Report-Nov-1-2016.md

[rahlk]

Weekly update

Done

• Algorithmic update: metrics matches were inconsistent (changed every run) with SEER.
  • Fix: Implement Bootstrap sampling when comparing metrics in a round-robin fashion.
  • Issue: Very Slow!!
• Computed Pd, Pf for HDP
  • Alarmingly high False alarm rate (See issue report)
• Deployed All experiments on HPC for faster result turnout

Doing

• Trying combinations of various SMOTE parameters.
  • Only over/undersampling/both/nothing
• Expts. now running on HPC (will update once complete)

To Do

• Write Paper (Will start as soon as experiments terminate)
• Draft in 2 weeks.

Impediments

• Experimentation takes too long. One test on all dataset takes 24 hours
• Fixed. Now have access to ARC Cluster to run multiple experiments.

Admin

• None