LauDPDepSumPdf.cc

Go to the documentation of this file.

// Copyright University of Warwick 2009 - 2013.
// Distributed under the Boost Software License, Version 1.0.
// (See accompanying file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)

// Authors:
// Thomas Latham
// John Back
// Paul Harrison

#include <iostream>
#include <vector>
using std::cout;
using std::cerr;
using std::endl;
using std::vector;

#include "TMath.h"
#include "TSystem.h"

#include "LauConstants.hh"
#include "LauDaughters.hh"
#include "LauEffModel.hh"
#include "LauParameter.hh"
#include "LauDPDepSumPdf.hh"

ClassImp(LauDPDepSumPdf)


LauDPDepSumPdf::LauDPDepSumPdf(LauAbsPdf* pdf1, LauAbsPdf* pdf2,
                const LauDaughters* daughters,
                const TH2* dpHisto, Bool_t upperHalf, Bool_t useSpline) :
        LauAbsPdf(pdf1 ? pdf1->varNames() : vector<TString>(), vector<LauAbsRValue*>(), pdf1 ? pdf1->getMinAbscissas() : LauFitData(), pdf1 ? pdf1->getMaxAbscissas() : LauFitData()),
        daughters_( new LauDaughters(*daughters) ),
        pdf1_(pdf1),
        pdf2_(pdf2),
        frac_(0),
        fracVal_(0.5),
        dpDependence_( new LauEffModel(daughters, 0) ),
        dpAxis_( CentreDist )
{
        // Constructor for the sum PDF.
        // We are defining the sum as:
        // f x (PDF1/S(PDF1)) + (1-f) x (PDF2/S(PDF2))
        // where f is the fraction, x is multiplication, PDFi is the i'th PDF,
        // and S(PDFi) is the integral of the i'th PDF.
        // The value of the fraction is read from the DP histogram.

        if(useSpline) {
                dpDependence_->setEffSpline( dpHisto, kFALSE, 0.0, 0.0, upperHalf, daughters->squareDP());
        } else {
                dpDependence_->setEffHisto( dpHisto, kTRUE, kFALSE, 0.0, 0.0, upperHalf, daughters->squareDP() );
        }

        // So the first thing we have to do is check the pointers are all valid.
        if (!pdf1 || !pdf2) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: one of the 2 PDF pointers is null."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Next check that the abscissa ranges are the same for each PDF
        if (pdf1->getMinAbscissa() != pdf2->getMinAbscissa()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: minimum abscissa values not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }
        if (pdf1->getMaxAbscissa() != pdf2->getMaxAbscissa()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: maximum abscissa values not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Also check that both PDFs are expecting the same number of input variables
        if (pdf1->nInputVars() != pdf2->nInputVars()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: number of input variables not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Also check that both PDFs are expecting the same variable name(s)
        if (pdf1->varNames() != pdf2->varNames()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: variable name(s) not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Then we need to grab all the parameters and pass them to the base class.
        // This is so that when we are asked for them they can be put into the fit.
        // The number of parameters is the number in PDF1 + the number in PDF2.
        UInt_t nPar(pdf1->nParameters()+pdf2->nParameters());
        vector<LauAbsRValue*> params;  params.reserve(nPar);
        vector<LauAbsRValue*>& pdf1pars = pdf1->getParameters();
        vector<LauAbsRValue*>& pdf2pars = pdf2->getParameters();
        for (vector<LauAbsRValue*>::iterator iter = pdf1pars.begin(); iter != pdf1pars.end(); ++iter) {
                params.push_back(*iter);
        }
        for (vector<LauAbsRValue*>::iterator iter = pdf2pars.begin(); iter != pdf2pars.end(); ++iter) {
                params.push_back(*iter);
        }
        this->addParameters(params);

        // Cache the normalisation factor
        this->calcNorm();
}

LauDPDepSumPdf::LauDPDepSumPdf(LauAbsPdf* pdf1, LauAbsPdf* pdf2,
                LauParameter* frac,
                const LauDaughters* daughters,
                const std::vector<Double_t>& fracCoeffs,
                DPAxis dpAxis) :
        LauAbsPdf(pdf1 ? pdf1->varNames() : vector<TString>(), vector<LauAbsRValue*>(), pdf1 ? pdf1->getMinAbscissas() : LauFitData(), pdf1 ? pdf1->getMaxAbscissas() : LauFitData()),
        daughters_( new LauDaughters(*daughters) ),
        pdf1_(pdf1),
        pdf2_(pdf2),
        frac_(frac),
        fracVal_( frac ? frac->unblindValue() : 0.0 ),
        dpDependence_( 0 ),
        fracCoeffs_( fracCoeffs ),
        dpAxis_( dpAxis )
{
        // Constructor for the sum PDF.
        // We are defining the sum as:
        // f x (PDF1/S(PDF1)) + (1-f) x (PDF2/S(PDF2))
        // where f is the fraction, x is multiplication, PDFi is the i'th PDF,
        // and S(PDFi) is the integral of the i'th PDF.
        // The value of the fraction has a polynomial dependence on one of
        // the DP axes.

        // So the first thing we have to do is check the pointers are all valid.
        if (!pdf1 || !pdf2) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: one of the 2 PDF pointers is null."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }
        if ( !frac ) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: the fraction parameter pointer is null."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Next check that the abscissa ranges are the same for each PDF
        if (pdf1->getMinAbscissa() != pdf2->getMinAbscissa()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: minimum abscissa values not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }
        if (pdf1->getMaxAbscissa() != pdf2->getMaxAbscissa()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: maximum abscissa values not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Also check that both PDFs are expecting the same number of input variables
        if (pdf1->nInputVars() != pdf2->nInputVars()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: number of input variables not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Also check that both PDFs are expecting the same variable name(s)
        if (pdf1->varNames() != pdf2->varNames()) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: variable name(s) not the same for the two PDFs."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Then we need to grab all the parameters and pass them to the base class.
        // This is so that when we are asked for them they can be put into the fit.
        // The number of parameters is the number in PDF1 + the number in PDF2.
        UInt_t nPar( pdf1->nParameters() + pdf2->nParameters() + 1 );
        vector<LauAbsRValue*> params;  params.reserve(nPar);
        params.push_back(frac);
        vector<LauAbsRValue*>& pdf1pars = pdf1->getParameters();
        vector<LauAbsRValue*>& pdf2pars = pdf2->getParameters();
        for (vector<LauAbsRValue*>::iterator iter = pdf1pars.begin(); iter != pdf1pars.end(); ++iter) {
                params.push_back(*iter);
        }
        for (vector<LauAbsRValue*>::iterator iter = pdf2pars.begin(); iter != pdf2pars.end(); ++iter) {
                params.push_back(*iter);
        }
        this->addParameters(params);

        // Now check that we can find the fraction parameter ok
        frac_ = this->findParameter("frac");
        if (frac_ == 0) {
                cerr<<"ERROR in LauDPDepSumPdf constructor: parameter \"frac\" not found."<<endl;
                gSystem->Exit(EXIT_FAILURE);
        }

        // Cache the normalisation factor
        this->calcNorm();
}

LauDPDepSumPdf::~LauDPDepSumPdf() 
{
        // Destructor
        delete daughters_; daughters_ = 0;
        delete dpDependence_; dpDependence_ = 0;
}

void LauDPDepSumPdf::calcLikelihoodInfo(const LauAbscissas& abscissas)
{
        // Check that the given abscissa is within the allowed range
        if (!this->checkRange(abscissas)) {
                gSystem->Exit(EXIT_FAILURE);
        }

        LauAbscissas noDPVars(1);
        noDPVars[0] = abscissas[0];

        // Evaluate the normalised PDF values
        if ( pdf1_->isDPDependent() ) {
                pdf1_->calcLikelihoodInfo(abscissas);
        } else {
                pdf1_->calcLikelihoodInfo(noDPVars);
        }
        if ( pdf2_->isDPDependent() ) {
                pdf2_->calcLikelihoodInfo(abscissas);
        } else {
                pdf2_->calcLikelihoodInfo(noDPVars);
        }
        Double_t result1 = pdf1_->getLikelihood();
        Double_t result2 = pdf2_->getLikelihood();

        // Determine the fraction
        // The DP variables will be abscissas[nInputVars] and
        // abscissas[nInputVars+1] (if present).
        UInt_t nVars = this->nInputVars();
        if ( abscissas.size() == nVars+2 ) {
                Double_t m13Sq = abscissas[nVars];
                Double_t m23Sq = abscissas[nVars+1];
                LauKinematics* kinematics = daughters_->getKinematics();
                if ( dpDependence_ ) {
                        kinematics->updateKinematics( m13Sq, m23Sq );
                        fracVal_ = dpDependence_->calcEfficiency( kinematics );
                } else {
                        fracVal_ = frac_->unblindValue();
                        Double_t dpPos(0.0);
                        if ( dpAxis_ == M12 ) {
                                dpPos = kinematics->calcThirdMassSq(m13Sq,m23Sq);
                        } else if ( dpAxis_ == M13 ) {
                                dpPos = m13Sq;
                        } else if ( dpAxis_ == M23 ) {
                                dpPos = m23Sq;
                        } else if ( dpAxis_ == MMIN ) {
                                dpPos = TMath::Min( m13Sq, m23Sq );
                        } else if ( dpAxis_ == MMAX ) {
                                dpPos = TMath::Max( m13Sq, m23Sq );
                        } else {
                                dpPos = kinematics->distanceFromDPCentre(m13Sq,m23Sq);
                        }
                        this->scaleFrac( dpPos );
                }
        }

        // Add them together
        Double_t result = fracVal_ * result1 + (1.0-fracVal_) * result2;
        this->setUnNormPDFVal(result);
}

void LauDPDepSumPdf::scaleFrac( Double_t dpPos )
{
        Int_t power = 1;
        for (std::vector<Double_t>::const_iterator iter = fracCoeffs_.begin(); iter != fracCoeffs_.end(); ++iter) {
                Double_t coeff = (*iter);
                fracVal_ += coeff * TMath::Power(dpPos,power);
                ++power;
        }
}

void LauDPDepSumPdf::calcNorm() 
{
        // Nothing to do here, since it is already normalized
        this->setNorm(1.0);
}

void LauDPDepSumPdf::calcPDFHeight( const LauKinematics* kinematics )
{
        // This method gives you the maximum possible height of the PDF.
        // It combines the maximum heights of the two individual PDFs.
        // So it would give the true maximum if the two individual maxima coincided.
        // It is guaranteed to always be >= the true maximum.

        // Update the heights of the individual PDFs
        pdf1_->calcPDFHeight( kinematics );
        pdf2_->calcPDFHeight( kinematics );

        // Get the up to date parameter values
        if ( dpDependence_ ) {
                fracVal_ = dpDependence_->calcEfficiency( kinematics );
        } else {
                fracVal_ = frac_->unblindValue();
                Double_t dpPos(0.0);
                if ( dpAxis_ == M12 ) {
                        dpPos = kinematics->getm12Sq();
                } else if ( dpAxis_ == M13 ) {
                        dpPos = kinematics->getm13Sq();
                } else if ( dpAxis_ == M23 ) {
                        dpPos = kinematics->getm23Sq();
                } else if ( dpAxis_ == MMIN ) {
                        Double_t m13Sq = kinematics->getm13Sq();
                        Double_t m23Sq = kinematics->getm23Sq();
                        dpPos = TMath::Min( m13Sq, m23Sq );
                } else if ( dpAxis_ == MMAX ) {
                        Double_t m13Sq = kinematics->getm13Sq();
                        Double_t m23Sq = kinematics->getm23Sq();
                        dpPos = TMath::Max( m13Sq, m23Sq );
                } else {
                        dpPos = kinematics->distanceFromDPCentre();
                }
                this->scaleFrac( dpPos );
        }

        // Find the (un-normalised) individual PDF maxima
        Double_t height1 = pdf1_->getMaxHeight();
        Double_t height2 = pdf2_->getMaxHeight();

        // Get the individual PDF normalisation factors
        Double_t norm1 = pdf1_->getNorm();
        Double_t norm2 = pdf2_->getNorm();

        // Calculate the normalised individual PDF maxima
        height1 /= norm1;
        height2 /= norm2;

        // Combine these heights together
        Double_t height = fracVal_ * height1 + (1-fracVal_) * height2;
        this->setMaxHeight(height);
}

// Override the base class methods for cacheInfo and calcLikelihoodInfo(UInt_t iEvt).
// For both of these we delegate to the two constituent PDFs.

void LauDPDepSumPdf::cacheInfo(const LauFitDataTree& inputData)
{
        // delegate to the two sub-PDFs to cache their information
        pdf1_->cacheInfo(inputData);
        pdf2_->cacheInfo(inputData);

        // now check that the DP variables are included in the data
        Bool_t hasBranch = inputData.haveBranch( "m13Sq" );
        hasBranch &= inputData.haveBranch( "m23Sq" );
        if (!hasBranch) {
                cerr<<"ERROR in LauDPDepSumPdf::cacheInfo : Input data does not contain Dalitz plot variables m13Sq and/or m23Sq."<<endl;
                return;
        }

        // determine whether we are caching our PDF value
        Bool_t doCaching( this->nFixedParameters() == this->nParameters() );
        this->cachePDF( doCaching );

        if ( !doCaching ) {
                // in this case we seem to be doing a fit where the parameters are floating
                // so need to mark that the PDF height is no longer up to date
                this->heightUpToDate(kFALSE);
        }

        // clear the vector and reserve enough space
        UInt_t nEvents = inputData.nEvents();
        fractions_.clear();
        fractions_.reserve(nEvents);

        // loop through the events, determine the fraction and store
        for (UInt_t iEvt = 0; iEvt < nEvents; iEvt++) {

                const LauFitData& dataValues = inputData.getData(iEvt);

                Double_t m13Sq = dataValues.find("m13Sq")->second;
                Double_t m23Sq = dataValues.find("m23Sq")->second;

                LauKinematics* kinematics = daughters_->getKinematics();
                if ( dpDependence_ ) {
                        // if we're using the histogram then just
                        // determine the fraction and store
                        kinematics->updateKinematics( m13Sq, m23Sq );
                        fracVal_ = dpDependence_->calcEfficiency( kinematics );
                } else {
                        // if we're scaling the fraction parameter then we
                        // just store the scaling info since the parameter
                        // might be floating
                        fracVal_ = frac_->unblindValue();
                        Double_t dpPos(0.0);
                        if ( dpAxis_ == M12 ) {
                                dpPos = kinematics->calcThirdMassSq(m13Sq,m23Sq);
                        } else if ( dpAxis_ == M13 ) {
                                dpPos = m13Sq;
                        } else if ( dpAxis_ == M23 ) {
                                dpPos = m23Sq;
                        } else if ( dpAxis_ == MMIN ) {
                                dpPos = TMath::Min( m13Sq, m23Sq );
                        } else if ( dpAxis_ == MMAX ) {
                                dpPos = TMath::Max( m13Sq, m23Sq );
                        } else {
                                dpPos = kinematics->distanceFromDPCentre(m13Sq,m23Sq);
                        }
                        this->scaleFrac( dpPos );
                        fracVal_ -= frac_->unblindValue();
                }

                fractions_.push_back( fracVal_ );
        }
}

void LauDPDepSumPdf::calcLikelihoodInfo(UInt_t iEvt)
{
        // Get the fraction value for this event
        fracVal_ = fractions_[iEvt];
        if ( frac_ ) {
                // if we're scaling the parameter then need to add the
                // current value of the parameter
                fracVal_ += frac_->unblindValue();
        }

        // Evaluate the normalised PDF values
        pdf1_->calcLikelihoodInfo(iEvt);
        pdf2_->calcLikelihoodInfo(iEvt);
        Double_t result1 = pdf1_->getLikelihood();
        Double_t result2 = pdf2_->getLikelihood();

        // Add them together
        Double_t result = fracVal_ * result1 + (1-fracVal_) * result2;
        this->setUnNormPDFVal(result);
}