Notes from meeting with Dr. Greiner

466.md

• forbidden set is a huge cluster of dups ==> randomly select a subset of the forbidden ones

• test it on never before seen data

• test without Anahita's data

• which of Anahita's data did we use? ==> did we use Anahita's word lists?

• make sure that there's a clear demarcation between Anahita's

• read decision tree and make sure that we're actually making legit decisions

• make sure that it's not learning off identifiers

• analyse features, explain which contribute most, can we identify where it didn't work ==> what's useful/useless?

• Find table from Anahita's thesis, false positive, true positive, etc.

• Same problems as Anahita?

• Find marginal effects of stack trace evaluations, only word lists, etc.

• Use a set of only stack traces to perform stack trace evaluations

• Use this on eclipse?

• try removing stack traces => easy or difficult predictors?

• Look at crash repositories

• "oblation studies"