Making 2d plots of a multidimensional simultaneous PDF

Roofit and RooStats
1

Hi all,

I have a PDF to fit four dimensions and slices defined by three categories. I produce 1d plots using the following code:

cat1->setLabel("foo");
cat2->setLabel("bar");
cat3->setLabel("baz");
RooArgSet* cats = new RooArgSet(*cat1, *cat2, *cat3);

TString cuts = Form("cat1==cat1::%s && cat2==cat2::%s && cat3==cat3::%s", cat1->getLabel(), cat2->getLabel(), cat3->getLabel());

// create a binned projection dataset of the categories to speed up plotting
RooDataSet*  catData1 = new RooDataSet("catData1", "catData1", dataset, *cats);
RooDataHist* catData  = catData1->binnedClone("catData");
	
dataset->plotOn(frame, Cut(cuts));
pdf->plotOn(frame, Slice(*cats), ProjWData(*catData));

Now I want to look at 2d subsets. I did

// as before:
cat1->setLabel("foo");
cat2->setLabel("bar");
cat3->setLabel("baz");
TString cuts = Form("cat1==cat1::%s && cat2==cat2::%s && cat3==cat3::%s", cat1->getLabel(), cat2->getLabel(), cat3->getLabel());

TCanvas* c = new TCanvas("c", "Plot", 1240, 480);
c->Divide(2);

c->cd(1);
TH1* dataseth2d = dataset->createHistogram("dataset2d", *var1, Binning(25), YVar(*var2, Binning(25)), Cut(cuts));
dataseth2d->Draw("colz");

c->cd(2);
TH1* pdfh2d     = pdf->createHistogram("pdfh2d", *var1, Binning(25), YVar(*var2, Binning(25)));
pdfh2d->Scale(dataseth2d->Integral()/pdfh2d->Integral());
pdfh2d->Draw("colz");

The createHistogram() interface doesn’t support the Slice() modifier. So it came quite as a surprise, that the plots actually look reasonable! Is this the correct way to do it, that is can I trust the plots? Can somebody explain what’s going on?

  • Moritz[/code]
2

Hi Moritz,

Anything produced by createHistogram() is implicitly sliced (i.e. no integration happens over observables that are not plotted), so in your case they are indeed equivalent to the slice plots.

This is different behavior than in a RooPlot, but since it has historically grown that way I didn’t want to change that.

NB: In ROOT 5.24 you can explicitly request integration over additional observables (if you like) through the argument IntegratedObservables() and the ConditionalObservables() keyword is accepted as well.

Wouter

3

Ok, thanks!