I have a binned dataset that I want to fit using a binned PDF. For this I created my dataset importing the TH2D as a RooDataHist, and I tried to implement a RooAbsPDF defining the evaluate() function. However I seem unable to force the RooAbsPDF sampling to go over the bins when I try to plot on one of the observables. First it took too long, then I saw that it was doing numerical integration to get the integral and normalise the PDF, but it was sampling too many points.
Then I tried to force it binned by overriding:
virtual Bool_t isBinnedDistribution(const RooArgSet&) const override { return kTRUE; }
From that moment onward it started to use the roobinintegrator and loop over only a few values but still more than it should, so not only bin centers. My RooRealVars have a binned set beforehand, but is seems it doesn´t matter.
I think I cannot use a RooHistPDF because my event prediction depends on physics parameters I want to fit but the prediction is already binned. In the “evaluate” function I just check on what bin corresponds to the value it is sampling from my RooRealVars and then I call my prediction function over the corresponding bin.
I tried to search for a similar case in the examples and the manual but I couldn´t find a similar use case in which the PDF is binned and depends on parameters. If some presentation / documentation / example is available that I overlooked it please let me know.