I am not sure if this is right place to request additional features. If not, please point me a right place.
Now ModelConfig is nicely handling parameters for parameter of interest, nuisance parameters and global parameters, but entirely misses another category: normalization factors. Many analyses are introducing unconstrained normalization factors to scale the number of background events and perform the fitting to obtained these normalization factors. But these factors are not the parameter of interest nor being constrained.
I propose in the ModelConfig to add a category to label normalization factors so that they could be treated differently.
However, I don’t think these normalization factors should be part of the model config. The ModelConfig stores information on how to setup your likelihoods for a general model. But these normalization factors are specific to one kind of model and a specific usecase.
If we would introduce entries in the model context for all subsets of parameters that fulfill a certain function in a certain model, then it would expand forever. So I think this should not be addressed.
What would be better for your usecase is to store these normalization factors as a named RooArgSet in your workspace with defineSet(), such that you can easily retrieve them with set().
I think like this you can also track your normalization factors just as easily as with the ModelConfig. It’s just in the workspace directly (so might be even easier).