I believe the OptAtmoPSF can be divided into two separate pieces:
- Fitting optics parameters and simultaneously fitting tied atmospheric parameters for every star.
- Independently fitting atmospheric parameters for every star while holding the optics parameters fixed.
I can imagine adding additional steps too: For example, iterating between fitting optics and fitting atmospheric parameters.
The HSC PSF model presented by Claire Saunders on the PSF TF telecon also has a sequence structure:
- Fit parameters of Moffat profile.
- Fit pixel grid model to residuals.
If we're being really ambitious, then I could also imagine following the OptAtmoPSF model with a PixelGrid over the residuals.
In all cases, I think the intermediate steps form valid PSF objects. This makes me wonder if we ought to develop a SequencePSF abstraction that would take an ordered list of other PSF objects as input, plus some kind of mapping to tell what parameters of PSF2 come from PSF1, and apply those PSFs in order. This may also help break OptAtmoPSF down into smaller pieces.
I believe the
OptAtmoPSFcan be divided into two separate pieces:I can imagine adding additional steps too: For example, iterating between fitting optics and fitting atmospheric parameters.
The HSC PSF model presented by Claire Saunders on the PSF TF telecon also has a sequence structure:
If we're being really ambitious, then I could also imagine following the
OptAtmoPSFmodel with a PixelGrid over the residuals.In all cases, I think the intermediate steps form valid
PSFobjects. This makes me wonder if we ought to develop aSequencePSFabstraction that would take an ordered list of otherPSFobjects as input, plus some kind of mapping to tell what parameters of PSF2 come from PSF1, and apply those PSFs in order. This may also help breakOptAtmoPSFdown into smaller pieces.