Learning/Event Service
Minutes: 28 September 1998 1:30PM - Smith Hall 101
PRESENT
  • Jon Hsieh
  • Jon Wildstrom
  • Wing Leung
  • Andy Zimdars
  • James Lampe
  • Eric Stein
  • Rudy Setiawan
  • Brian Woo
  • Yun Ching Lee
  • Michael Smith
    • GOALS OF MEETING
  • Select team leaders for rest of project phases
  • Do RAD, SPMP, HW2
    • STATUS
  • Andy: At Architecture meeting, discussion of client side learning (Network considering possibility), and Database triggers learning
  • Wing: Found statistical learning packages that seem promising, may cost around $500
  • Jon W: What is budget? Answer unsure, Eric will check
  • Jon H: Does documentation liason exist? Not on SPMP template.
  • Eric: Bernd is in Germany
  • Yun-Ching: Need things to do - responsible for programming environment (as toolsmith).
  • New webmaster: Rudy
  • Preferences for programming environments: Mostly prefer NT/UNIX running JDK 1.2 (or is it 1.12?)
    • DISCUSSION
  • SPMP
    • 2.2.1 Teams and Tasks -- Learning team will intelligently analyze data flows and recommend actions to expedite data exchanges based on analyzation
    • 2.4 Project Responsibilities
      • 2.4.1 Project Management
        Leader Dates Phase
        Jon Hshieh 9/11-10/8 Requirements Analysis
        Andy Zimdars 10/9-10/26 System Design
        Rudy Setiawan 10/28-11/7 Object Design
        Jon Wildstrom 11/8-11/20 Implementation
        Wing Leung 11/22-12/7 Testing

      • 2.4.2-10 Team Assignments
        2.4.2 Coach Eric Stein
        2.4.4 Architecture Liason Andy Zimdars
        2.4.6 Documentation Editor James Lampe
        2.4.7 Configuration Manager Jon Wildstrom
        2.4.8 Toolsmith Yun-Ching Lee
        2.4.9 CASE Tool Manager Wing Leung
        2.4.10 WebMaster Rudy Setiawan
        2.4.?? Documentation Liason Jon Hsieh

    • 2.32 Meeting Times
      Mondays 6:30-7:50 Smith Hall 101

    • 3.2.1 Assumptions
      • Learning is on server side
      • Different types of data to analyze; one general system
      • Using 3rd party software whenever possible
      • Rely on database triggers for actions
      • Get A's if it works :)
    • 3.2.2 Dependencies
      • Rely on databse team for persistent storage (log, behavior files)
      • Rely on databse for actions
    • 3.2.3 Constraints
      • Learning problem essentially data mining
      • Everything needs to be Java or wrapped in Java
      • Domain specific knwledge may be limited
    • 3.3 Risk Management
      1. Risk: Code could be compute intensive
        • Write fast code; don't buy slow 3rd party code
        • Do processing off-line (data mining)
        • Have default behavior to avoid increased iterations
      2. Risk: No ability to use 3rd party code
        • Write decision tree
        • Don't overapply learning techniques
        • Don't expect too much domain specific code
      3. Risk: No packages in Java
        • Write wrappers
        • Get source for packages to enable cross compiles
      4. Risk: Need more than 1 learning algorithm
        • Accept suboptimal single solution for all scenarios
        • Use multiple packages
      5. Risk: Assumptions change because of other teams assumptions
        • Worry about it in systems design
      6. Risk: Subsystem assumptions invalidated
        • Retrench, restart
  • RAD
    • 1.0 Objective -- Same as problem statement
    • 2.0 Current Situation -- They ahvve CDs that get sent out every month; Intelligent learning nonexistent
    • 3.1 Overview -- Monitor System Activity, intelligently recommending optimal activity
    • 3.2 Functional Requirements
      • Analyze behaviors, suggest actions to implement
      • Monitor behavior noted by DB
      • Response triggered by analyzing triggers
      • Respond to databse triggers
      • Local databse storage of frequently accessed records
      • Rescheduling of updates over net to avoid network congestion and "irate customers"
    • 3.3 Nonfunctional requirements -- need to finish this
    • 3.3.1 User Interface and Human Factors -- We have a "learning preferences" panel for the client
    • 3.3.2 Documentation -- Documentation will include an explanation of the "learning preferences" panel. It will include a description (as examples) of the functionality that selection of the feature enables
    • 3.3.3 Hardware Consideration -- The system runs on NT, considerable RAM will be needed to handle events from as many as 6,000 dealers
    • 3.3.4 Performance Characteristics -- Needed to be fast preferrably, or offline computation is necessary
    • 3.3.5 Error handling and Extreme Conditions -- ??
    • 3.3.6 System interfacing -- from Event Service/DB - input and output
    • 3.3.7 Quality Issues -- Looking for near optimal intelligent recommendations
    • 3.3.8 System Modifications -- System should be easily extendable and modifyable, can perhaps reuse data mining process to learn other scenarios
    • 3.3.9 Physical environment -- None
    • 3.3.10 Security Issues -- Learning is an internal system; probably no concern for security
    • 3.3.11 Resource Issues -- Only persistent data on DB, DB handles it
    • 3.4 Constraints -- See SPMP
    • 3.5 System Model -- See HW 1 865?OpenDocument | | | |------------- posted document contents ------------- | |3.3.1 User Interface and Human Factors | |We have a "learning preferences panel" for the client. | | | |3.3.2 Documentation | |The Documentation for the learning system will be a brief description of the | |"learning preference panel". It will include a description of the functionalit| |y | |(as examples) that selection of the feature enables. | | | |3.3.5 Error handling and Extreme Conditions | |The major sections of our subsystems are persistant storage (event logs and | |behavior files), the dataminers, and our connection to the database/event | |service. | | | |We assume that the persistant data is robust and stable. In the situation tha| |t | |we are unable to reach the our persistant data, we are crippled. | Copy/Excerpt done! | |---------- posted document contents ends ----------- process, we fall back to | |a | |"dumb" algorithm (logging, and a default response) to process requests. They | |will be handled laters by the data mining cycles. | | | | | | | | | |
    • ACTION ITEMS
  • Eric: Pester about status of event service
  • Yun-Ching: Looking into platforms supporting CVS and JDK 1.2
  • Post your class schedules
  • Get spellings of names right in DB/documents: Yun-Ching Lee, Rudy Setiawan, Jonathan Wildstrom, Jonathan Hsieh, James Lampe
  • New webmaster, Rudy, get a hold of Joyce Johnstone about posting the websites
  • James: Submit draft of SPMP by friday (suggest corrections everyone??)
  • Jon H: Bug client about functional/nonfunctional requirements
  • Post to bboard, read Chapter 6 about nonfunctional requirements -- we need to finish this section
  • Post to BBoard by friday: what types of data patterns are we looking for; what data needed
  • Rough draft of team RAD due by wed; James will finish rough draft by tues to be approved by Eric
  • Get your pictures taken; we want cookies
  • All document posts need to be in html format
  • Jon W: Handle CVS
  • New research project people:
    • Machine learning: James and Rudy
    • Genetic Algorithms: Jon H. and Jon W.
    • Statistical Modeling: Wing, Andy and Yun-Ching
    • Neural Networks: Andy, Yun-Ching
    • CRITIQUE
  • Organization has improved.
  • Let's just try to get all our documents done; post to bboard with suggestions

    • This page is hosted by the Chair for Applied Software Engineering of the Technische Universität München.
    Imprint (Impressum)