Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: pipe_tasks
-
Labels:
-
Team:External
Description
The current default PSF-based image selection leaves large holes in the distribution of coadds. Takashi Hamana recommends adopting:
max_medianE = 0.007
max_scatterSize/medianSize^2 = 0.009
This involves modifying the current defaults and adding a new selection criterion.
Attachments
Activity
If we want PsfWcsSelectImagesTask to be the default for obs_subaru (as opposed to WcsSelectImagesTask), I believe the following needs to be added to obs_subaru/config/safeClipAssembleCoadd.py since that is the config override file that gets read when running assembleCoadd.py:
from lsst.pipe.tasks.selectImages import PsfWcsSelectImagesTask
|
config.select.retarget(PsfWcsSelectImagesTask)
|
I suppose this could affect ci_hsc as I believe it runs the command lines separately, (i.e. does not use the driver scripts which do have the correct retargeting in the config override file). Hhhmmm...looking more closely at obs_subaru/config/coaddDriver.py, it loads the following config files:
for sub in ("makeCoaddTempExp", "backgroundReference", "assembleCoadd", "processCoadd"):
|
First, backgroundReference and processCoadd don't exist in obs_subaru/config (they do in obs_subaru/config/suprimecam). Should they be removed from the list?
Second, should assembleCoadd ==> safeClipAssembleCoadd?
I realize this isn't really part of this ticket so feel free to punt, but you are probably the authority on this anyway!
Cleaning up the coadd configs is DM-10634. I'd love to be able to put that off until the warping and coadding is refactored....
Paul Price
added a comment - Cleaning up the coadd configs is DM-10634 . I'd love to be able to put that off until the warping and coadding is refactored.... Ok, thanks. I will point out that this means that the PsfWcsSelectImagesTask is not exercised anywhere in CI as there is no test for it and ci_hsc does not use it. This seems a bit dangerous...what are you running to ensure the appropriate CCDs are getting omitted with this change (I assume the next HSC run will be scrutinized, but a sanity check ahead of a production run is never a bad idea)?
Furusawa-san sent me a list of visits for a couple of tracts, and pointed out which get included and rejected with the new cuts. I'm working on that now.
Paul Price
added a comment - Furusawa-san sent me a list of visits for a couple of tracts, and pointed out which get included and rejected with the new cuts. I'm working on that now. Based on the testing against the list Furusawa-san sent, I'm rejecting CCDs that should be rejected. There was one instance where we reject a CCD that he didn't, but I think it's because of some operational difference that I don't currently understand (our medianE values for that CCD are very different). Nevertheless, the code works (though most images are rejected on the basis of the medianE cut, not the scaled size scatter).
I believe I've addressed all the comments on GitHub, and so am about to merge...
Paul Price
added a comment - Based on the testing against the list Furusawa-san sent, I'm rejecting CCDs that should be rejected. There was one instance where we reject a CCD that he didn't, but I think it's because of some operational difference that I don't currently understand (our medianE values for that CCD are very different). Nevertheless, the code works (though most images are rejected on the basis of the medianE cut, not the scaled size scatter).
I believe I've addressed all the comments on GitHub, and so am about to merge... Merged to master.
Thanks for the useful suggestions, Lauren!
Paul Price
added a comment - Merged to master.
Thanks for the useful suggestions, Lauren! 
Thanks for volunteering to look at this, Lauren MacArthur!
price@pap-laptop:~/LSST/pipe_tasks (tickets/DM-11243=) $ git subcommit 3ed9f4e630839bff682865941b9ca0660abee751Author: Paul Price <price@astro.princeton.edu>Date: Mon Jul 10 12:19:40 2017 -0400PsfWcsSelectImagesTask: make thresholds optionalWe may decide not to use a particular threshold, so they should beoptional.python/lsst/pipe/tasks/selectImages.py | 10 ++++++----1 file changed, 6 insertions(+), 4 deletions(-)commit 2634f9e6df4bbf6147d5a8520b9f1a0cc7b36db5Author: Paul Price <price@astro.princeton.edu>Date: Mon Jul 10 13:10:55 2017 -0400PsfWcsSelectImagesTask: add selection on scaled size scatterTakashi Hamana recommends a selection involving:max_scatterSize/medianSize^2 = 0.009python/lsst/pipe/tasks/selectImages.py | 12 ++++++++++++1 file changed, 12 insertions(+)commit d133c9b1cbdd5015f00d8ca9c7c79d390716787fAuthor: Paul Price <price@astro.princeton.edu>Date: Mon Jul 10 13:14:02 2017 -0400PsfWcsSelectImagesTask: adopt Hamana-san's recommended limitsTakashi Hamana recommends:max_medianE = 0.007max_scatterSize/medianSize^2 = 0.009We update the maxEllipResidual to correspond to Hamana-san'srecommendation and remove the maxSizeScatter limit (maxScaledSizeScatteris already set appropriately).python/lsst/pipe/tasks/selectImages.py | 3 +--1 file changed, 1 insertion(+), 2 deletions(-)