Flux maps combination

[QRemy]

Create a new FluxMaps by combining a list of flux maps with the same geometry.
Under the gaussian error approximation the likelihood is given by the gaussian distribution. The product of gaussians is also a gaussian so can derive dnde, dnde_err, and ts for the combined map.

[codecov]

Codecov Report

Attention: Patch coverage is 98.63014% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 94.47%. Comparing base (a55f23c) to head (3f54576).
Report is 1 commits behind head on main.

Files	Patch %	Lines
gammapy/estimators/utils.py	97.82%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5161   +/-   ##
=======================================
  Coverage   94.46%   94.47%           
=======================================
  Files         235      235           
  Lines       35493    35566   +73     
=======================================
+ Hits        33528    33600   +72     
- Misses       1965     1966    +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[bkhelifi]

Hello @QRemy,
Thanks a lot for this nice utility function. I have two questions:

• maybe one could add a check and a test of maps with different geometries. This might happen frequently
• I am wondering whether one could not extend this function for non-gaussian errors, when the scan profile is present. Maybe one could make that later into an other PR, and add a ToDo in this one....

[QRemy]

• I am wondering whether one could not extend this function for non-gaussian errors, when the scan profile is present. Maybe one could make that later into an other PR, and add a ToDo in this one....

Yes I had that in mind but first we need to implement the stat_scan option on the ExcessMapEstimator and TSMapEstimator (at least so 2 other PRs).

[registerrier]

From the dev call, let's try to find a more explicit name for the function. It could be possibly an independent function rather than a classmethod.

[registerrier]

Thanks @QRemy. This is nice. I have left some inline comments.

A general question is whether applying this to a regular FluxMaps is safe or not. What use cases does this cover?

[registerrier]

There is a merge conflict here. Can you rebase @QRemy ?

[AtreyeeS]

Thanks @QRemy ! I have some questions inline, but it is okay from my side

[AtreyeeS]

Thanks @QRemy for the updates! No further comments

[registerrier]

Thanks @QRemy. Looks good to me.

I simply propose to call the function average_flux_maps as this is more explicit. Joint flux map would rather be applied to flux maps extracted jointly from a group of datasets.Actually maybe stack is even more clear as this is consistent with typical gammapy semantics.

[QRemy]

I simply propose to call the function average_flux_maps as this is more explicit. Joint flux map would rather be applied to flux maps extracted jointly from a group of datasets.

Maybe combine_flux_maps as we have get_combined_significance_maps. I think the term combined is often used when combining likehoods from different experiments a posteriori, especially in cosmology like in https://arxiv.org/pdf/1807.06209 .

[registerrier]

Thanks @QRemy. Looks good!