Result of using a udeep-wide trained SGS classifier on a sample of ~ 850 objects. We plot cmodel_mag v/s psf_mag - cmodel_mag and color by P(star) for each of the r & i bands. At low magnitudes (below ~ 23.5), most objects that have psf_mag - cmodel_mag ~ 0 are classified as stars (red) by the SGS classifier. At higher magnitudes, we see that while some objects with low psf_mag - cmodel_mag still end up being classified as stars (red), there is a large number of objects that are inferred to be galaxies (blue) by the SGS classifier.

Hi John,
Could you please review the work done on this ticket? The SGS code is now able to classify objects as stars or galaxies using the script classifyObjects.py added by me to the repository.

I'm going to punt on this review for now, and return to it when DM-9002 and DM-9125 are in. I think it'll be a bit clearer to see what's going on once it can be rebased cleanly against master, rather than being based on a branch which is based on a branch, ...

As I read through this, I'm really confused about why we have multiple repositories here at all. Rather than transfer individual files back and forth, I'm going to merge the entirety of sgs into sgsupervised. Makes it easier to copy code back and forth while preserving history, and I don't think it loses us anything.

I agree - sgs and sgsupervised should be merged into a single repo. Can we also make it eups manged?

Not on this ticket — please file another!

Ok, repositories merged. Closing this ticket.

Also deleting the old sgs repository.