Paul Price added a comment - 20/Mar/17 10:25 AM Chris Morrison [X] , would you mind giving this simple change a quick look, as I believe you grok this stuff. price@pap-laptop:~/LSST/obs_subaru (tickets/DM-9874=) $ git sub-patch commit e378a39f4967f0b8567b335288e9f621115e1118 Author: Paul Price <price@astro.princeton.edu> Date: Mon Mar 20 13:00:34 2017 -0400   config: set matcher.numBrightStars=150 This greatly increases the stability of the matcher, at least for MADCASH dwarf galaxy fields. There's also evidence from UW astrometry experiments that it helps in general.   diff --git a/config/processCcd.py b/config/processCcd.py index a1a7d2e..2d9755d 100644 --- a/config/processCcd.py +++ b/config/processCcd.py @@ -34,6 +34,10 @@ for refObjLoader in (config.calibrate.astromRefObjLoader, ): refObjLoader.load(os.path.join(getPackageDir("obs_subaru"), "config", "filterMap.py")) +# Better astrometry matching +config.calibrate.astrometry.matcher.numBrightStars = 150 +config.calibrate.photoCal.matcher.numBrightStars = 150 + # Set to match defaults curretnly used in HSC production runs (e.g. S15B) config.charImage.catalogCalculation.plugins['base_ClassificationExtendedness'].fluxRatio = 0.95 config.calibrate.catalogCalculation.plugins['base_ClassificationExtendedness'].fluxRatio = 0.95

Chris Morrison [X] (Inactive) added a comment - 20/Mar/17 10:35 AM Hey Paul, This is basically one of the major problems with the current matcher set up. Turns out in denser fields 50 bright stars is too few to successfully match on. One quick question before I approve this change, have you tried excluding saturated and interpolated objects from the source sample? I found that in dense stellar fields the first 50 objects by brightness are all saturated and causes the matcher to fail or give bad fits. I'm likely going to change the matcherSourceSelector to exclude these objects completely and instead rely on a S/N cut to pull in more objects if needed. Would be interesting to hear if you also have problem with saturated objects.

Paul Price added a comment - 20/Mar/17 10:37 AM I haven't tried that. Is there a quick config option to enable that?

Chris Morrison [X] (Inactive) added a comment - 20/Mar/17 10:43 AM Unfortunately no. You'd have to change the ordering of the isUseable and isGood flags in matchOptimsiticBTask or just switch isUseable for isGood. I'm going to hopefully implement this change as part of DM-9750 . I'll mark this ticket as reviewed. These are the visit numbers you sent me previously correct? I'll take a look at them once I start my validating my replacement matcher against a large volume of data.

Chris Morrison [X] (Inactive) added a comment - 20/Mar/17 10:46 AM Change in number of brightStars approved. This change does help the success rate of a generic implementation of matchOptimisticB as has been worked on in DM-7366 . Future releases may also want to remove saturated and interpolated objects from matching. This is planned to happen as part of DM-9750 .

Paul Price added a comment - 20/Mar/17 10:59 AM Thanks Chris! This passed Jenkins .

Paul Price added a comment - 20/Mar/17 11:01 AM Merged to master.