Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-31388

Investigate low fake injection completeness in HSC COSMOS

    XMLWordPrintable

Details

    • Story
    • Status: Done
    • Resolution: Done
    • None
    • None

    Description

      In DM-31340 we see a premature falloff of injected fakes before reaching the six-sigma limiting magnitude. This ticket is to investigate why we are losing the fainter injected DIASources.

      Attachments

        Issue Links

          Activity

            ebellm Eric Bellm added a comment -

            Suggestion for followup at the QA meeting (https://confluence.lsstcorp.org/display/DM/AP+Pipeline+QA+Meeting%2C+2021-09-13) included (in rough order of priority)

            • use sky sources in the diffim to get a more accurate estimate of the limiting magnitude
            • plot number counts of direct image sources and look for turnover
            • check if fakes are being injected on masked regions, or otherwise bad data?
              → make scatter plots of unmasked fakes
            • maybe the template doesn't have enough coverage and doesn’t have enough data to pick the variance plane scale factor right? Current config val is 0.1
            ebellm Eric Bellm added a comment - Suggestion for followup at the QA meeting ( https://confluence.lsstcorp.org/display/DM/AP+Pipeline+QA+Meeting%2C+2021-09-13 ) included (in rough order of priority) use sky sources in the diffim to get a more accurate estimate of the limiting magnitude plot number counts of direct image sources and look for turnover check if fakes are being injected on masked regions, or otherwise bad data? → make scatter plots of unmasked fakes maybe the template doesn't have enough coverage and doesn’t have enough data to pick the variance plane scale factor right? Current config val is 0.1
            ebellm Eric Bellm added a comment -

            We also see this on HiTS in DM-32245. It appears that some of the injected sources in the matchedFakesCatalogs are outside the physical bbox, as in DM-30030. Prefiltering in Ra/Dec space before applying the bbox in https://github.com/lsst/pipe_tasks/blob/aca5567d74a05fb166ef81240f5dd9cf5a539e7c/python/lsst/pipe/tasks/matchFakes.py#L151 should fix this, and should provide some improvement in completeness.

            ebellm Eric Bellm added a comment - We also see this on HiTS in DM-32245 . It appears that some of the injected sources in the matchedFakesCatalogs are outside the physical bbox, as in DM-30030 . Prefiltering in Ra/Dec space before applying the bbox in https://github.com/lsst/pipe_tasks/blob/aca5567d74a05fb166ef81240f5dd9cf5a539e7c/python/lsst/pipe/tasks/matchFakes.py#L151 should fix this, and should provide some improvement in completeness.
            ebellm Eric Bellm added a comment -

            I note that the switch in DM-32988 to multi-tract difference imaging led to big changes in the CI fakes completeness metrics for both HSC and DECam, which may be related.

            ebellm Eric Bellm added a comment - I note that the switch in DM-32988 to multi-tract difference imaging led to big changes in the CI fakes completeness metrics for both HSC and DECam, which may be related.
            mrawls Meredith Rawls added a comment - - edited

            It appears that the RA/Dec trimming we added makes the fake completeness a bit better for DECam (and fixes the ridiculous "fakes are everywhere outside the visit" problem), but I see no appreciable change in HSC. More investigation is needed. But, the changes to the fakes insertion and trimming are definitely needed to make DECam fakes not ridiculous, so this was still useful.

            The completeness plots I attached to the ticket show the small but real improvement for DECam and basically no change for HSC.

            mrawls Meredith Rawls added a comment - - edited It appears that the RA/Dec trimming we added makes the fake completeness a bit better for DECam (and fixes the ridiculous "fakes are everywhere outside the visit" problem), but I see no appreciable change in HSC. More investigation is needed. But, the changes to the fakes insertion and trimming are definitely needed to make DECam fakes not ridiculous, so this was still useful. The completeness plots I attached to the ticket show the small but real improvement for DECam and basically no change for HSC.
            mrawls Meredith Rawls added a comment - - edited

            Jenkins passed, and I investigated more thoroughly to conclude (1) DECam completeness is definitely improved and (2) HSC completeness is definitely not worse. Putting this in review for the fakes trimming portion, which appears to duplicate DM-32835. I will file a new ticket for further investigation of how HSC fakes completeness still isn't as good as we would like.

            This is the "new" HSC completeness plot, which is basically identical to DM-31340 completeness plots. 

            mrawls Meredith Rawls added a comment - - edited Jenkins passed, and I investigated more thoroughly to conclude (1) DECam completeness is definitely improved and (2) HSC completeness is definitely not worse. Putting this in review for the fakes trimming portion, which appears to duplicate DM-32835 . I will file a new ticket for further investigation of how HSC fakes completeness still isn't as good as we would like. This is the "new" HSC completeness plot, which is basically identical to DM-31340 completeness plots. 

            Made some changes to insertFakes during pair coding, specifically trimming fakes in insertFakes to the region plus a buffer, and doing it in RA/Dec space. sophiereed, per discussion, please test we didn't do anything horrible as part of your review, and then I can rerun Jenkins once we're 99% sure we're happy.

            mrawls Meredith Rawls added a comment - Made some changes to insertFakes during pair coding, specifically trimming fakes in insertFakes to the region plus a buffer, and doing it in RA/Dec space. sophiereed , per discussion, please test we didn't do anything horrible as part of your review, and then I can rerun Jenkins once we're 99% sure we're happy.

            Jenkins passed, and this now uses the awesome new containsSkyCoords exposure method from DM-33591!

            mrawls Meredith Rawls added a comment - Jenkins passed, and this now uses the awesome new containsSkyCoords exposure method from DM-33591 !

            People

              mrawls Meredith Rawls
              ebellm Eric Bellm
              Sophie Reed
              Eric Bellm, Meredith Rawls, Sophie Reed
              Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved:

                Jenkins

                  No builds found.