Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: None
-
Labels:None
-
Story Points:4
-
Epic Link:
-
Team:Data Release Production
Description
Apparently coincident with the merge of DM-18643, there was a significant regression in astrometric metrics associated with validate_drp as tracked by SQuaSH. (e.g. between 2019-04-13 and 2019-04-16, AM1 went from 7.9 to 12.3 mas, AD1 from 14.6 to 119.6 mas, etc). Please investigate what happened and (if appropriate) put in a fix.
Attachments
Issue Links
- relates to
-
- Done
Activity
Here is the key plot from the fixed version on cfht. Will post hsc validation when it's done.
And here's the cfht json file, AM1 is now 7.75 mas, and AD1 is 14.6, in line with _14 prior to DM-18643. Cfht_output_r.json
Code changes look fine. Sorry for not catching this. Assuming Jenkins passes and validation_data_hsc gives acceptable results, go ahead and merge.
validate_drp on HSC looks good. AM1 is 3.65 / 7.36 / 5.80 mas in R, I, Y and AD1 is 5.77 / 8.26 / 8.82 mas.
Just catching up after travel — thanks Eli Rykoff & Chris Morrison [X] for the fast turnaround on this one!
Turns out this was a canary in the coal mine, and the situation was worse than a simple regression on this metric.
I had not realized in
DM-18643that the wcs fitter was updating the coordinates with the new wcs. By handing the wcs fitter a down-selected catalog, only some positions were getting updated (those possibly used in the astrometry fitting) while the rest of the positions were left alone. This led to (a) inconsistent positions in the persisted calexp source catalog; (b) strange validate_drp metrics because the bright well-measured sources were good, and the fainter sources (and galaxies) were not good at all.