In - DM-19292Getting issue details... STATUS we determined that our current AP HiTS processing shows much higher rates of false positive detections than the ~50 DIASource detections per CCD reported at 5 sigma with decorrelation on in DMTN-021. The goal of this sprint is to explain (not necessarily correct or remove) the excess detections.
Daily standups (BlueJeans): time TBD
Weekly meeting: Monday 2pm Pacific, if Yusra can make it?
Slack channel: #dm-alert-prod
- start with "False Positive Census"
- counts of single/few/many detection DIASources
- aggregates by CCD, visit, night
- aggregates by DIASource flag type
- aggregates by seeing
- replication of DMTN-021 → current ap_pipe
Hypotheses & Potential Tests/Mitigation
- Overfitting of kernel basis functions
- reduce number of basis functions
- try alternate basis set
- regularize kernel solution
- compare to ZOGY
- Lack of obs_decam illumination correction
- exclude bad observations from counts
- count only science > template images
- Need better use of flags to discriminate edge detections, etc.
- counts excluding flagged edge pixels
- Template problems
- compare to calexp template case