BlueJeans: https://bluejeans.com/426716450
attending: Eric Bellm John Swinbank Krzysztof Findeisen Unknown User (emorganson)Unknown User (gkovacs)Ian Sullivan Unknown User (cmorrison) Meredith Rawls
Topics for discussion:
- HiTS reprocessing (Meredith Rawls):
- also has a couple tickets of work to extract from pre-AAS notebooks designed to make nice skyplots, etc.
- getting cutouts from the template given source coordinates is still super gross (has to iterate through all the patches). suggestion to ask on #dm
- comments at the poster: Matthew Penny from OSU has a DECam time-variability dataset, would be interested in comparing to his pipeline
- Eric Bellm reports that HiTS DR1 doesn't seem like it's going to be useful–variable vs. non-variable cut is machine-learned in a weird way, features aren't stored, etc.
- Looking source by source: lots of bad kernel-match failures (task for Unknown User (gkovacs) to examine)
- Also an issue that constructing "DC fluxes" from template flux + diffim flux shows systematic effects
- Probably should update the HiTS dataset templates to be compareWarp–we are still using the two-year-old versions?
- DES processing (Unknown User (emorganson)):
- has been thinking about integrating whole system from raw exposure → alert production; but don't think we're close to be able to do this realistically in terms of triggering processing, preloading catalogs, etc. Engineering team not using anything realistic for pipeline payloads.
- relevant official milestone is probably two years out
- no blockers Eric is aware of on the pipelines side
- Have gotten something similar to Maria's mini-brokers running on kubernetes
- looking at processing times: about 10x too slow
- has been thinking about integrating whole system from raw exposure → alert production; but don't think we're close to be able to do this realistically in terms of triggering processing, preloading catalogs, etc. Engineering team not using anything realistic for pipeline payloads.
- CI (Krzysztof Findeisen ):
- CI change simon called out: none of us has tracked this down. What's this workflow? John suggests we review the dashboard regularly in this meeting.
- John Swinbank investigate code change that led to change in AP squash dashboard. — This is now DM-17225 - Getting issue details... STATUS
- John Swinbank ask Frossie if chronograph dashboards are going to get clobbered regularly
- Krzysztof Findeisen will work on improving dashboards if work won't be wasted.
- DCR (Ian Sullivan ):
- for AAS, put quasars in his sim and they didn't show up as excesses in any sub-band as they should have. Has some ideas but no success yet–recovers overall slope very well but not catching within-band features. Didn't have ability to do this with previous prototype. Hard to say if it's a bug in the simulations, a feature of the simulated observations, or something else.
- Eric: science utility of figuring out above is clear; is it necessary to address false-positive reduction?
- Ian: next steps: stabilize the code after pre-AAS hacking (DM-16864). At AAS Gordon Richards gave him catalogs of quasars in HiTS which could be used for checking.
ap_association (Unknown User (cmorrison)):- Got most of the way through DIAForceSourceTask, but hit some problems with SDM standardization stack. Will probably have to wait for help with Yusra in February, do new matcher in the meantime
- image differencing (Unknown User (gkovacs) ):
- met with Yusra last week; learned of the existence of a flowchart for ip_diffim. David worked on Task-level things and ZOGY but didn't do anything with A&L implementation (which was written in C++ by Becker during R&D). Looking at old Winter13/14 performance reports. Learned about how to deal with debug outputs. Now understands the context of the refactor.
- Metrics & ap_verify (Krzysztof Findeisen ):
- working on un-prototyping the prototype code; documentation enxt.
- AOB/table round
- next week is a holiday–no meeting
- next week is a holiday–no meeting