Fix Version/s: None
As noted in
DM-9535, the method used to compute second derivatives in the Brighter-Fatter correction implementation can lead to significant (up to 1%, which is particularly significant for weak lensing considerations) differences in the trace radii of sources, with the differences becoming larger with brighter magnitudes (see figures in DM-9535). This ticket is to investigate the optimal algorithm to use for the Brighter-Fatter correction.
DM-9870 Release hscPipe 5.0-beta1
Just for clarity: we actually have several months of time set aside in the long term plan for figuring out how to handle brighter-fatter. My assumption (and Lauren MacArthur, please correct me if I'm wrong) is that this ticket is not supposed to capture all that effort, but simply to figure out what the best approach is in the short term (ie, on the timescale of the next HSC software release).
Indeed. The algorithm selected on this ticket should result in behaviour that is at least as good as that seen with the HSC stack. That may end up being a straight adoption of what is there now, but some effort will me made (likely by Bob Armstrong) to homogenize the method for computing the derivative between the 1st & 2nd (they currently use different methods).
According to the discussion at the HSC call of 207-03-14, Bob Armstrong will not make an effort to homogenize methods (as described by Lauren MacArthur above), but will simply revert to the old (HSC) way of doing things.
That's fine in the short term, but we need to figure out whether there's a larger problem here that we need to address in the intermediate term (based on the assumption that the whole system needs revising in the long term). Bob Armstrong, can you comment on that please?
Yes, this was seen as a short term fix only since we would like to use the older code for the next HSC processing. We certainly still need to investigate why the newer version is not working and decide how to fix it.
Ok — could you please create the follow-up ticket when you're working on this? Thanks!
Give me a sec to rebase and I'll push the branch I was using with the update (revert!).
Bob Armstrong — Lauren MacArthur suggests that we assign this ticket to her, since she already has a fix on that branch, and have you act as reviewer. If you object, do shout!
Bob Armstrong, do you have time to review this?
Thanks Bob Armstrong. I ran every combination of Jenkins and it passed all it currently can, i.e. through ci_hsc on all py2 os' (centos6&7 and OS X), but only lsst_distrib on py3 (since ci_hsc hasn't been py3'd yet).
I don't think this ticket is immediately critical for LSST (although it certainly needs resolving), but it likely will be for future HSC data releases — we'd better make sure Robert Lupton is aware of it.