BlueJeans: https://bluejeans.com/426716450
attending: Eric Bellm Meredith Rawls Ian Sullivan Unknown User (cmorrison) Unknown User (gkovacs) Krzysztof Findeisen
Topics for discussion:
- HiTS reprocessing (Meredith Rawls):
- was sick all last week, just returning to work. Trying to tie up some loose ends from a review from Krzysztof, and closing out a review from Lauren on making standalone scripts.
- needs to start digging in to flux correlations, and then will start on "building calibs"
- DES processing (Unknown User (emorganson)):
- Eric is out with a sick kid
ap_
association (Unknown User (cmorrison)):- went through Yusra's prototype SDM code and got it to put out pandas tables, etc. matching current output.
- one awkward part is getting into pandas since we're not starting with parquet
- also some baked-in assumption that all columns exist ahead of time (e.g., create empty columns and wait for data)
- some extra looping required to pick up calibrations, etc. since YAML doesn't know
- Metrics & ap_verify (Krzysztof Findeisen ):
- Metrics migration ticket is in review, will work on other items in the meantime
- image differencing (Unknown User (gkovacs) ):
- chased down different number of detected sources between his processing and Meredith's–only difference was reference catalogs (gaia vs. PS1). Tried swapping those on a small region and the problem is not resolved
- then got bogged down trying to plot with WCSAxes in astropy–doesn't quite work.
- John Parejko suggests we should think about what WCS to include with alert cutouts–approximate linear probably sufficient
- next back into the rogue's gallery of bad subtractions
- DCR (Ian Sullivan ):
- made ~1000 simulated observations for several fields over multiple years based on OpSim baseline
- scheduler has constraint on airmass but a bad one on parallactic angle–on absolute value, but that's not helpful–only gets effectively two independent measurements! even with more observations. so a double failure–doesn't avoid DCR, but also doesn't train the model
- also doing some sims with more effective potential observations patterns.
- starting to need metrics/means of evaluating. Will start with false positives (any detection in these static sims is a false positive)
- Review CI (https://chronograf-demo.lsst.codes/):
- no changes in last 30 days
- AOB/table round:
- fake insertion
- Unknown User (cmorrison) asked about this at DESC; RHL thinks it should in principle work maybe with some data type
- sounds like HSC code is just putting things on a spatial grid
- fake solar system attribution
- no meeting next week (Eric and John both on vacation)
- John Parejko reports that ARCH is now determining if AP header contents are sufficient to reconstruct ObservatoryInfo
- John Parejko asking about impact of astrometry on false positives:
- Unknown User (cmorrison) thinks that matcher change indicates that's ~1%
- jointcal to produce optical distortion model to be used in single-frame astrometry
- dipole fits are in DIASource catalogs, but Unknown User (cmorrison) says it's not making into the PPDB yet because there are more translations needed to get the columns to map
- could imagine editing WCSs output by jointcal and just rerunning image differencing to see the difference
- jointcal vs. SFM give 50-100 mas offsets; DCR give 0.2 pixel motion at 1.5 airmass–so comparable.
- fake insertion