RooHistPdf and Variable Binning

juliasilva · March 17, 2025, 5:56pm

Hi experts!
For the first time I am trying to use RooHistPdf with variable binning. Is this supported?
When I plot the RooHistPdf I created from my variable binning histogram, the plot is wrong.
Is there some documentation that could help me on this?
Cheers,
Júlia

bellenot · March 17, 2025, 6:47pm

Did you try to search the forum? Search results for 'RooHistPdf variable binning' - ROOT Forum.
(for example RooPlot normalization with variable binning)

jonas · March 18, 2025, 8:32am

Hi @juliasilva!

I understand that the RooHistPdf might be confusing with variable binning. There was recently a discussion about this on GitHub:

github.com/root-project/root

[RF] RooDataHist from a variable binning TH1D: different shape, different bin content

opened 10:41AM - 13 Jul 24 UTC

closed 10:51AM - 15 Feb 25 UTC

santocch

in:RooFit/RooStats

### Check duplicate issues. - [ ] Checked for duplicates ### Description … This problem has been already mentioned here [0] but no solution has been proposed. When you compare a RooDataHist obtained from a TH1 with fixed size binning everything looks ok but when you do the same from a variable size binning the comparison fails. See code below to reproduce the comparison. [0] https://root-forum.cern.ch/t/roodatahist-and-th1d-why-different-shape-bin-content/56971 ### Reproducer just run root mytest.cpp ``` using namespace RooFit; void mytest() { const int nBin=100; double xbins[nBin+1]; const float left_Val = 150; const float rightVal = 1150; const float logxmin = TMath::Log10(left_Val); const float logxmax = TMath::Log10(rightVal); const double dxLog = (logxmax-logxmin)/nBin; for (int i=0;i<=nBin;i++) xbins[i] = TMath::Power(10., logxmin + i * dxLog); TF1 *f1 = new TF1("f1","expo(0)",left_Val,rightVal); f1->SetParameters(1e2,-1e-2); TH1D *h1 = new TH1D("h1","",nBin,xbins); h1->FillRandom("f1",10000000); RooRealVar mInv("mInv","m [GeV]",150,1150); RooDataHist hist_test1("hist_test1","hist_test1",mInv,Import(*h1,false)); TCanvas *cExpo1 = new TCanvas(); gPad->SetLogx(); gPad->SetLogy(); auto plot_test1 = mInv.frame(); hist_test1.plotOn(plot_test1,DataError(RooAbsData::SumW2)); plot_test1->Draw(); h1->Draw("same"); TH1D *h2 = new TH1D("h2","",nBin,left_Val,rightVal); h2->FillRandom("f1",10000000); RooDataHist hist_test2("hist_test2","hist_test2",mInv,Import(*h2,false)); TCanvas *cExpo2 = new TCanvas(); gPad->SetLogx(); gPad->SetLogy(); auto plot_test2 = mInv.frame(); hist_test2.plotOn(plot_test2,DataError(RooAbsData::SumW2)); plot_test2->Draw(); h2->Draw("same"); } ``` ### ROOT version ROOT 6.30/06 ### Installation method homebrew ### Operating system OSX Sonoma 14.5 ### Additional context _No response_

Maybe you were confused by the same thing, i.e. that the RooHistPdf divides the histogram counts by bin widths to get a probability density? And this is what makes you think the plot is wrong?

Cheers,
Jonas

juliasilva · March 18, 2025, 12:21pm

Hi!
Thanks for pointing these out to me.
So does this mean that it should be merely a plotting issue? In principle performing a fit to a variable binning RooDataHist with a variable binning RooHistPdf should work within RooFit’s technical capabilities?
Cheers,
Júlia

system · April 1, 2025, 12:22pm

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.