|
PRESENT
|
- Jon Hsieh
- Jon Wildstrom
- Wing Leung
- Andy Zimdars
- James Lampe
- Eric Stein
- Rudy Setiawan
- William Ross
- Brian Woo
- Yun Ching Lee
- Michael Smith
|
|
GOALS OF MEETING
|
- Brian Woo from UI needs questions from us to ask the client for a visit this wednesday
- Event service may be networking's responsibility, so we will put off naming event service
expert until further notice
- Will focus on learning use cases for HW 1
- Address non-functional/pseudo requirements
|
|
STATUS
|
- Jon W: No event service report, see above
- Wing: Found public statistic modeling learning methods in C, still looking for Java
packages. Suggestions for finding commercial packages
- James: Found machine learning, expert systems, data mining packages, mostly in C/C++ and Lisp,
will also look for commercial Java packages. Rudy mentioned he had experience working with the MLC++
package, but is skeptical that we will find a good use for it, although it is possible.
- Jon H: Found ABE package, details were posted on BBoard.
|
|
DISCUSSION/ ACTION ITEMS
|
- Pseudo Requirements:
- Need to know platform requirements, Eric believes NT is necessity
- We assume database team will handle our database needs
- Seems to be the case that we are between MBNA and the dealer, not dealing
with Germany. Need clarification.
- Need to find out how frequently client expects learning to process new data
- Use Case General Model
- Wing will post graphical version of all models
- Going on assumption that we are on the server side
- Our four potentially relevant scenarios (#3, 4, 5, 6) all fit into this general model,
the specific implementations of our data mining functions will differ from scenario to scenario.
We are still figuring out how to model the specific data mining functions and may need more information
from the client and learning method researchers.
- Actors: Event service, clock
- Entry: Event service publishes event we are subscribed to
- Event logged into Log database
- Data miner periodically "wakes up" (by clock actor) and reads log database (it is expected data miner will be expensive and
should remain off-line)
- Data miner finishes computation and updates behavior file
- Behavior file will send a request to event service based on recommendations of data miner
- Object Model
- Log database: Logs all events that learning subscribes to from event service
- Event Record: Many event records in log database, they contain information such as time and
location of event, etc.
- Data Miner: Implementation TBD, will sort through data and find patterns
- Behavior file: Updated by Data Miner, sends request to event service
- Request: For example, in scenario 4, behavior file will send a request to update database to
event service
- Dynamic Model
- Sequence diagram #1: Event service sends event to behavior file, which will then send event back
to event service??
- Sequence diagram #2: Clock periodically "wakes up" Data Miner, which reads log database, which sends
update to Data Miner. Data Miner processes new data, finds behavior, then writes a behavior pattern
to behavior file
- Sequence diagram #3: Event service sends event to log database
- Action Items:
- Andy: Posting to bboard about handin format for HW 1 and also posting further explanation of diagrams
- Wing: Posting to bboard graphical versions of diagrams, wants suggestions for improvements
- James: Putting together Wing, Andy and anyone else's contributions for hand in.
|
|
WRAP UP
|
- Everyone read bboard to iron out ambiguities of the models
- PLEASE POST TO BBOARD AND RESPOND TO POSTS
- Everyone still working on research to define specific data mining techniques
|
|
CRITIQUE
|
- Working on overall meeting organization, efficiency and preparation
- Trying to get more ideas down in the minutes
|