Robert Lupton – could you please add pointers to code, documentation or other relevant material on the SDSS algorithm?

Lauren MacArthur would you be able to review this ticket?

I've not done a code review on this (that's Lauren MacArthur's job) but I did quickly glance at it and I was a bit concerned about the lack of documentation. This is difficult because the deblender as a whole isn't well documented so there's nowhere obvious to add an explanation of what this new algorithm does, but I do think we need to capture more information — even if just in code comments — than has been done so far. Are there published papers (or unpublished proto-papers) we could refer to, for example?

John Swinbank beat me to it! +1 on more details (wherever they end up) about the algorithm. Perhaps a figure or two demonstrating it in action would be illuminating as well.

I was also not quite sure how to document this. There is no reference except that this procedure helped in SDSS. I am happy to put more comments in the code, but given that the deblender is to going to be completely rewritten (with documentation!) I don't know how useful these will be.

Lauren MacArthur Is there a place to put documentation with figures?

I'd be fine with a few "before" and "after" figures attached to the ticket.

I'd be fine with a few "before" and "after" figures attached to the ticket.

Hmm, I'm not sure I would. At least, I'd like to discuss in more detail what our longer term plans for the deblender and its documentation will be before agreeing to that. In particular:

1. Does the "complete rewrite" literally involve throwing away everything and starting again with completely new algorithms? If not, it seems like anything we can capture on this new algorithm now will continue to be useful;
2. While I think Lauren is asking for documentation on this ticket that helps her understand the impact of the work that has been carried out, I'd actually like to see documentation we can present to folks trying to use the deblender to get work done in the future;
3. There is some documentation in the doc/design.tex file written by Dustin. That hasn't been touched in a couple of years; I've not read it properly, I don't know if it reflects reality, and I'm not sure if it's actually in a format that we'd find useful, but we should at least consider updating it to reflect this new functionality.

Oh, I didn't mean to imply the documentation should be attached to the ticket, just that if the location of it (whether temporary or not) doesn't easily accommodate demonstrative figures, a few of the latter attached to the ticket would nice.

That sounds fine to me.

Here are a couple of examples of what the algorithm does from a list of merging galaxies in HSC that are known to have their flux underestimated as compared to SDSS. (You can find a list of these in DM-8139).

In the following pictures I have labeled

• blue "x" = peak the deblender thought was a PSF
• green "+" = peak that is not a PSF
• purple "o" = objects identified as having their template degenerate with another one.
• red "x" = object that had underestimated flux compared to SDSS

Here are a few examples of cases that look like this is helping to not shred galaxies

Here are some examples where objects were identified as having a degenerate template, but perhaps should not have been.

Looks great Bob Armstrong. A few minor comments on the PR and a request for more test coverage (which can be pushed to another ticket if you'd like).