Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: jointcal
-
Labels:
-
Story Points:4
-
Epic Link:
-
Sprint:Alert Production F17 - 9, Alert Production F17 - 10, Alert Production F17 - 11, AP S18-1, AP S18-2, AP S18-3, AP S18-4, AP S18-5, AP S18-6, AP F18-1, AP F18-2, AP F18-3
-
Team:Alert Production
Description
Push jointcal and meas_mosaic catalogs through validate_drp, generate plots, and look at the results. John Parejko will do a first cut look to check for obvious problems, before handing off to Jim Bosch for the final comparison.
If changes to jointcal are necessary, we can add them to this epic and then re-compare.
Attachments
Issue Links
Activity
DM-14786-performance.tar.gz
I now have an "enough complete" run on PDR1 (using DM-14786) that we can finally start asking some useful questions.
I've uploaded a tarball containing the RST and converted HTML summary tables from validate_drp, and some PNG plots created from that. I'll let Jim Bosch summarize the results once he's had a chance to look at them, so I don't bias things too much.
There is one obviously bad astrometry tract+filter: HSC-Y 15832. Looking at the log (/project/parejkoj/DM-11783/logs/jointcal-SSP_WIDE_15832_HSC-Y-141536.2.log), jointcal definitely had trouble with it, ending with chi2/ndof=827, which is much higher than the others. Is there anything in this tract+filter that might cause problems with jointcal's astrometry, Hsin-Fang Chiang?
I guess "In review" in this context means "please look at the plots and numbers"?
John Parejko When you run these photometric repeatability numbers in the tarfile, what object selection has been done? All objects? Just stars? S/N cut? Color cut? Etc.
That would be a question for Michael Wood-Vasey, on the details of validate_drp.
S/N > 100
base_ClassificationExtendeness_value < 1
No color cut. No deblending cut.
Michael Wood-Vasey
added a comment - S/N > 100
base_ClassificationExtendeness_value < 1
No color cut. No deblending cut. This does indeed look quite promising!
Please do take a close look at the one outlier tract; I'd both like to understand what went wrong (too many "outliers" rejected? not enough? bad matches?) and hopefully find a configuration that gets it through. Strongly recommend you start by looking at the y-band coadd images.
Once that's working, I think the next step in validation would be to build coadds based on jointcal outputs (which still has some outstanding tickets, I believe).
Outlier tract is investigated on DM-15344.
This is a filters-per-node matchedVisitMetricsTask executor script for the verification cluster for meas_mosaic. Waiting to see if the results are sensible.