|
PRESENT
|
- Jon Hsieh
- Jon Wildstrom
- Wing Leung
- Andy Zimdars
- James Lampe
- Eric Stein
- Yun Ching Lee
- Rudy Setiawan
|
|
STATUS
|
- Andy: Found jaNet Neural net. Consolidating RAD's at Arch meeting
- Jon W: Project managment insists on using Voyager. May have conflicts with being implementation leader, posting schedules again necessary to devise new meeting times. For those who have not added anything to CVS: 1)Set env (as in handout), 2)cvs update creates directories 3)copy file to that directory 4) cvs add and commit commands (from handout)
- Rudy: To specify the data mining process, can use resources worked on from last summer job. Our data miner is almost
equivalent to a Builder object on his project, which interfaced with packages such as ID3, NaiveBayes, etc. Also uses MLC++ somewhere.
Builder constructs a model object using the optimal ML package (similar to ur behavior file, not a file but a set of rules).
Goo-goo-ga-ga may work with this.
- James: HW 1 is in Elizabeth's office, HW 2 is on web page (see action)
- Wing: Waiting approval from a professor to use CMU AI repository packages, need wrappers. Found java based Bayes Net, will post specific findings (see action)
- Jon H: Found 2 packages, 1) JAG - Simple genetic algorithm, easy integration; 2) GOO-GOO-GA-GA - Java front end for C back end, fuller and faster, harder to integrate. Had documentation meeting, Dan needs latest RAD (see action). We also need to prepare scenario demonstrations showing off our subsystem inupcoming presentations.
- Eric: SCS DB backup problem led to our BBoard woes, 413 will take over backup responsibility so it shouldn't happen again. Will also check if notifications are working (see action).
- Yun-Ching: Joyce gave him info on software package that runs on win95, Mac and afs, not sure if it will run outside of afs, joyce has yet to get back to him.
|
|
DISCUSSION
|
- More RAD Revisions
- Andy made basic nonfunctional changes
- Dealer profiles: we're doing it anyway; assume its up to us or differentiate from DB -- since we need less information than DB and our dealer profiles may be useless for the rest of the system, DB should
handle it and grant us access to necessary info.
- Eureka! Problem statement says we should use Visibroker (see action).
- Data builder suggested by Rudy would make Behavior file a rule base rather than a DB. Will there be a way to save these objects if system goes down? This backup of PAID system software needs to be addressed in quality issues nonfunctional req. section in RAD
- Decision trees will categorize groups of dealers with similar characteristics, one decision tree would be made for the entire data set instead of trees for each dealer. What about case where tree incorrectly assigns dealer into wrong branch of tree? Need extensive training data to build accurate tree then rebuild tree when new data (i.e. dealer does something unexpected) is available (build tree with old and new training data). This will require more thought, Wing will post with concerns about this algorithm (see action).
- Glossary
- behavior file - Object consisting of rules for recommending actions
- event service
- database trigger - Database action that learning is interested in monitoring, not a network event but a creation of new record, etc.
- ORB
- subscriber
- publish
- event channel
- learning preferences panel
- scheme
- naming service (dealer/machine level? D-B servers)
- API's we provide for interaction for other teams
- Event service
- publish, subscribe, unsubscribe, register as publisher, add channel, delete channel
- In case they switch from Voyager to Visibroker, need to allow extendability
- Learning preferences panel
- Scheduler
- Learning (force data miner to mine, including model and building)
- UI reports
- API's others provide
- DB trigger handling
- Database access (dealer profiles)
- Report format for UI
- Concurrency Identification Section of SDD
- Network: Rudy
- Jon H: DB
- Jon W: UI
- Yun-Ching, Wing - Learning/Event Service
- James: Auth
- Andy: Arch
|
|
ACTION ITEMS
|
- Andy
- Posting definitions
- Posting about network reports
- Asking mgmt about Visibroker
- Figure out status of next meeting
- Rudy
- James
- Notify swati location of HW1
- Send Dan in Documentation the latest RAD
- Find HW2, notify Wing/Jon W
- Jon W
- CVSing HW 2
- Try to provide research directory in CVS for others to submit findings
- Wing
- Will post specifics of Bayes net.
- Posting about concerns over d-tree algorithm's applicability to this problem
- Find HW2, email tloh its whereabouts, tell Jon W so he can CVS it.
- Jon H
- Notify James about missed items in last week's minutes (attributes we need to monitor).
- Still figuring out API with DB team
- Eric
- Checking into notification problem.
- Posting about concerns over d-tree algorithm's applicability to this problem
- Everyone
- Post class schedules AGAIN
- Need derogatory acronym to name database
- Post definitions for glossary
- Don't post to arch glossary
- Concurrency identification due by friday; see above for individual assignments.
Can someone please tell me exactly what we're supposed to do with that again? In one ear out the other.
|
|
CRITIQUE
|
|