Hello, in RooFIt we have two kind of datasets, unbinned (RooDataSet) or binned (RooDataHist).
In many analyses people are using categories, with very different number of events (~10 in some categories, ~10k in others). It would be nice to have the possibility to mix binned and unbinned data. Presenlty people are solving it using a RooDataSet to mimic the binning: one entry is added at the center of each bin with a weight equal to the number of events. I see that this approach is also using inside Roofit, for example when using genBinned or by the ToyMCSamples.
Is that ok? I guess there should be a small difference between a RooDataHist and a RooDataSet with one entry per bin, since in the computation of the likelihood should be a bit different (considering only the weighted point in the bin center, or the integral in the bin). Is it?
Is there a plan to provide a way to mix proper binned dataset (RooDataHist) and unbinned dataset in a dataset to be used in simultaneous fit?
Is there the possibility to mix binned dataset with different binnings (now people are solving it using several observables, 1 per category, with different binning, but this complicate the generation of toys. By the way why the binning is a property of the RooRealVar and not of the dataset?)?