AP Pipeline Meeting, 2020-07-13

2pm Pacific Time

BlueJeans: https://bluejeans.com/426716450

attending:

• Eric Bellm
• Krzysztof Findeisen
• Unknown User (cmorrison)
• John Swinbank
• Meredith Rawls
• Yusra AlSayyad
• Krzysztof Suberlak
• Jeffrey Carlin
• Andrew Connolly
• Bryce Kalmbach
• Unknown User (emorganson)
• Ian Sullivan
• Spencer Nelson

regrets:

Topics for discussion:

• Discussion of monthly reprocessing results (Meredith Rawls and all):
  • showing June 2020 reprocessing (July is done but hasn't yet been analyzed)
  • DM-25125
    • https://github.com/lsst-dm/ap_pipe-notebooks/blob/tickets/DM-25125/DM-25125-June2020-ap_pipe-HiTS.ipynb
  • number of sources has increased modestly since April (which we assume means more bad sources: would be helpful to look at the cutouts here)
    • originally thought this was ISR, but no change after ISR fixes
    • so fixes to decorrelation afterburner seem more likely.
    • no obvious sky pattern
    • Eric thinks we should think about how to enhance the "looking at stamps" aspect of QA, and how that might fit into real/bogus training infrastructure
  • also showing diffim runs on HSC (all five filters)
    • (and it overlap with HiTS!)
    • lots of overdensities around bright stars, along the camera edge
    • plotting DIAObjects sized by the number of DIASources; needed to exclude single-apparition DIASources to have the plot even be viewable
    • also plotting focal plane small stamps of each visit–clear CCD failures, which seem to drop out more and more in time?
    • Yusra thinks this is all the variance plane that needs rescaling–has a ticket she can send to Meredith for review
    • flag selections from DECam (bad/suspect/saturated center) don't remove as many sources as in DECam
    • has looked into the logs a bit for the failing CCDs; most are due to failure to find templates that match?  This is puzzling
  • date for next meeting
    • August 17
  • Actions
    • Eric Bellm send commissioning team reminder 17 Aug 2020 
• WCS for out-of-focus images (Andrew Connolly )
  • showing (phosim-simulated) out of focus image from Comcam
  • want to take wavefront sensor images, detect and identify isolated donuts (perhaps using input catalog), deblend others, than then get Zernike polynomials to adjust mirror supports
    • use both a DM and a T&S code for detection
    • have a model for the donut–preconvolve with the model
    • have questions about how to configure the DM detection task
      • Yusra greps the outputs of the monthly DRP reprocessing...
      • Doxygen documentation is the other place to look–Andy has difficulty walking through all the retargeted tasks...
      • John S requesting a concrete example to feed back to JSick–Andy will do so
      • Yusra: lower-level methods of detection are pretty flexible, can be adapted as needed
      • detection is not actually on pipelines.lsst.io (meas_algorithms and pipe_tasks are not converted–summer student at UW working on it...)
    • after finding sources, run astrometry–which fails here on a low-dec field
    • T&S code lets you detect only isolated donuts, after which the WCS solution converges.  Assumes won't have full camera WCS in time from DM
    • also not sure how dynamically to limit magnitude range of input reference catalog
      • Eric wonders about counting number of detections and sort reference catalog by brightness
      • Yusra interpreted this as a config question–it's in ScienceSourceSelector
    • also would like to learn to use the ReferenceStarSelector to assist with understanding deblending
    • discussion:
      • Gabor cautioning to attend to `doSmooth` in detection, to avoid assumptions about PSF shape
      • JDS: if the T&S code works, do we need the DM code to work here?  Andy: still under discussion; commissioning wants to be able to analysis donuts at scale (full focal plane) rapidly, which suggests DM code for parallelization.  So at minimum want to be able to work with both systems for now

• Review CI (https://chronograf-demo.lsst.codes/):
  • nothing new

Action Items