Use histogram as both prior and input data?

[Moelf]

as of:

• 1D-Histograms as prior distributions #70

we can use histogram as prior, in HEP (as @oschulz knows), the input data is also histogram (with same binning of course), and each uncertainty variation as one additional histogram.

Can we run likelihood for parameter extraction in such case?

[oschulz]

Sure, that should work! EmpiricalDistributions.jl turns histograms into (stepwise, no KDE capability and gradient yet, so not compatible with HMC) distributions, which can just be plugged into a VariableShapes.NamedTupleDist(a = ..., b = ...) as components of the prior.

The likelihood can use an arbitrary log-likelihood function, that has to be written, of course.

each uncertainty variation as one additional histogram.

The uncertainty is on the expectation?