​​Replacing PsfAttributes DM-11026
​​Child task of DM-2259
​​Question: does computeShape().getDeterminant return the right answer?
​​
​​Reasonable for 1pix < sigma < 9pix  for  simple 40X40 pixel model SingleGaussianPsf(40, 40, inputSig).
​​
 
User-uploaded image: figure_1-24.png
​​
​​And I think this ticket (https://jira.lsstcorp.org/browse/DM-5824) is saying that sub-pixel sigmas can’t be estimated (I wouldn’t expect that they would)  filed under ticket.
​​

​​​✅​ ModelPsfMatchTask (WarpedPSF (warped PsfExPsfs)):

​​
​​Running makeCoaddTempExp on HSC-I band data on ci_hsc:
​​
​​makeCoaddTempExp.py /Users/yusra/lsst_devel/LSST/repos/HSC/ci_hsc/DATA --rerun ci_hsc_ps1_2  --id patch=5,4 tract=0 filter=HSC-I      --selectId visit=903986 ccd=23 --selectId visit=903986 ccd=100 --selectId visit=904014 ccd=1 --selectId visit=904014 ccd=6 --selectId visit=904014 ccd=12 --selectId visit=903990 ccd=18 --selectId visit=903990 ccd=25 --selectId visit=904010 ccd=4 --selectId visit=904010 ccd=10 --selectId visit=904010 ccd=100 --selectId visit=903988 ccd=16 --selectId visit=903988 ccd=17 --selectId visit=903988 ccd=23 --selectId visit=903988 ccd=24 --doraise -c doApplyUberCal=False  modelPsf.size=39  makePsfMatched=True makeDirect=False --configfile ./psfMatchedWarpConfig.py  --no-versions --clobber-config
​​
​​Shows that PsfAttributes gives nonsensical Gaussian widths.  A FWHM of 0.860 is a common failure model. computeShape().getDeterminant() gives approx same FWHM for calexps in a single visit (they’re consistent). 
​​
​​Visit 903988
​​
 
User-uploaded image: fwhmScience7.94754.png
​​
User-uploaded image: fwhmScience7.97293.png
​​
User-uploaded image: fwhmScience7.92891.png
​​
​​
 
User-uploaded image: fwhmScience7.94173.png
​​
​​
​​Visit 903986
​​
 
User-uploaded image: fwhmScience6.92705.png
​​
User-uploaded image: fwhmScience6.93925.png
​​
User-uploaded image: fwhmScience6.85823.png
​​
​​
 
User-uploaded image: fwhmScience6.84967.png
​​
​​Visit 904010
​​
 
User-uploaded image: fwhmScience5.72415.png
​​
User-uploaded image: fwhmScience5.57754.png
​​
User-uploaded image: fwhmScience5.56060.png
​​
​​visit 904014
​​
​​
 
User-uploaded image: fwhmScience5.94112.png
​​
User-uploaded image: fwhmScience5.99582.png
​​
User-uploaded image: fwhmScience5.97417.png
​​
​​
​​Conclusion:  Replacing PsfAttri with computeShape is safe for this use case, and should be done ASAP. 
​​

​​​✅​ ModelPsfMatchTask (CoaddPsf):

​​
​​PsfAttr quits on CoaddPsf, making it impossible to compare, but computeShape gives FWHMs of 
​​~0.70 arcseconds which are sane.
​​
​​
 
User-uploaded image: 4.0875.png
​​
​​
​​

​​​✅​ ImageDifferenceTask 

​​
​​When testing on  on SDSS Stripe82: (PcaPsf vs. CoaddPsf(PcaPsf)) and Decam PcaPsf vs PcaPsf, 
​​computeShape returns a very different answer:
​​
​​.computeShape() does not give the same answer as PsfAttributes for Stripe82 PcaPsfs: 
​​
​​To reproduce on lsst-dev:
​​processCcd.py /lsst7/stripe82/dr7/runs/  --output /scratch/yusra/calexp_dir --id run=4192 camcol=4 filter=r field=104
​​
​​processCcd.py /lsst7/stripe82/dr7/runs/  --output /scratch/yusra/calexp_dir_psf --id run=4192 camcol=4 filter=r field=104 --config charImage.doMeasurePsf=True