How to extract the parameter from the fitted Gaussian function to pullplot in RooMCStudy?

Roofit and RooStats
1

Dear Experts,

I am doing some toy mc study using RooMCStudy module. I can do the pull plot for signal and background and fit it with gaussian using the plotPull() function and display the fitted parameter (mean and sigma) on the plot. But I want to retrieve the mean and sigma inside the root macro and save it in the txt file. Is there any easy way how I could do that? Because I am doing so many pulls for different signal and background hypotheses (and also different mass hypotheses) it would be easier for me if I can retrieve them and save it in txt file. Any help into this matter will be highly appreciated.

Root version: 6.21
Os : MacOs

Best Regards
Rajesh

2

Dear all,

Any help into this topic will be highly appreciated.

Thanks
Rajesh

3

Perhaps @jonas can help

4

Hi @RAJESH_KUMAR_MAITI, sorry for the very late reply!

There is no way to get the numerical values of the pull distribution fit parameters when you use RooMCStudy::plotPull. But you can copy-paste the relevant code from said function where the Gaussian fit is done, and write your own function that returns the pull distribution fit results. It could look like this:

std::pair<double,double> fitMCStudyPulls(
        RooMCStudy& mcStudy, RooAbsReal const& param) {

  using namespace RooFit;

  auto& fitParData = const_cast<RooDataSet&>(mcStudy.fitParDataSet());
  auto pullName = std::string(param.GetName()) + "pull";
  auto pullTitle = std::string(param.GetTitle()) + " Pull";
  RooRealVar pvar(pullName.c_str(), pullTitle.c_str(), -100, 100) ;
  pvar.setBins(100) ;
  RooRealVar pullMean("pullMean","Mean of pull",0,-10,10) ;
  RooRealVar pullSigma("pullSigma","Width of pull",1,0.1,5) ;
  RooGenericPdf pullGauss("pullGauss","Gaussian of pull",
                          "exp(-0.5*(@0-@1)*(@0-@1)/(@2*@2))",
                          RooArgSet(pvar,pullMean,pullSigma)) ;
  pullGauss.fitTo(fitParData, Minos(0), PrintLevel(-1)) ;
  return {pullMean.getVal(), pullSigma.getVal()};
  // Or have some sort of output structure where you can
  // also return the uncertainty if you want. You can get them like
  // pullMean.getError(), pullSigma.getError()
};

You would use it like this:

   auto pullFitParams = fitMCStudyPulls(*mcstudy, mean);
   std::cout << "Pull mean: " << pullFitParams.first << std::endl;
   std::cout << "Pull sigma: " << pullFitParams.second << std::endl;

Instead of printing the results, you can dump them into a text file or whatever you like.

I hope this helps! Please let me know if something is unclear or if you have any further questions.

Cheers,
Jonas

5

Hi @jonas

Thanks for your reply.

Yes I understood this. Actually I Did something more simpler, I came to know from RooMCStudy module (line 922 - 924) that, this all pull related variables are saved in fitParDataSet(). I mean if one would like to calculate pull for some variable sig, then corresponding pull variable is “sig_pull”. So one could just do, TH1 *frame1 = mcstudy->fitParDataSet().createHistogram(“sig_pull”); frame1->Fit(“gaus”); frame1->GetFunction(“gaus”)->GetParameter(1); frame1->GetFunction(“gaus”)->GetParameter(2);

Is that okay?

Thank you,
Regards
Rajesh

6

Hi @RAJESH_KUMAR_MAITI,

sure, you can do that too. But keep in mind that since you created a histogram, you will now make a binned fit of the pull means instead of an unbinned fit with the RooDataSet directly like I proposed. So your results might be different from what is shown on the plot.

I would argue that the unbinned fit is better here, since it uses more information. If you stay with the binned fit, I would still manually verify that the binning that is used for the histogram makes sense.

Cheers,
Jonas

7

Hi @jonas

Ah, this is good point, I missed that. Yes I will convert to unbinned one (as my all of fittings are unbinned).

Thank you.

Cheers
Rajesh