Unbinned analysis

[israelmcmc]

General issue collecting presentations, ideas, and general steps towards an unbinned analysis:

Presentations on the topics:

• 20230817-UnbinnedCont-Israel.pdf

• 20240222-RLunbinned.pdf

  • Notebook

General steps:

• Change how we create and handle the response. Use a relative parametrization scheme.
  • MEGAlib side: update the response creator with this new binning
  • Cosipy side: response handling that interpolates and returns probabilities
• RL imaging
  • Generalize the iterative operation to handle any 2D matrix (see Major updates on the image deconvolution modules #188 )
  • Write down the matrix creator for the unbinned case
  • Parallize (see Parallel Richardson-Lucy Algorithm Implementation #189)
  • Optimize, optimize, optimize
• COSILike: code likelihood function
  • Paralellize
  • Optimize optimize optimize

[israelmcmc]

Copy-paste from #188 so it doesn't get lost:
With #188 merged, for the unbinned RL analysis, ImageDeconvolutionDataInterfaceBase._event needs to be set to 1 (a scalar is fine), calc_expectation and _bkg_models need to be calculated differentially (per phase space unit), and the likelihood needs to be the one described on slide 6 here 20230817-UnbinnedCont-Israel.pdf. The _exposure_map needs to take into account the full phase space, whether we had events or not (the normalization is decoupled from the full response matrix)