330 likes | 345 Views
CORE Bringing the GSBPM to life! J. Linnerud & J.-P. Kent. Main points. An ideal development process for a statistical system Why this ideal usually is not met How CORE aims at supporting this ideal development process. Statistics: How?. Specify a statistic
E N D
CORE Bringing the GSBPM to life! J. Linnerud & J.-P. Kent
Main points • An ideal development process for a statistical system • Why this ideal usually is not met • How CORE aims at supporting this ideal development process
Statistics: How? • Specify a statistic • Design a process that will produce this statistic • Build a system that will execute this process
What is the product? • Define a statistic • What does it say? • Measures, dimensions, explanations… • What does it look like? • Tables, Press release, Analytic paper… • What is the input? • Population, variables, data sources… • What is the relation between input and output? • Methods to apply
Statistics: How? • Specify a statistic • Design a process that will produce this statistic • Build a system that will execute this process
How to produce the statistic • Model the data • Input, output, intermediary results • Specify process steps to apply the chosen statistical methods • Integrate these steps in a process flow
Statistics: How? • Specify a statistic • Design a process that will produce this statistic • Build a system that will execute this process
Let the machine do it • Implement the data models • Implement the process steps • Implement the process flow
Why is this approach good? (1) • Variability vs. stability • Statistical products are specific • There is a great variety of products • A given product will vary in time • Statistical processes are generic • The same method can be applied to many products • Process steps implementing methods can be reused • A significant change in the product can be implemented with some simple changes in some process steps
Why is this approach good? (2) • It allows a clean specification of the product • In terms of what it is • In terms of what is used • In terms of what the relation is between input and output
Why is this approach good? (3) • It separates product design from IT • The product is defined in terms of what it is (and not how it is produced) • The process is defined in terms of what it does (and not how it is implemented) • Only the system is defined in technical terms
Why is this approach good? (4) • It supports optimalisation of process development • Possibility of developing standardised, re-usable process steps • Generic process steps are not defined for an actual statistic, but for use in different statistics
Main points • An ideal development process for a statistical system • Why this ideal usually is not met • How CORE aims at supporting this ideal development process
The usual approach • Statisticians present a project in which product and process are combined • IT people specify and build a system that creates the product by performing the process
Why is the usual approach inefficient? • Complexity • Process & product are tightly coupled • Rigidity • Maintenance is labour-intensive • Specificity • It is not easy to devise a generic solution when developing for a specific product
Main points • An ideal development process for a statistical system • Why this ideal usually is not met • How CORE aims at supporting this ideal development process
Promoting the better approach • The CORA and CORE projects (Jenny) • Bringing the results into practice (Jean-Pierre)
CORA ESSnet • COmmon Reference Architecture (CORA) • Financed by Eurostat under 2009 Statistical Workprogramme • Countries involved: it (coordinator), ch, dk, lv, nl, no, se • Duration: October 2009 - October 2010
CORA deliverables • Questionnaire • Set of Requirements • State of the Art • Definition of the Layered Model • Technical Annex • Instruction Manual • Commercial and Legal Foundations for the Exchange of Software between Statistical Offices • Requirements Checklist for CORA Tools • Recommendations for CORA Tools
After CORA … CORE! • COmmon Reference Environment (CORE) • Financed by Eurostat under 2010 Statistical Workprogramme • Countries involved: it (coordinator), fr, nl, no, pt, se • Duration: December 2010 - January 2012
CORE Workpackages • Design of the information model according to GSBPM and alignment with NSI's information models • Generic interface design for interconnecting GSBPM sub-processes • Research workflow solutions for process management • Implementation library for generic interface and production chain for .NET • Implementation library for generic interface and production chain for Java
Practical usage of CORA / CORE • Modeling a process in terms of services (CORA) • Classifying services (CORA) • Making services platform-independent (CORE)
A transport statistic Input: Loading reports Unloading reports Date, time, place, type & quantity goods, type vehicle Output: Monthly transport data Same data also used for time series An example process
Use the CORA space grid Modeling approach
Macrodata Microdata
Use the CORA space grid Display statistical services in the appropriate cells Modeling approach
Aggregate Macroediting
Use the CORA space grid Display statistical services in the appropriate cells Join services with arrows to show the dependencies Modeling approach
Aggregate Macroediting
Confidentialty control Monthly Transport Publication Archive Publication data Select period data Archive Time Series data Archive Statistic data Integrate data Supply period data Aggregate Archive Unit data Archive obs. vars. Download Combine Microediting Macroediting Outlier detection Error detection Correct outliers Correct variables Compute distance ? ?
A traditional service Script (X) Model (X) Model (X) Input (X) Tool X Output (X)
A CORA service CV CV CV CV CV Script (Y) Model (Y) Model (Y) Input (Y) Tool Y Output (Y) Script (CORA) Input (CORA) Output (CORA) Model (CORA) Model (CORA) Logging CV CV CV CV CV Script (X) Model (X) Model (X) Input (X) Tool X Output (X) CV = Convertor