today's almost demo
- show data set that we created in outlook
- show that data in kb
- sensitivity field had to be hand amended due to re-harvest problem
- ___will ask Oliver for batch harvesting
- ___no emailFinal event appears in logs
- ___will ask Chris to look into problem with incremental re-harvesting
- ___can we copy a kb from one machine to another
- demonstrate procedure for ITL
- show reduced field count in ITL and less cluttered screens
- save procedure
- ___still waiting green light from Will Haines on pulling in updated ITL code to do messaging hookup with ITL server
- Parse procedure from sparkle (now using hardwired procedure)
- SATService is persisted for use by other UIL components
- ___still need to host this in desktop
- so will pop up UI for feature relevance from eclipse
- user interacts with a few fields to use as features (note ui improvements)
- user dismisses UI
- ___ui not yet pushing info to the Learner
- Learner should train on the data set
- ___still to be done: TrainingSetCreator needs to hook in newly revised SATService to do hardwired sensitivity labelling of emails in KB
- Classifier is created
- ___still to do: use newly created persistence layer to persist the classifier
- this marks the end of the learning phase
- email sent by user
- ___to be done: PredictorManager uses real classifier rather than dummied up alternating answer
- ___to be done: PredictorManager passes prediction and pojo to SATService for sensitivity_forget labeling
- if user forgot, they will see a dialog in Outlook.
- ___need to use Procedure name in dialog that appears in outlook.
Open issues
- ___need ability to reharvest existing email data set. (batch preferred. Currently, Judah and Jed think that "single click on email" to reharvest won't work because the UID is set to "notyetset" on the relevant event, so harvest will be thwarted.(asap)
- ___outlook addin: no To address is put into harvesting request (may need for post-demo research)
- ___outlook addin: no sendEmailFinal event flows (no q*.xml file in IrisTracer logs show it) (need for post-demo research)
- ___incremental harvesting not working (if fix, can remove ID hack)
- ___dialog to show training example added? three emails in demo story - one pos one neg one forget case
Hacks in play
- bcc field is being used to convey sensitivity
- repeated harvesting requests are thwarted by putting bogus UID in harvesting request for all but send email initial and send email final event, thought latter doesn't flow. This works, allows all fields present to show in KB.