esheldon
Pass on ignore_zero_weight to medsreaders. Fill in AM missing pixels
This PR has two parts.
The first is to expose the ignore_zero_weight to the medsreaders, so we can create Observations with zero weights or uberseg with these zero weight pixels in the pixels array. This faciliates filling in the pixels in AM
Second is to 1) calculate a flux for the current AM model and 2) fill in zero weight pixels using the properly normalized model.
I also am now using this flux/flux error and flags in the admom result
Filing only occurs when there are zero weight pixels, which is not the default for Observations. No examples in the /examples director showed using it.
Anyone using admom with masked data in the past was getting a biased result.
Right but I want to try symmetrized zero-weight pixels. I can't with this PR.
Yes you can, just construct the Observation with the default ignore_zero_weight=True
This results in an extra evaluation of the model over the pixels during each iteration, so is sub-optimal. I think we can probably eliminate some of the passes by pre-evaluating the model.
But I first want to test how this works in some simulations before optimizing.