a aa dita seatiunbis a aeaa E AA 28 Step 3 Analyzing and Understanding the Generated MOdel cccccccccccecseceeeeceeeeeceeeueeeeeeesaaeeeeeeenaneeeees 29 3 4 1 Model NC ONS TUM ahs a a a aa Err ATE O 30 SAL Foreca ocenenia a a ae a a E ee haces 35 3 4 3 Sinal OX 0 818 1k tS ee a ee a E 40 3 4 4 Regressions Contribution by Variables ssssnnusssrruusrrrrurrrrrisrrrrterrrrrterrrrrterrrrrrrrrrrrerrrrrrerrrn 45 3 4 5 a Cal e DOr re N E Tn eee TN na 46 SCD ASI thie Mode eese sntaigacdes lt tacinns E R A se batenpaesia
48. etection Seasonal variable detection Extra predictable usage as Periodics Autoregressive modeling CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series23 2014 SAP AG or an SAP affiliate company All rights reserved 23 vV To Modify the Modeling Procedure 1 Inthe section Modify the Modeling Procedure check the desired option Specific Parameters of the Model General Modify the Modeling Procedure Default Only Based on Extra predictable Variables Disable the Polynomial Trends Customized Trends Lagi Lag Second Order Differencing Linear in Time Polynomial in Time Linear in Extrapredictables Linear in Time and Linear in Extrapredictables Polynomial in Time and Linear in Extrapredictables Periodic Extrapredictables m Fluctuations Autoregressive Previous 2 Ifyou have selected the Customized option uncheck the types of models you want to disable 3 Click the OK button The panel Summary of Modeling Parameters is displayed Activating the Autosave Option The panel Model Autosave allows you to activate the option that will automatically save the model at the end of the generation process and to set the parameters needed when saving the model vV To Activate the Autosave Option 1 Inthe panel Summary of Modeling Parameters click the Autosave button The panel Model Autosave is displayed 2 Check the option Enable Model Autosave
49. he Samples folder located in C Program Elles SAP Infinicernsighrc iniiniceInsignt Yz 727 SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series13 2014 SAP AG or an SAP affiliate company All rights reserved 13 4 Select the file R_ozone la txt then click OK The name of the file appears in the Data Set field 3 2 2 Selecting a Cutting Strategy To generate a Time Series model you must select a cutting strategy to cut your training data set into the three sub sets estimation validation and test Since the order of the observations in the data set is important for the modeling only two types of cutting strategies are available in Time Series Sequential with Test Sequential without Test For more information on Cutting Strategies see the Infinitelnsight User Guide For this Scenario Do not change the cutting strategy by default vV To Select a Cutting Strategy 1 On the screen Select a Data Source click the button Cutting Strategy The panel Cutting Strategy is displayed 2 Inthe Predefined list select the cutting strategy you want to use Cutting Strategy f Predefined Sequential without test 3 Click the OK button 4 Back to the panel Select a Data Source click the Next button 5 The screen Data Description appears Data Description Add Filter in Data Set spp Analyze LD open Description l Save Description Q View Data D
50. he picks amplitude is not correctly forecasted the model appears noisy The solution to refine this model is to add extra predictable variables 4 4 2 Understanding Forecasts with Extra Predictable Variables The following screen displays the forecasts generated by Time Series when using extra predictable variables ally View Forecasts A Ss el en View Graph Apply Settings Forecasts vs Signal tt L P yN ay Sts ETTET ETTET Jan 2001 Feb 2001 Mar2001 Apr2001 May2001 Jun 2001 Jul 2001 Aug2001 Sep 2001 Oct2001 Nov2001 Dec 2001 Jan 2002 Date E Cash w Cash_Forecasts m Cash_Trend Outliers G In this scenario the addition of extra predictable variables has improved the trend detection and therefore the model quality The three points that needed to be refined in the previous section are improved the error bars are reduced especially for the forecasted pick the picks amplitude is correctly forecasted the noise has been almost completely attenuated The extra predictable variable selected by the nfinite lnsight Modeler Time Series engine to refine the model is the monday in month index It is found in the model definition Cyclic PeriodicExtrasPred_MondayMonthInd polynom Date R CUSTOMER SAP Infinitelnsight 6 5 SP5 66 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables vV To Display the Periodics 1 On the Using the Mo
51. he values of a target variable saved for later use IN THIS CHAPTER PAD VIN ME MOGE asri erena EE A E E EE E EESE E R 49 NG ME Node e r esac seacecete esseeeate 53 Opening a Model 3 5 1 Applying the Model vV To Apply the Model 1 On the screen Using the Model click Apply Model The screen Applying the Model appears gt Applying the Model Application Data Set Data Type TextFiles v a fee Samples KTS H Browse Data R_ozone ta txt F a El Browse Define Mapping Generation Options Generate Forecasts with their Components and Residues Results Generated by the Model Data Type Text Files Data P H Browse Define Mapping Number of Forecasts Used to Build the Model 24 Number of Forecasts for the Next Application 24 C Use Direct Apply in the Database nos SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series49 2014 SAP AG or an SAP affiliate company All rights reserved 49 a 2 Inthe section Application Data Set the information concerning the data set is already filled See the section Application Data Set Requirements on page 50 for further information 3 Inthe section Generation Options select the type of output you want see section Type of Results Available see Understanding the Applying Mode on page 51 4 Check the option View Generated Outputs to also display the apply results in the int
52. hich have the following form Model Polynom Date Cyclic PeriodicExtrasPred_MondayMonthInd R Training the Model a 28 Hh Current Report All Reports Modeling Warnings Continuous Targets Number Model Components Keen Time Series Model Performance KTS Model Performance Cancel SAP Infinitelnsight 6 5 SP5 Min Max Mean Standard Deviation Trend Cycles Fluctuations 2014 03 10 11 36 29 TS Kxen Time Series natacha yam 2001 01 02 12 00 00 2001 12 28 12 00 00 7 1 579 57 24 659 2 4 930 13 3 471 31 Polynom Date PeniodicExtrasPred_MondayMonthind Previous CUSTOMER Scenario 2 Modeling with Extra Predictable Variables63 2014 SAP AG or an SAP affiliate company All rights reserved 63 vV To View the Corresponding Forecasts 1 Onthe menu Using a Model click the option View Forecasts 2 Click the Next Button the screen View Forecasts appears ally View Forecasts ETE View Graph Apply Settings Forecasts vs Signal PEE Se LULU AN hia Jan 2001 Feb 2001 Mar2001 Apr2001 May2001 Jun 2001 Jul 2001 Aug2001 Sep 2001 Oct2001 Nov2001 Dec 2001 Jan 2002 Date E Cash m Cash_Forecasts w Cash_Trend Outliers Fac CUSTOMER SAP Infinitelnsight 6 5 SP5 64 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables 4 4 Comparing the Forecasts With and Without Extra Predictable Variables To better u
53. hould be sufficient enough for users who are already familiar with the KJWizard For detailed procedures and more information see the following sections Replace the options given in the scenario by the following ones Task s Screen Settings e Specifying the data source Data to be Modeled e In the field Data Set select the file CashFlows txt e Selecting a cutting strategy Cutting strategy sequential without test Specifying a description file Data Description In the field Description select the file KxDesc_CashFlows txt Defining the extra predictable Selecting Variables e Select and exclude all the variables from the field inputs number Predictable Variables Kept Defining the Forecasts Number Summary of Modeling In the field Number of Forecasts enter 20 Parameters 4 2 2 Selecting a Data Source and a Cutting Strategy For this Scenario Use the file CashFlows txt as the training data set Select the cutting strategy Sequential without test For the detailed procedures refer to sections 3 2 1 Selecting a Data Source on page 13 and 3 2 2 Selecting a Cutting Strategy on page 14 SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables57 2014 SAP AG or an SAP affiliate company All rights reserved 57 4 2 3 Describing the Data For this Scenario Use the description file KxDesc_CashFlows txt For the detailed procedure refer to section 3 2 3 Describing the Data Selected
54. ht User Guide A Comment about Database Keys For data and performance management purposes the data set to be analyzed must contain a variable that serves as a key variable Two cases should be considered If the initial data set does not contain a key variable a variable index Kx ndex is automatically generated by nfinitelnsight Modeler Time Series This will correspond to the row number of the processed data If the file contains one or more key variables they are not recognized automatically You must specify them manually in the data description See the procedure To Specify that a Variable is a Key On the other hand if your data is stored in a database the key will be automatically recognized SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series15 2014 SAP AG or an SAP affiliate company All rights reserved 15 eee nies vV To Specify that a Variable is a Key 1 Inthe Key column click the box corresponding to the row of the key variable 2 ype in the value 1 to define this as a key variable oO Guessed Description For this Scenario Select Text Files as the file type Use the Analyze function to describe the R_ozone la txt data file Set the TIME variable as the key Set the Kx ndex variable Key to O Set the TIME variable Order to 1 vV To Analyze a Data File 1 Onthe screen Data Description click the Analyze button Th
55. iew Signal Components 2 Apop up is displayed asking you to confirm or update the name and location of the training data set file 3 Update the information if you have renamed or moved the training data set file or if its type has been changed l Note Steps 2 and 3 are required in case you open a saved model and the data set used to train it has been moved for example CUSTOMER SAP Infinitelnsight 6 5 SP5 40 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 4 Click OK The screen View Signal Components appears aly View Signal Components ASA View Graph Apply Settings Display Options Signal vs Forecasts Signal vs Forecasts 1958 1859 1960 1861 1962 1963 1964 1965 1866 1967 1968 1969 1870 1871 18972 18973 1874 TIME E R_ozone la kts_1 Outliers iG 5 Inthe list Display Options select the component you want to display Understanding the Signal Components The signal components are ordered as listed below the trend the periodics the fluctuation A N a the residuals 5 the outliers To display a component Time Series removes the previous existing components from the signal For example to display the fluctuation Time Series removes the trend and the periodics from the signal More information on the signal components available in Understanding the Model Debriefing Signal Components on page 31
56. in the time window defining the training data set For example for the ozone model a data set ozone without the first 10 rows is a valid application data set while a data set starting with the date value 1973 03 01 is not since this date is not contained in the training data set which ends on 1971 12 28 CUSTOMER SAP Infinitelnsight 6 5 SP5 50 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie Type of Results Available In the Generate pull down menu you can choose to generate three types of results If you choose the option The result file will contain Predicted Values Only a NeienIeS the predicted variables that is the forecasts for every date of the training data set Forecasts with their aMnputVaTaNIES Components the predicted variables that is the forecasts for every date of the training data set the components value trend cycles fluctuation for each forecast Forecasts with their a put varians Components and Residues the predicted variables that is the forecasts for every date of the training data set the components value trend cycles fluctuation for each forecast the remaining values residue obtained after extracting each component from each forecast Only First Forecasts Column all input variables and their Error Bars the first predicted variable that is the first forecast for e
57. iontnrctednrenecocctn se GES GES 6 Goto the section Describing the Data Selected CUSTOMER SAP Infinitelnsight 6 5 SP5 14 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 2 3 Describing the Data Selected To describe your data you can Either use an existing description file that is taken from your information system or saved from a previous use of SAP Infinitelnsight features Or create a description file using the Analyze option available to you in SAP Infinitelnsight modeling assistant In this case it is important that you validate the description file obtained You can save this file for later re use A Caution The description file obtained using the Analyze option results from the analysis of the first 100 lines of the initial data file In order to avoid all bias we encourage you to mix up your data set before performing this analysis Why Describe the Data Selected In order for SAP Infinitelnsight features to interpret and analyze your data the data must be described To put it another way the description file must specify the nature of each variable determining their Storage format number number integer integer character string string date and time datetime or date date Type continuous nominal ordinal or textual For more information about data description see the Infinitelnsig
58. lay Options Signal Trend vs Fluctuations Signal Trend vs Fluctuations 1958 18959 1960 1964 1862 1963 1964 1965 1866 1967 1968 1969 1870 1871 18972 1873 1874 TIME ekts_1ResiduesTrend m kts_14AR A detailed explanation of the fluctuation can be found in section Understanding the Model Debriefing gt Signal Components gt The Fluctuation see The Fluctuation on page 34 SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series43 2014 SAP AG or an SAP affiliate company All rights reserved 43 Signal vs Residuals For this Scenario The residuals found by the model appear in blue on the plot Signal vs Final Residuals ally View Signal Components sda View Graph Apply Settings Display Options Signal vs Final Residuals 1958 1859 1960 1861 1962 1963 1964 1965 1866 18967 1968 18969 1870 1871 18972 18973 1874 TIME ER _ozone la m kts_1ResiduesAR A detailed explanation of the residuals can be found in section Understanding the Model Debriefing gt Signal Components gt The Residuals see The Residuals on page 34 CUSTOMER SAP Infinitelnsight 6 5 SP5 44 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie KO 3 4 4 Regressions Contribution by Variables The Contributions by Variables plot allows you to examine the relative significance of each of
59. led The Outliers The outliers are points where the predictive curve is very distant from the real curve They are represented by a red square on the plot An outlier is detected when the absolute value of the residuals is over twice the standard deviation computed on the Estimation data set CUSTOMER SAP Infinitelnsight 6 5 SP5 34 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie The Model Performance This section gives the model performance Name Significance For this Scenario Horizon Wide MA PE This quality indicator for the forecasting model is the mean of MAPE values 0 178 observed over all the training horizon A value of zero indicates a perfect model while values above 1 indicate bad quality models A value of 0 09 means that the model takes into account 91 of the signal or in other words the forecasting error model residues is relatively of 9 MAPE Mean Absolute Percentage Error The MAPE value is the average of the sum of the absolute values of the percentage errors It measures the accuracy of the model s forecasts and indicates how much the forecasts differ from the real Signal value 3 4 2 Forecasts vV To Display the Forecast Plot 1 On the Using the Model Menu click the View Forecasts option 2 A pop up is displayed asking you to confirm or update the name and location of the training data set file 3 Update the infor
60. lt use open the ApplyiIn data set model newDataSet ApplyIn CashFlows txt bind model DataSet ApplyIn myDataset myDataset getParameter myDataset changeParameter Parameters LastRow 251 myDataset validateParameter delete myDataset open the ApplyOut data set model newDataSet ApplyOut out_CashFlows txt f fapply the model model sendMode apply open the ApplyOut data set model openNewStore Kxen FileStore Saved model newDataSet ApplyOut out_CashFlows txt fapply the model model sendMode apply apply with a different horizon 6 bind model TransformInProtocol Default 0 myKTS myKTS getParameter myKTS changeParameter Parameters AutoFeedCountApplied 6 myKTS validateParameter model sendMode apply SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables73 2014 SAP AG or an SAP affiliate company All rights reserved 73 vV To Display Cyclic Details 1 On the panel Using the Model select the option Statistical Reports 2 Onthe left menu select the item Cyclic Variables gt Extra Predictable Variables Analysis The following screen appears ally Model Reporting YF gaja AGASHBOR E i Descriptive Statistics E Variables Engine kxen TimeSeries ai amp t ee Extra predictable Variable PeriodicExtrasPred_MondayMonthind e gt E Continuous Targets Number Target Minus Cash Regression Cash Date SquareTime SquareRootTime be E D
61. mation if you have renamed or moved the training data set file or if its type has been changed a Note Steps 2 and 3 are required in case you open a saved model and the data set used to train it has been moved for example SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series35 2014 SAP AG or an SAP affiliate company All rights reserved 39 4 Click OK The screen View Forecasts appears ally View Forecasts As el en View Graph Apply Settings Forecasts vs Signal 1965 1866 18967 1968 196 1970 1871 18972 1858 1850 1860 19861 1862 1863 19864 TIME R_ozone la m R_ozone la_Forecasts m R_ozone la_Trend Outliers l Note When you copy and paste this graph the confidence interval information are not made available To get this information you can go to Tables Debriefing gt Error Bars The interval lower bound equals the signal value minus two times the value of the error bar and the upper bound equals the signal value plus two times the value of the error bar Displaying the Apply Plot The panel Apply Settings allows you to graphically preview the results of an apply of your model on a new data set l Note Be aware that this feature does not generate any output files apart from the graph To generate an output file see Step 4 Applying the Model see Applying the Model on page 49 CUSTOMER SAP Infinitelnsight 6 5 SP5 36
62. nderstand how extra predictable variables can improve a model this section will analyze and compare the forecasts obtained with both a standard modeling and a modeling with extra predictable variables IN THIS CHAPTER Understanding Forecasts without Extra Predictable Variables ccccccccccccceseceeeeeeeeceeeceeeeeeeeseeeeeseeeesseeeeesees 65 Understanding Forecasts with Extra Predictable Variables ccccccceccccceseceeseeceeeeeseeeseeeceeeesecesseeeeseeeeeesseeeeesaes 66 4 4 1Understanding Forecasts without Extra Predictable Variables The following screen displays the forecasts generated by Time Series when the extra predictable variables have been excluded ally View Forecasts sda View Graph Apply Settings Forecasts vs Signal 1 ia WV Aa AN les ald A M jV 4 N Apr2001 May 2001 Jun 2001 Jul 2001 Aug 2001 Sep 2001 Cet2001 Now 20014 Dec 2001 Jan 2002 Feb 2002 Date E Cash Cash_Forecasts m Cash_Trend Outliers SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables65 2014 SAP AG or an SAP affiliate company All rights reserved 65 In this model the engine uses its own variables cyclics trend fluctuations to generate the more predictive model possible The trend and the picks position are correctly detected but the following points could be improved the error bars are very extended meaning that the confidence of the model is low t
63. nts that is the components of the polynomial used to generated the forecasts A model is a combination of at least one of the three following types of elements one trend one or more cycles one fluctuation SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series31 2014 SAP AG or an SAP affiliate company All rights reserved 31 e The Trend The trend is the general orientation of the signal The four types of trends are detailed in the following table Type of trend Can be displayed as Polynom Time a curve corresponding to the detected polynomial A polynomial is modeled using Classification Regression The available functions for the polynomial are Time square Time sqrt Time where Time equals the value of the DateCo umnName parameter Polynom Time a curve corresponding to the detected polynomial ExtraPredictable This is the same function using in addition the extra predictable variables as inputs Linear Time a straight line It is a special case of Polynom Time Linear Time a straight line ExtraPredictables This is the same function using in addition the extra predictable variables as inputs Polynom ExtraPredictables a curve corresponding to the detected polynomial This function could be very next to a cyclic representation because Classification Regression is only using the extra predictable variables as inputs L the signal mov
64. o is to confirm the decreasing trend of the ozone rate by predicting the next 18 months and describing the different signal elements based on the ozone rate 2 2 Scenario 2 This scenario demonstrates how to use the nfinite nsight Modeler Time Series feature to create a model with extra predictable inputs In this scenario you are an executive of a financial entity that manages cash flows Your role is to make sure that credits are available with the correct amount at the correct date to provide the best management possible of your financial flows Time Series provides you with two methods for reaching your objective creating a standard model creating a model with extra predictable variables SAP Infinitelnsight 6 5 SP5 CUSTOMER General Introduction to Scenarios 2014 SAP AG or an SAP affiliate company All rights reserved 7 a 2 3 Introduction to Sample Files SAP Infinitelnsight is provided with sample data files allowing you to evaluate the nfinite nsight Modeler Time Series feature and take your first steps in using it The file R_ozone la txt is the sample data file that you will use to follow Scenario 1 of the Time Series feature It is an excerpt from the book Time Series Analysis Forecasting and Control G E P Box and G M Jenkins Third Edition Prentice Hall 1994 This file presents monthly averages of hourly ozone O3 readings in downtown Los Angeles from 1955 to 1972 Each observation is charact
65. odel in the current directory model saveModel DefaultBankFlows_Model txt With Extras Predictable Inputs and 20 forecasts Opening an Existing Model Once saved models may be opened and reused in SAP Infinitelnsight For this scenario createStore Kxen FileStore myRestoreStore myRestorestore openstore setDefaultUserPassword myRestoreStore restoreLastModelD model CUSTOMER SAP Infinitelnsight 6 5 SP5 72 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables Applying the model A model generated by Time Series can be applied ONLY to data sets for which the first date of the time variable is located between the first date and the last date of the time variable of the training data set By default a model generated by Time Series is applied to the training data set To apply the model you have to open a data set containing the data to use the Applyln data set You have to open as well a data set that will contain the output of the apply session the ApplyOut data set As the training data set specifications you have to set the end of the training section i e the end of the known values of the signal By default the model is applied with the same horizon as the horizon used for training The user can however apply with a different horizon by setting the parameter AutoFeedCountApplied For this scenario Applying the model on the training data set defau
66. r R_ozone la Min 1 33 Max T 54 Mean 3 662 Standard Deviation 1 29 Model Components Trend Linear TIME Cycles Fluctuations AR 37 Model Performance KTS Model Performance Horizon wide MAPE 0 178 If you have built more than one model in the same session all model debriefing will be displayed on this screen sorted by Date of Build For more information on the model summary go to section Understanding the Model Debriefing A Caution In some cases the message No Model Found is displayed instead of the signal information It means that none of the models found can predict accurately the signal evolution CUSTOMER SAP Infinitelnsight 6 5 SP5 28 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 4 Step 3 Analyzing and Understanding the Generated Model The suite of plotting tools within the SAP Infinitelnsight allows you to analyze and understand the model generated he performance of the model The forecasts generated by the model IN THIS CHAPTER Model Debriefing eee eee eee ees ieee ie ee mee eee ee eee ee ee ene eae ee ee eee eee eee ee ee ere 30 FO A e E E er secre oie EN arc eros cc ats sce N EAS E EENS A E EN ee 35 Reh ON SNS areca tee see Spaeth sein tse aetna Sco T E A A E SE E E A IAE E T 40 Regressions Contribution by Variables ccccccccsssccceccesseceeceeeseceeeeeeseceeseeeseceeceease
67. roup for informational purposes only without representation or warranty of any kind and SAP Group shall not be liable for errors or omissions with respect to the materials The only warranties for SAP Group products and services are those that are set forth in the express warranty statements accompanying such products and services Br if any Nothing herein should be construed as B constituting an additional warranty SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks F or registered trademarks of SAP AG in Germany and _ other countries Please see Www sap com corporate en legal copyright i ase ndex epx trademark for additional trademark information and notices i ot Es O D 41
68. s E Continuous Variables z im Continuous Targets Number Data Set Size cy Cross Statistics with the L Parie Indicators i Outliers E KTS Advanced Settings SAP Infinitelnsight 6 5 SP5 Model Reporting H a Alw a D a obdo Variable TIME Hel Data Set Estimation e Target 2 ozone a J EJE Bigle Pelele Ple elele q gt db b gt 4 89 8 L662 ap ap aj Lm 1 a ti ia 92 G0 co co 4 oo in J on E fa a ha e D En io SEINE op B a 5 b i A e lj pa in s 2 F H SIS F Pd w Previous CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series47 2014 SAP AG or an SAP affiliate company All rights reserved 47 Statistical Reports Options A tool bar is provided allowing you to modify how the current report is displayed to copy the report print it or Save it Display Options HH This option allows you to display the current report view in the graphical table that can be sorted by column This option allows you to display the current report view as a HTML table Some reports can be displayed as a bar chart This bar chart can be sorted by ascending or descending values or by ascending or descending alphabetical order You can also select which data should be displayed EEE Some reports can be displayed as a pie chart When the current report i
69. s displayed as a bar chart this option allows you to change the orientation of the bars from horizontal to vertical and vice versa amp Usage Options This option allows you to copy the data from the current view of the displayed report The data can then be pasted in a text editor a soreadsheet a word processing software If the current report contains more than one view for various variables data sets and so on this option allows you to copy all the views of this report If the current report is displayed as a graph this option allows you to copy it as an image and paste it in a word processing software or a graphic application gt e w amp This option allows you to print the current view of the selected report depending on the chosen display mode HTML table graph This option allows you to save under different formats text html pdf rtf the data from the current view of the selected report This option allows you to save under different formats text html pdf rtf the data from all the views of the selected report i g CUSTOMER SAP Infinitelnsight 6 5 SP5 48 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 5 Step 4 Using the Model Once generated the model may be applied to additional data sets The model thus allows you to perform predictions on these application data sets by predicting t
70. seeeeeeesseeueeeeeeesseasaeeeeeeeesssaaasess 63 4 3 1Summary of the Modeling Settings to Use In this step you will execute Scenario 2 using extra predictable variables The table below summarizes the modeling settings that you must use It should be sufficient enough for users who are already familiar with the Infinitelnsight modeling assistant For detailed procedures and more information see the following sections Replace the options given in the scenario by the following ones Task s Screen Settings Specifying the data source Data to be Modeled Inthe field Data Set select the file Selecting a cutting strategy CashFlows txt Cutting strategy sequential without test Specifying a description file Data Description In the field Description select the file KxDesc_CashFlows txt Defining the extra predictable inputs Selecting Variables Keep all the variables in the field Predictable Variables Kept Defining the Forecasts Number Summary of In the field Number of Forecasts enter 21 Modeling Parameters 4 3 2 Selecting a Cutting Strategy and a Data Source For this Scenario Use the file CashFlows txt as the training data set Select the cutting strategy Sequential without test SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 2 Modeling with Extra Predictable Variables61 2014 SAP AG or an SAP affiliate company All rights reserved 61 For the detailed procedures refer to sections 3 2 1 Selecting a
71. sions txt text file in which the data is separated by tabs or csv text file in which the data is separated by commas SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series53 2014 SAP AG or an SAP affiliate company All rights reserved 53 3 5 3 Opening a Model Once saved models may be opened and reused in the SAP Infinitelnsight v To Open a Model 1 Onthe main screen of Infinitelnsight modeling assistant select Load a Model The screen Opening a Model appears Opening a Model Data Type Text Files dass Censusi Kxen Classification Kxen Classification Kxen Classification dass Census01 Kren dlassfication dass Census01 hours per neekCensus0l Kxen Classification dass_Census01 Kxe 6 sf 2044401 09 15 17 31 ours per week_Census01 Kxen Regression 2 sd 4 01 09 15 35 21 age_Census01 fenRegresson SL SSCSC C i 0 09 15 36 37 age Census01 fenRegresson h 2014010915 37 35 dass_Census01 Kxe 2014 01 10 09 34 19 dass Census01_decisiontree Exen Classification dass Census01 Exen Classification 10 09 36 05 ge Refresh Af Delete Selected 2 Select one of the following Data Type options Text files Database Flat files depending upon the format of the model that you want to open 3 Click the Browse button A selection dialog box appears if Data Selection Select Source Folder for Data or E gal m g C Wsers natacha
72. t checked by default Modifying the Modeling Procedure Four types of modeling procedure are available Default which corresponds to the standard Time Series modeling Based on Extra Predictable Variables which works as a Classification Regression model build on the extra predictable variables with the signal as the target This mode can be used to refine and validate the extra predictable variables or to identify the useless ones Disable the Polynomial Trends which generates all the models but those containing a polynomial trend Customized gives the possibility to enable disable the types of models that will be generated by Time Series when analyzing the signal The following table lists the types of models that can be disabled Component Model Type Trends Lagl Lag2 Linear in Time Polynomial in Time Linear in ExtraPredictables Linear in Time and Linear in ExtraPredictables Polynomial in Time and Linear in ExtraPredictables Periodicities Cyclics Seasonalities Periodic Extrapredictables Fluctuations Autoregressive SAP Infinitelnsight 6 5 SP5 Description Previous value of the signal Value before previous Linear regression on the time Polynomial regression on the time Linear regression on the extra predictable variables Linear regression on the time and extra predictable variables Polynomial regression on the time and linear regression on the extra predictable variables Cyclic variable d
73. the Generating Process Se err oie NEIE EEE N TEA I NE E E 27 Visualizing the Model FSSUTS sarin cisicoweiivewaimcaniensametiebanadannwedinewaswcxxensamelistuabaeldctisewelve wehacaveusaivcbieednlndystowediverisbwsavicalauwetiosieds 28 CUSTOMER SAP Infinitelnsight 6 5 SP5 26 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 3 3 1Following the Generating Process There is two ways for you to follow the progress of the generation process The Progress Bar displays the progression for each step of the process It is the screen displayed by default The Detailed Log displays the details of each step of the process vV To Display the Detailed Log Click the Show Detailed Log button The following screen appears Training the Model a lo G5 Computing statistics Statistics Statistics Statistics Statistics Statistics Statistics Statistics For the final model an outlier has been detected at time point 1955 06 78 For the final model an outlier has been detected at time point 1965 10 78 For the final model an outlier has been detected at time point 1967 10 28 Computing statistics for the chosen model Computing statistics Computing statistics Computing statistics Computing statistics The final model is Sum Regression R_ozone la TIME Regression Minus R_ozone 1la Regression R_ozone la
74. the signal AmonthOfYear cyclic representing 12 months that is a periodic of one year The following graph presents this cycle compared with part of the signal once the trend has been extracted Signal Trend Cyclic 12 ai A ame Dem TA E AL A os ME 2 ae oe 2 ee ae S WO AA WOO WO W wv y J _lyv vy MS A Cyclic 52 representing 52 months that is a periodic of about four years SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series33 2014 SAP AG or an SAP affiliate company All rights reserved S _ __ _ a The Fluctuation The fluctuation is what is left when the trend and the cycles have been extracted It is modeled with an auto regression that uses a window of past data to model the current residue The number of observations in the window is determined by Time Series depending on the total number of observations in the Estimation data set For this scenario Fluctuations have been detected for this scenario The following graph presents the auto regression The orange area represents the window of past observations which are based on the past 37 months The point in the red circle is the point calculated by the auto regression 3 The Residuals The residuals is what is left when the trend the cycles and the fluctuation have been extracted from the signal This part called white noise and made of random elements that cannot be mode
75. uction to SAP Infinitelnsight The essential concepts related to the use of SAP Infinitelnsight features No prior knowledge of SQL is required to use Data Manipulation only knowledge about how to work with tables and columns accessed through ODBC sources Furthermore users must have read access on these ODBC sources To use the Infinitelnsight modeling assistant users need write access on the tables KxAdmin and ConnectorsTable which are used to store representations of data manipulations For more technical details regarding SAP Infinitelnsight please contact us We will be happy to provide you with more technical information and documentation CUSTOMER SAP Infinitelnsight 6 5 SP5 4 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide 1 1 3 What this Document Covers This document introduces you to the main functionalities of the nfinite nsight Modeler Time Series feature Using the application scenario you can create your first models with confidence Infinitelnsight Modeler Time Series formerly known as KTS lets you build predictive models from data representing time series Thanks to nfinite nsight Modeler Time Series models you can Identify and understand the phenomenon represented by your time series Forecast the evolution of time series in the short and medium term that Is predict their future values To know more about the basic concepts underpinning
76. very date of the training data set the error bars for the predicted variable Understanding the Applying Mode A Time Series model can only be applied on all or part of the training data set The result file contains the input variables that is the time and the signal and as many predicted variables noted KTS_x as the number of forecasts requested The following table describes a Time Series result file where TIME is the time variable and R_ozone la is the Signal variable TIME kxlndex F ozone la kts 1 kts 2 kts 3 1959 01 29 1959 02 29 1959 03 25 1959 04 20 Bz 9 210 4 320 4 250 Bae 1959 02 26 an 1959 03 25 od 1959 04 20 Legend is an estimator of KTS_lis an estimator of the current value of the signal at date 4 KTS_2is an estimator for the next value of the signal at date t KTS_3 is an estimator for the value of the signal at date 2 andsoon SAP Infinitelnsight 6 5 SP5 CUSTOMER Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Series51 2014 SAP AG or an SAP affiliate company All rights reserved 51 A Time Series model uses its own previous predictions to compute the next ones The values used to compute a forecast are presented below KTS_1 is computed using the current date known as the known extra predictable variables at this date and past values of the signal TIME kxlndex F ozone la kts_1 kts 2 kts 3 1959 01 45 1959 02 29 1959 03 26 1959 04 2
77. ving the Model Once a model has been generated you can Save it Saving it preserves all the information that pertains to that model that is the modeling parameters the forecasts view and so on vV To Save a Model 1 On the screen Using the Model click Save Model The screen Saving the Model appears 2 Saving the Model Model Name Rozonela Rozonelae ts Description TT Data Type TextFiles gt Folder j Samples KTS H Browse File Table J Browse 2 Above the Browse button select one of the following options e Text files to save the model in a text file e Database to save the model in a database e Flat Memory to save the model in the active memory 3 Complete the following fields e Model Name This field allows you to associate a name with the model This name will then appear in the list of models to be offered when you open an existing model e Description This field allows you to enter the information of your choosing such as the name of the training datasetused or the number of forecasts calculated This information will help you identify your model for a later use e Folder Depending upon which option you selected this field allows you to specify the ODBC source the memory store or the folder in which you want to save the model e File Table This field allows you to enter the name of the file or table that is to contain the model The name of the file must contain one of the following format exten
78. w sits 71 Learning Ne IN ONS ia cece cpettec center teisinei i arnir a aa iecnetaacinsaee bgrerseneoucepeeneuseneeeeee lt eieaeetes 71 SAVING WN TOO Uh epoe saseseeccaecececesemaeeuesegesenneeeeeeseceseasesee ac oscbeanen nc oete cleseuencemaesuacsuanteeeeseye ceplesnee cen avawacsuosceecesaueeteres 72 Opening UL Mode heisen E E 72 PAO DY UNG Me MOUE krisi E E A EAE E EEA oceans 73 Creating the time series model The following code describes the method to create a Forecasting model A default model is created including a Forecasting transform For this scenario creating a model with a Forecasting transform createModel Kxen SimpleModel model model pushTransformInProtocol Default Kxen TimeSeries setting the model parameters The cutting strategy is the only one model parameter which has to be set For this scenario setting the model s parameters model getParameter model changeParameter Parameters CutTrainingPolicy sequential with no test model validateParameter CUSTOMER SAP Infinitelnsight 6 5 SP5 68 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 2 Modeling with Extra Predictable Variables SSIS Opening the Training Data Set In this step the training data set CashFlows txt is opened with its description KxDesc_CashFlows txt As the file contains training information and predictive information the line index of the end of training is fixed For this scenario open the training dat
79. yam Documents Se afafa Samples E El Census H JapaneseData H I KAR H l KelData Text Files dat data csv txt User Password co CUSTOMER SAP Infinitelnsight 6 5 SP5 54 2014 SAP AG or an SAP affiliate company All rights reserved Scenario 1 Standard Modeling with Infinitelnsight Modeler Time Serie 4 Select the folder that holds the model that you want to open The list of models contained in that folder appears Opening a Model Data Type Text Files i Samples KTS es a a R_ozonejta R_lozoneda Kwen TimeSeries 2014 03 10 11 10 35 qe Refresh Af Delete Selected cle lt The following table lists the information provided for each model allowing to identify the model you want to reload Column Description Values Name Name under which the model has Character string been saved Class Class of the model that is the Kxen Classification Classification Regression with nominal type of the model target Kxen Regression Classification Regression with continuous target Kxen Segmentation Clustering with SQL Mode Kxen Clustering Clustering without SQL Mode Kxen TimeSeries Time Series Kxen AssociationRules Association Rules Kxen SimpleModel Classification Regression and Clustering multi target models any other model Version Number of the model version Integer starting at 1 when the model has been saved several times Date Date when the

User Guide

Contents

Download Pdf Manuals

Related Search

Related Contents