RooParametricStepFunction.cxx

Go to the documentation of this file.

/*****************************************************************************
 * Project: RooFit                                                           *
 * Package: RooFitBabar                                                      *
 * @(#)root/roofit:$Id$
 * Authors:                                                                  *
 *    Aaron Roodman, Stanford Linear Accelerator Center, Stanford University *
 *                                                                           *
 * Copyright (c) 2004, Stanford University. All rights reserved.        *
 *
 * Redistribution and use in source and binary forms,                        *
 * with or without modification, are permitted according to the terms        *
 * listed in LICENSE (http://roofit.sourceforge.net/license.txt)             *
 *****************************************************************************/
 
/** \class RooParametricStepFunction
    \ingroup Roofit
 
The Parametric Step Function PDF is a binned distribution whose parameters
are the heights of each bin.  This PDF was first used in BaBar's B0->pi0pi0
paper BABAR Collaboration (B. Aubert et al.) Phys.Rev.Lett.91:241801,2003.
 
This PDF may be used to describe oddly shaped distributions.  It differs
from a RooKeysPdf or a RooHistPdf in that a RooParametricStepFunction
has free parameters.  In particular, any statistical uncertainty in
sample used to model this PDF may be understood with these free parameters;
this is not possible with non-parametric PDFs.
 
The RooParametricStepFunction has Nbins-1 free parameters. Note that
the limits of the dependent variable must match the low and hi bin limits.
 
Here is an example showing how to use the RooParametricStepFunction to fit toy
data generated from a uniform distribution:
 
~~~ {.cpp}
// Define some constant parameters
const int nBins = 10;
const double xMin = 0.0;
const double xMax = 10.0;
const double binWidth = (xMax - xMin) / nBins;
const std::size_t nEvents = 10000;
 
// Fill the bin boundaries
std::vector<double> binBoundaries(nBins + 1);
 
for(int i = 0; i <= nBins; ++i) {
  binBoundaries[i] = i * binWidth;
}
 
// The RooParametricStepFunction needs a TArrayD
TArrayD binBoundariesTArr{int(binBoundaries.size()), binBoundaries.data()};
 
RooRealVar x{"x", "x", xMin, xMax};
 
// There are nBins-1 free coefficient parameters, whose sum must be <= 1.0.
// We all set them to 0.1, such that the resulting step function pdf is
// initially uniform.
RooArgList coefList;
for(int i = 0; i < nBins - 1; ++i) {
  auto name = std::string("coef_") + std::to_string(i);
  coefList.addOwned(std::make_unique<RooRealVar>(name.c_str(), name.c_str(), 0.1, 0.0, 1.0));
}
 
// Create the RooParametricStepFunction pdf
RooParametricStepFunction pdf{"pdf", "pdf", x, coefList, binBoundariesTArr, int(nBins)};
 
// Generate some toy data, following our uniform step function pdf
std::unique_ptr<RooAbsData> data{pdf.generate(x, nEvents)};
 
// Fit the step function to the toy data
pdf.fitTo(*data);
 
// Now we plot the data and the pdf, the latter should not be uniform
// anymore because the coefficients were fit to the toy data
RooPlot *xframe = x.frame(RooFit::Title("Fitting uniform toy data with a step function p.d.f."));
data->plotOn(xframe);
pdf.plotOn(xframe);
xframe->Draw();
~~~
*/
 
#include <RooParametricStepFunction.h>
 
#include <RooCurve.h>
#include <RooRealVar.h>
 
ClassImp(RooParametricStepFunction);
 
////////////////////////////////////////////////////////////////////////////////
/// Constructor
 
RooParametricStepFunction::RooParametricStepFunction(const char* name, const char* title,
              RooAbsReal& x, const RooArgList& coefList, TArrayD const& limits, Int_t nBins) :
  RooAbsPdf(name, title),
  _x("x", "Dependent", this, x),
  _coefList("coefList","List of coefficients",this),
  _nBins(nBins)
{
 
  // Check lowest order
  if (_nBins<0) {
    std::cout << "RooParametricStepFunction::ctor(" << GetName()
              << ") WARNING: nBins must be >=0, setting value to 0" << std::endl ;
    _nBins=0 ;
  }
 
  _coefList.addTyped<RooAbsReal>(coefList);
 
  // Bin limits
  limits.Copy(_limits);
 
}
 
////////////////////////////////////////////////////////////////////////////////
/// Copy constructor
 
RooParametricStepFunction::RooParametricStepFunction(const RooParametricStepFunction& other, const char* name) :
  RooAbsPdf(other, name),
  _x("x", this, other._x),
  _coefList("coefList",this,other._coefList),
  _nBins(other._nBins)
{
  (other._limits).Copy(_limits);
}
 
////////////////////////////////////////////////////////////////////////////////
 
Int_t RooParametricStepFunction::getAnalyticalIntegral(RooArgSet& allVars, RooArgSet& analVars, const char* /*rangeName*/) const
{
  if (matchArgs(allVars, analVars, _x)) return 1;
  return 0;
}
 
////////////////////////////////////////////////////////////////////////////////
 
double RooParametricStepFunction::analyticalIntegral(Int_t /*code*/, const char* rangeName) const
{
  // Case without range is trivial: p.d.f is by construction normalized
  if (!rangeName) {
    return 1.0 ;
  }
 
  // Case with ranges, calculate integral explicitly
  double xmin = _x.min(rangeName) ;
  double xmax = _x.max(rangeName) ;
 
  double sum=0 ;
  Int_t i ;
  for (i=1 ; i<=_nBins ; i++) {
    double binVal = (i<_nBins) ? (static_cast<RooRealVar*>(_coefList.at(i-1))->getVal()) : lastBinValue() ;
    if (_limits[i-1]>=xmin && _limits[i]<=xmax) {
      // Bin fully in the integration domain
      sum += (_limits[i]-_limits[i-1])*binVal ;
    } else if (_limits[i-1]<xmin && _limits[i]>xmax) {
      // Domain is fully contained in this bin
      sum += (xmax-xmin)*binVal ;
      // Exit here, this is the last bin to be processed by construction
      return sum ;
    } else if (_limits[i-1]<xmin && _limits[i]<=xmax && _limits[i]>xmin) {
      // Lower domain boundary is in bin
      sum +=  (_limits[i]-xmin)*binVal ;
    } else if (_limits[i-1]>=xmin && _limits[i]>xmax && _limits[i-1]<xmax) {
      sum +=  (xmax-_limits[i-1])*binVal ;
      // Upper domain boundary is in bin
      // Exit here, this is the last bin to be processed by construction
      return sum ;
    }
  }
 
  return sum;
}
 
////////////////////////////////////////////////////////////////////////////////
 
double RooParametricStepFunction::lastBinValue() const
{
  double sum(0.);
  double binSize(0.);
  for (Int_t j=1;j<_nBins;j++){
    RooRealVar* tmp = static_cast<RooRealVar*>(_coefList.at(j-1));
    binSize = _limits[j] - _limits[j-1];
    sum = sum + tmp->getVal()*binSize;
  }
  binSize = _limits[_nBins] - _limits[_nBins-1];
  return (1.0 - sum)/binSize;
}
 
////////////////////////////////////////////////////////////////////////////////
 
double RooParametricStepFunction::evaluate() const
{
  double value(0.);
  if (_x >= _limits[0] && _x < _limits[_nBins]){
 
    for (Int_t i=1;i<=_nBins;i++){
      if (_x < _limits[i]){
   // in Bin i-1 (starting with Bin 0)
   if (i<_nBins) {
     // not in last Bin
     value = static_cast<RooRealVar*>(_coefList.at(i-1))->getVal();
     break;
   } else {
     value = lastBinValue();
     if (value<=0.0){
       value = 0.000001;
       //       cout << "RooParametricStepFunction: sum of values gt 1.0 -- beware!!" <<endl;
     }
     break;
   }
      }
    }
 
  }
  return value;
 
}
 
 
std::list<double>* RooParametricStepFunction::plotSamplingHint(RooAbsRealLValue& obs, double xlo, double xhi) const
{
   if(obs.namePtr() != _x->namePtr()) {
      return nullptr;
   }
 
  // Retrieve position of all bin boundaries
  std::span<const double> boundaries{_limits.GetArray(), static_cast<std::size_t>(_limits.GetSize())};
 
  // Use the helper function from RooCurve to make sure to get sampling hints
  // that work with the RooFitPlotting.
  return RooCurve::plotSamplingHintForBinBoundaries(boundaries, xlo, xhi);
}