AP Pipeline QA Meeting, 2020-09-21

4pm Pacific Time

BlueJeans: https://bluejeans.com/426716450

attending:

• Eric Bellm
• Andrew Connolly
• Bryce Kalmbach
• Unknown User (cmorrison)
• Jeffrey Carlin
• John Swinbank
• Krzysztof Suberlak
• Meredith Rawls
• Krzysztof Findeisen
• Ian Sullivan
• Unknown User (gkovacs)

regrets:

Topics for discussion:

• Precursor processing (Meredith Rawls ):
  • DM-26380: HiTS reprocessing.  Essentially no changes since August
  • DM-24252: HSC COSMOS processing.
    • previously, we were hitting database lock errors with sqlite: now using Postgres for apdb, no more trouble
      • all of the visits are completely populated, unlike last time where some ccd-visits didn't get recorded
    • previously, had lots of background noise sources.  Hoped that applying variance rescaling it would help–but HSC is still showing a large number of sources.
      • coincidentally, one of our HiTS fields is cosmos, although HSC is much deeper
    • still losing some visits in i and z due to some database complaints about visit duplication, which shouldn't be correct
    • now all filters are getting loaded into a single apdb, which worked–and is exciting!
      • Jeffrey Carlin thinks the visits should all be even numbered
    • 5M diasources, 2M diaobjects
    • some discussion about individual visits–edges are pretty bad everywhere, but some individual visits seem okay
      • haven't yet dug into correlation with individual visit properties
      • also maybe useful to compare to analysis Yusra did last summer
      • found some negative values in totFlux–probably a background subtraction problem
    • question: is the postgres table queryable by others?
      • Meredith Rawls see what's needed for others to be able to access this 28 Sep 2020 
      • Eric Bellm chat with NCSA about productizing APDB processing 05 Oct 2020 
• Crowded fields processing (Ian Sullivan ):
  • showing Saha DECam fields (~0.5M sources / sq deg)
  • lack of isolated sources for PSF/backgrounds tends to make the pipelines fall over in single frame processing with default configuration
    • ~30% of the visits fail with defaults, 6000 visits
    • main failure point turned out to be having PCA PSF estimator as the default–switching to PsfEx fixes most of the problems, down to a 1000 failures even with default PsfEx parameters–with some tuning thinks he can get almost all frames through, even without any kludges to chi-sq fits, etc.
    • in some cases, cosmic ray detection finds everything --iterating with PSF modeling seems to work
    • obviously good if we can get a good PSF. A&L diffim doesn't need a PSF–we should consider how to make single frame robust if needed
      • but since it seems to be a very small number of images we can probably defer past the report to Zeljko
      • Ian discussed with Colin–they agreed that we want the LSST stack to at least complete some processing, so science users can come in and work more with the data if needed without going to the raws
    • showing a first diffim–pretty sparse, although there seem to be a reasonable number of dipoles
• SITCOM team updates (Andrew Connolly  et al.):
  • simulating out-of-focus images with ComCam and LSSTCam, plus wavefront sensors
    • Comcam loop working: detect images, measure donuts, calculate zernikes, determine correction to mirror, re-simulate with correction.  converges!  at least on fields of medium density.  next will try on higher density sources.
    • now trying to do the same with LSSTCam and wavefront sensors. 
    • next want to integrate with T&S software
    • then want to think about deblending in the donut context
    • question: does DM plan to process out-of-focus images at scale?  John Swinbank and Eric Bellm don't think this is a current DM deliverable, although may be useful to connect with Josh Meyers to share knowledge
    • how good is the pointing knowledge?
      • Chris
• Next meetings (all at 4pm Pacific):
  • Oct 19
  • third Mondays of the month, thereafter?  no, too many conflicts.
    
    • November 16
    • December 14
    • Jan 25
• Review CI (https://chronograf-demo.lsst.codes/):
  • no changes in the last 30 days
• AOB

Action Items