Asymmetry plots with Roofit

[sanjeeda]

Dear Experts,

I have the working code for a weighted 2D simultaneous PDF. It has 2 mass variables which I fit and two separate datasets corresponding to the D0 and D0bar samples. How do I get the asymmetry of as a function of the two mass variables from the D0 and D0bar PDFs corresponding to each mass variable.

I can do it for histograms using TH1::GetAsymmetry but I am not able to do it in the Roofit code.
Here’s the code that I am using:asym_proj_forum.zip (10.4 KB) and the root file is at the link:

component_wise_trees_for_fit_kk_belle_mc_4Sonly.root

Will you please have a look.

[jonas]

Hi @sanjeeda,

very nice fits you have there, and an interesting analysis! I should take your code as an example of 2D analysis numer numeric integrals that we should optimize to run faster, your script takes quite some time to run Also, the numeric integrals have a hard time converging, maybe we can do something about this. Is this kind of analysis you are doing common in your collaboration?

As for the actual question: is there a reason why this wouldn’t work with a RooFormulaVar?

[sanjeeda]

Thank you @Jonas. I think such analyses are common in any experimental high energy physics collaborations.

It looks there is a way if doing it using arguments in RooSimultaneous. One can search “Asymmetry” in ROOT: RooSimultaneous Class Reference. am tryin to include the argument like:
simPdf.plotOn(kk_d0,Slice(sample,"D0"),Components(RooArgSet(model_d0)),ProjWData(sample,combData),Name("model_d0"), Asymmetry(sample));, but I am not able to get what I want. I want to have something like:

I am using Root 6.28.

[jonas]

I have some suggestions to massively speed up your script, in case this is helpful:

• in fitTo(), use our new , EvalBackend("cpu") (or BatchMode(true), as it was called prior to ROOT 6.30) See also the documentation for RooAbsPdf::fitTo()
• Restrict your RooDstD0BG parameter ranges a bit, such that you don’t get inf integrals:

  RooRealVar kappa("C", "C", 1.59953e-03,-1e4,1e4);
  RooRealVar alpha("A", "A", 3.75892e+01,0,1e3);
  

Also, use the Parameters() command arg in your call to paramOn:

simPdf.paramOn(d0pi_d0bar,Parameters(*dataxy_d0bar->get()));

The signature that directly takes the RooDataSet was deprecated for a long time and was removed in in 6.30.

[jonas]

Thanks for pointing out the Asymmetry() command argument! I was not aware of it actually.

So according to the documentation, your sample category must have the index values +1 and -1 for it to work:

  RooCategory sample("sample","sample", {{"D0", -1}, {"D0bar", +1}}) ;

So do this instead of the defineType() calls.

But that doesn’t seem to solve the problem

When I try to plot the asymmetry, I get this error:

[#0] ERROR:Plotting -- RooAbsReal::plotAsymOn(simPdf) function doesn't depend on asymmetry category sample

I suspect the Asymmetry() command doesn’t play well with the ProjWData(), which you of course need to get the 1D plots I will try to reproduce this problem in a simple toy macro and then go on fixing it.

[sanjeeda]

Thank you @jonas.
What I did was:

RooCategory sample("sample","sample") ;
sample.defineType("D0",1);
 sample.defineType("D0bar",-1);

Then you can have
simPdf.plotOn(kk_d0,Slice(sample,"D0"),Components(RooArgSet(model_d0)),ProjWData(sample,combData),Name("model_d0"), Asymmetry(sample));

The code will run with this, but I don’t get the asymmetry plots as I have shown as an example.

[jonas]

Yes, I understand. Apparently the Asymmetry() plot for RooSimultaneous is simply broken. Here is a reproducer:

void repro()
{
   using namespace RooFit;

   RooWorkspace ws;
   ws.factory("Gaussian::gauss_A(x_A[-10, 10], -1.0, 1.0)");
   ws.factory("Gaussian::gauss_B(x_B[-10, 10], +1.0, 1.0)");
   ws.factory("ExtendPdf::pdf_A(gauss_A, n_A[10000.])");
   ws.factory("ExtendPdf::pdf_B(gauss_B, n_B[10000.])");
   ws.factory("SIMUL::simPdf(sample[A=-1, B=+1], A=pdf_A, B=pdf_B)");

   RooRealVar &xA = *ws.var("x_A");
   RooRealVar &xB = *ws.var("x_B");
   RooCategory &sample = *ws.cat("sample");

   std::unique_ptr<RooDataSet> data{ws.pdf("simPdf")->generate({xA, xB, sample})};

   auto frame = ws.var("x_A")->frame();
   ws.pdf("simPdf")->plotOn(frame, Asymmetry(sample), ProjWData(sample, *data));

   frame->Draw();
}

Also here, the output is:

[#1] INFO:Plotting -- RooSimultaneous::plotOn(simPdf) plot on x_A represents a slice in the index category (sample)
[#0] ERROR:Plotting -- RooAbsReal::plotAsymOn(pdf_A) function doesn't depend on asymmetry category sample

If you want us to fix this, can you please open a GitHub issue that includes my reproducer? It’s better if you open the issue instead of me, because if a user opens an issue instead of a dev, it’s clear that this is a priority. That helps with the planning.

Thanks!
Jonas

[sanjeeda]

Dear @jonas,

Thank you for taking the time to look into this. I have opened the issue Asymmetry plots with Roofit · Issue #14255 · root-project/root · GitHub.
Please let me know is anything is missing.

Gratefully,
Sanjeeda

[sanjeeda]

Dear @jonas, is there a way to get a TH1D histogram corresponding to the PDFs that are used for the fit. If yes, the, one can find the asymmetry between the two ?