BlueJeans: https://bluejeans.com/426716450
attending: Eric Bellm John Swinbank Meredith Rawls Unknown User (emorganson) Ian Sullivan Krzysztof FindeisenUnknown User (gkovacs)
Topics for discussion
- June false positive sprint (Eric Bellm Yusra AlSayyad Meredith Rawls Unknown User (gkovacs) )
- lots of good findings, capturing lots of followon action in tickets.
- Gabor has tried processing HiTS with instcals but there were lots of astrometry failures so it's not quite ready for any kind of significant comparison
- Gabor also trying to the calexp-calexp run
- DES processing (Unknown User (emorganson)):
- has done some profiling work:
- ran on DECam (half LSST size CCDs) on lsst-dev09, comparable to operations speeds
- total time 101 seconds (5x longer than desired, but still 5x faster than DESDM)
- convolution (smaller kernels?), fitsIO (pre-loading and post-saving could help), scipy.lmdif (tune this), meas_base wrappers (the measurement plugin code–fewer measurements?), warpExposure (smaller kernel?), makeStatistics are 50% of the runtime
- going to require lots of code changes, maybe accepting worse algorithms (e.g., smaller kernels) to get to target latencies
- or longer latencies!
- thread convolution or warping? (OpenMP) buying more processors is not a major budgetary issue relative to hard drives ((GPUs?))
- has done some profiling work:
ap_
association (Unknown User (cmorrison)):- away at DESC
- DCR (Ian Sullivan ):
- has been gone for a week and a half; tech note (DMTN-121) is in review with Eric and John.
- Review CI (https://chronograf-demo.lsst.codes/):
- Krzysztof Findeisenwill look into 0 unassociated DIASources on 10 July–looks like ProcessCcd crashed. John says there were a failure to open the templates (cFITSIO error)? Probably this should have shown up as a build failure in Jenkins, will ticket it
- PCW session review
- sessions that are high-priority for the team:
- Data Management All Hands
- Progress on QA Tooling
- Gen 3 porting in the hack session (as directed by Fritz, described in DM All Hands on day 1)
- others of interest:
- time domain day of focus
- sessions that are high-priority for the team:
- AOB
- Meredith and John are interested in understanding DESC DC2 DIA.
- we will continue to monitor slack channels etc. to see if there are algorithmic learnings emerging from DESC, and/or steal their cool plots
- KF will be giving a MetricTask presentation (& how to make it work for DRP) to Princeton Wednesday
- John is looking for volunteers to look at Tim's QA tooling
- Meredith and John are interested in understanding DESC DC2 DIA.