Suggestion to remove standard_llh

[pernicet]

I have been using Jannis Flaires catalogue to calculate its sensitivity:

• dataset: ps_v004_p02
• minimisation handler: fixed_weights
• weight: bolometric fluence
• 528 sources
• Gamma = 2

The two likelihoods tested are:

1. standard_llh
2. standard_matrix

The n_s bias plots look like:

standard_llh
bias_n_s-25.pdf
standard_matrix
bias_n_s-26.pdf

Similar behaviour appears in the gamma bias plots. I also tested sub catalogues of Flaires with 100, 200, 300 and 400 sources and the behaviour it's the same: a huge deviation is observed when using the standard_llh (even with fewer sources), which it seems to disappear when using standard_matrix. Do we know why? Should we perhaps reconsider using standard_llh at all?

[JannisNe]

Thanks, yes, I think we should make standard_llh unusable. Something like #375 should work.

[pernicet]

The same behaviour is observed when I use standard_kde_enabled when using KDEs with the northern tracks sample:
bias_n_s-45.pdf
Maybe we should reconsider using this one as well.

[mlincett]

Thanks, yes, I think we should make standard_llh unusable. Something like #375 should work.

Does standard_matrix have the same performance of standard_llh for a single pointsource? I would definitely drop standard_llh for any stacking because it's more likely to screw up than actually work, but I still wonder if it provides any performance benefit for a single source.

[JannisNe]

Do you mean performance in terms of runtime? I have never tested it, so I don't have numbers, but I feel that it will not make a big difference.

[sathanas31]

In principle the standard_kde_enabled can be deleted altogether on top of removing it from compatible_llh since it's not even tested. Also, for future reference, we could merge StandardOverlappingLLH with StandardLLH since the only difference between the 2 are the create_kwargs & calculate_test_statistic methods. This is too big of a change, and the existing tests need to be modified accordingly, so I second Jannis' solution for now.