FeldmanCousins.cxx

Go to the documentation of this file.

// @(#)root/roostats:$Id$
// Author: Kyle Cranmer   January 2009

/*************************************************************************
 * Copyright (C) 1995-2008, Rene Brun and Fons Rademakers.               *
 * All rights reserved.                                                  *
 *                                                                       *
 * For the licensing terms see $ROOTSYS/LICENSE.                         *
 * For the list of contributors see $ROOTSYS/README/CREDITS.             *
 *************************************************************************/


#include "RooStats/FeldmanCousins.h"

#include "RooStats/RooStatsUtils.h"

#include "RooStats/PointSetInterval.h"

#include "RooStats/ModelConfig.h"

#include "RooStats/SamplingDistribution.h"

#include "RooStats/ProfileLikelihoodTestStat.h"
#include "RooStats/NeymanConstruction.h"
#include "RooStats/RooStatsUtils.h"

#include "RooDataSet.h"
#include "RooDataHist.h"
#include "RooGlobalFunc.h"
#include "RooMsgService.h"
#include "TFile.h"
#include "TTree.h"

/** \class RooStats::FeldmanCousins
    \ingroup Roostats

The FeldmanCousins class (like the Feldman-Cousins technique) is essentially a
specific configuration of the more general NeymanConstruction.  It is a concrete
implementation of the IntervalCalculator interface that, which uses the
NeymanConstruction in a particular way. As the name suggests, it returns a
ConfidenceInterval.  In particular, it produces a RooStats::PointSetInterval,
which is a concrete implementation of the ConfInterval interface.

The Neyman Construction is not a uniquely defined statistical technique, it
requires that one specify an ordering rule  or ordering principle, which is
usually encoded by choosing a specific test statistic and limits of integration
(corresponding to upper/lower/central limits).  As a result, this class must be
configured with the corresponding information before it can produce an interval.

In the case of the Feldman-Cousins approach, the ordering principle is the
likelihood ratio -- motivated by the Neyman-Pearson lemma.  When nuisance
parameters are involved, the profile likelihood ratio is the natural
generalization.  One may either choose to perform the construction over the full
space of the nuisance parameters, or restrict the nuisance parameters to their
conditional MLE (eg. profiled values).

*/

ClassImp(RooStats::FeldmanCousins); ;

using namespace RooFit;
using namespace RooStats;
using namespace std;

////////////////////////////////////////////////////////////////////////////////
/// standard constructor

FeldmanCousins::FeldmanCousins(RooAbsData& data, ModelConfig& model) :
  fSize(0.05),
  fModel(model),
  fData(data),
  fTestStatSampler(0),
  fPointsToTest(0),
  fPOIToTest(0),
  fConfBelt(0),
  fAdaptiveSampling(false),
  fAdditionalNToysFactor(1.),
  fNbins(10),
  fFluctuateData(true),
  fDoProfileConstruction(true),
  fSaveBeltToFile(false),
  fCreateBelt(false)
{
}

////////////////////////////////////////////////////////////////////////////////
/// destructor

FeldmanCousins::~FeldmanCousins() {
  if(fPointsToTest) delete fPointsToTest;
  if(fPOIToTest) delete fPOIToTest;
  if(fTestStatSampler) delete fTestStatSampler;
}

////////////////////////////////////////////////////////////////////////////////
/// set the model

void FeldmanCousins::SetModel(const ModelConfig & model) {
  fModel = model;
}

////////////////////////////////////////////////////////////////////////////////

TestStatSampler*  FeldmanCousins::GetTestStatSampler() const{
  if(!fTestStatSampler)
    this->CreateTestStatSampler();
  return fTestStatSampler;
}

////////////////////////////////////////////////////////////////////////////////
/// specify the Test Statistic and create a ToyMC test statistic sampler

void FeldmanCousins::CreateTestStatSampler() const{
  // use the profile likelihood ratio as the test statistic
  ProfileLikelihoodTestStat* testStatistic = new ProfileLikelihoodTestStat(*fModel.GetPdf());

  // create the ToyMC test statistic sampler
  fTestStatSampler = new ToyMCSampler(*testStatistic,int(fAdditionalNToysFactor*50./fSize)) ;
  fTestStatSampler->SetParametersForTestStat(*fModel.GetParametersOfInterest() );
  if(fModel.GetObservables())
    fTestStatSampler->SetObservables(*fModel.GetObservables());
  fTestStatSampler->SetPdf(*fModel.GetPdf());

  if(!fAdaptiveSampling){
    ooccoutP(&fModel,Generation) << "FeldmanCousins: ntoys per point = " << (int) (fAdditionalNToysFactor*50./fSize) << endl;
  } else{
    ooccoutP(&fModel,Generation) << "FeldmanCousins: ntoys per point: adaptive" << endl;
  }
  if(fFluctuateData){
    ooccoutP(&fModel,Generation) << "FeldmanCousins: nEvents per toy will fluctuate about  expectation" << endl;
  } else{
    ooccoutP(&fModel,Generation) << "FeldmanCousins: nEvents per toy will not fluctuate, will always be " << fData.numEntries() << endl;
    fTestStatSampler->SetNEventsPerToy(fData.numEntries());
  }
}

////////////////////////////////////////////////////////////////////////////////
/// specify the parameter points to perform the construction.
/// allow ability to profile on some nuisance parameters

void FeldmanCousins::CreateParameterPoints() const{
  // get ingredients
  RooAbsPdf* pdf   = fModel.GetPdf();
  if (!pdf ){
    ooccoutE(&fModel,Generation) << "FeldmanCousins: ModelConfig has no PDF" << endl;
    return;
  }

  // get list of all parameters
  RooArgSet* parameters = new RooArgSet(*fModel.GetParametersOfInterest());
  if(fModel.GetNuisanceParameters())
    parameters->add(*fModel.GetNuisanceParameters());


  if( fModel.GetNuisanceParameters() && ! fModel.GetParametersOfInterest()->equals(*parameters) && fDoProfileConstruction) {
    // if parameters include nuisance parameters, do profile construction
    ooccoutP(&fModel,Generation) << "FeldmanCousins: Model has nuisance parameters, will do profile construction" << endl;

    // set nbins for the POI
    TIter it2 = fModel.GetParametersOfInterest()->createIterator();
    RooRealVar *myarg2;
    while ((myarg2 = dynamic_cast<RooRealVar*>(it2.Next()))) {
      myarg2->setBins(fNbins);
    }

    // get dataset for POI scan
    //     RooDataHist* parameterScan = NULL;
    RooAbsData* parameterScan = NULL;
    if(fPOIToTest)
      parameterScan = fPOIToTest;
    else
      parameterScan = new RooDataHist("parameterScan", "", *fModel.GetParametersOfInterest());


    ooccoutP(&fModel,Generation) << "FeldmanCousins: # points to test = " << parameterScan->numEntries() << endl;
    // make profile construction
    RooFit::MsgLevel previous  = RooMsgService::instance().globalKillBelow();
    RooMsgService::instance().setGlobalKillBelow(RooFit::FATAL) ;
    RooAbsReal* nll = pdf->createNLL(fData,RooFit::CloneData(false));
    RooAbsReal* profile = nll->createProfile(*fModel.GetParametersOfInterest());

    RooDataSet* profileConstructionPoints = new RooDataSet("profileConstruction",
                        "profileConstruction",
                        *parameters);


    for(Int_t i=0; i<parameterScan->numEntries(); ++i){
      // here's where we figure out the profiled value of nuisance parameters
      *parameters = *parameterScan->get(i);
      profile->getVal();
      profileConstructionPoints->add(*parameters);
    }
    RooMsgService::instance().setGlobalKillBelow(previous) ;
    delete profile;
    delete nll;
    if(!fPOIToTest) delete parameterScan;

    // done
    fPointsToTest = profileConstructionPoints;

  } else{
    // Do full construction
    ooccoutP(&fModel,Generation) << "FeldmanCousins: Model has no nuisance parameters" << endl;

    TIter it = parameters->createIterator();
    RooRealVar *myarg;
    while ((myarg = dynamic_cast<RooRealVar*>(it.Next()))) {
      myarg->setBins(fNbins);
    }

    RooDataHist* parameterScan = new RooDataHist("parameterScan", "", *parameters);
    ooccoutP(&fModel,Generation) << "FeldmanCousins: # points to test = " << parameterScan->numEntries() << endl;

    fPointsToTest = parameterScan;
  }

  delete parameters;

}

////////////////////////////////////////////////////////////////////////////////
/// Main interface to get a RooStats::ConfInterval.
/// It constructs a RooStats::PointSetInterval.

PointSetInterval* FeldmanCousins::GetInterval() const {
  // local variables
  //  RooAbsData* data = fData; //fWS->data(fDataName);

  // fill in implied variables given data
  fModel.GuessObsAndNuisance(fData);

  // create the test statistic sampler (private data member fTestStatSampler)
  if(!fTestStatSampler)
    this->CreateTestStatSampler();

  fTestStatSampler->SetObservables(*fModel.GetObservables());

  if(!fFluctuateData)
    fTestStatSampler->SetNEventsPerToy(fData.numEntries());

  // create parameter points to perform construction (private data member fPointsToTest)
  this->CreateParameterPoints();

  // Create a Neyman Construction
  NeymanConstruction nc(fData,fModel);
  // configure it
  nc.SetTestStatSampler(*fTestStatSampler);
  nc.SetTestSize(fSize); // set size of test
  //  nc.SetParameters( fModel.GetParametersOfInterest);
  nc.SetParameterPointsToTest( *fPointsToTest );
  nc.SetLeftSideTailFraction(0.); // part of definition of Feldman-Cousins
  nc.SetData(fData);
  nc.UseAdaptiveSampling(fAdaptiveSampling);
  nc.AdditionalNToysFactor(fAdditionalNToysFactor);
  nc.SaveBeltToFile(fSaveBeltToFile);
  nc.CreateConfBelt(fCreateBelt);
  fConfBelt = nc.GetConfidenceBelt();
  // use it
  return nc.GetInterval();
}