DM-32355: Develop function and unit test to remove Poisson contribution from sources in variance plane

[MorganSchmitz]

No description provided.

[beckermr]

What is the status of this one?

[MorganSchmitz]

The code itself is ready (it's just a very simple free function)
Iirc, I never got it to review because I ran into trouble trying to make unit tests for it, which had to do with creating an Exposure with several Amplifiers (see this slack thread)

and then I guess it completely slipped my mind, sorry

[TallJimbo]

Looks good!

In addition to the line comments, I'd like to make sure this is tested to robustly do nothing (or nearly nothing) when given an image with no signal, rather than (say) subtracting arbitrary constants.

I'm also interested in the use case of doing this on coadds, if that's even possible - I'd love for someone to work through the algebra to see if:

• there is an "effective gain" on the coadd that is a weighted sum of images with different gains;
• if not, what the error is when running this algorithm on such a coadd if (say) those gains are drawn from a Gaussian distribution with some mean and standard deviation.

[enourbakhsh]

Looks good!

In addition to the line comments, I'd like to make sure this is tested to robustly do nothing (or nearly nothing) when given an image with no signal, rather than (say) subtracting arbitrary constants.

I'm also interested in the use case of doing this on coadds, if that's even possible - I'd love for someone to work through the algebra to see if:

• there is an "effective gain" on the coadd that is a weighted sum of images with different gains;
• if not, what the error is when running this algorithm on such a coadd if (say) those gains are drawn from a Gaussian distribution with some mean and standard deviation.

I moved the unit test to ip_isr to avoid circular dependency. As requested, I added a test for when the image given has no signal to ensure nearly the same variance plane post-correction. Regarding its applicability to coadds, I've opened a new ticket linked to this one.