Gaussian mixtures model about piff

[esheldon]

I didn't suggest mixtures, but when I saw people were wanting to do gaussian/multi-gaussian fits I thought ngmix is a natural choice.

I see gaussian_model.py; Is the idea to replace that? And add an additional gmix_model.py or something like that?

[rmjarvis]

I think we would want an additional model class that does Gaussian Mixtures, not a replacement. I not exactly sure how the interpolation would go with a gmix model. Unless the multiple Gaussians are fixed in their relative positions, it might not make much sense to interpolate the parameters between two stars.

Maybe have a first pass that tries to figure out a good number of Gaussians and their relative positions, and then a second pass that forces that configuration on all the stars?

[esheldon]

I agree it is not clear how interpolation would work. And fixing the relative locations would possibly remove any advantage of using the mixture model.

[esheldon]

For the above reasons, I think we should close this and not bother with the gaussian mixtures.

[rmjarvis]

Sure. If we think of a way to use it, we can certainly reopen or make a new issue at that time.