General questions

1. Is grizli using the same grismconf as Nor's code (e.g. does it import https://github.com/npirzkal/GRISMCONF)?
   1. Ans: It is using an earlier version, still based on the aXe approach (detailed in the 2014 TIR)
2. How hard might it be to segregate all FITS and FITS-WCS dependencies into separate I/O routines (as a step towards moving over to ASDF + gWCS)?
   1. Ans: FITS dependencies are probably pretty widespread in the code
3. Is EAZY a strict dependency? 
   1. Ans: This is probably less intricately woven into the code than FITS
4. Looks like it's using pysynphot instead of synphot_refactor. Have you looked at updating?
   1. Ans: Yes; just using this for computing fluxes in direct images, so may not be a big deal to update

Model based extraction

1. How does grizli specify parameters to vary?
2. What is the merit-function for the fit? 
   1. Ans: chisq 
3. How are the data weighted ?
   1. Ans: Standard lsq. But can add in additional down weighting for contamination
4. What optimizer is used?
5. How are uncertainties calculated?
   1. Ans: detector noise model (straightforward because we are in FLT frames)
   2. Drawback – can be slow if there are many exposures (e.g. FIGS)
6. Which modules/functions/classes are doing the fitting?
   1. Most of the work is one in multifit

Ans:

• Turns input 1D spectrum & flux into 2D dispersed using object morphology
• LSQ fitting scaling the models (template library)  
• multifit is doing the chisq minimization
• Current approach – generate static contamination model  (first starting with flat spectrum, then polynomials)
• Start with brightest object, subtract fainter ones, fit polynomial, then iterate to fainter objects
• For now, stop there for contamination – could imagine going to the next step of iterating all the fitted spectra
• Possible to simultaneously fit in cutouts with multiple objects 

GrismDisperser

• In model.GrismDisperser, do compute_model and compute_model_psf correspond to pixel- and object-based dispersion, respectively?
  
  • Ans: the compute_model_psf is a module to use an "effective psf" to estimate the flux in each pixel in the dispersed image 

model.py

• Seems to have a lot of hard-coded instrument specifics. Is that common elsewhere in the code?
  • Ans: probably mostly confined to grismconf and model.py
  • maybe also some in multifit

Approach for JWST – just rotated 90 degrees the cross-dispersed images.