BCRooInterface.cxx

Go to the documentation of this file.

#include "BCRooInterface.h"
#include "../../BAT/BCMath.h"
#include "../../BAT/BCParameter.h"

#include <RooGlobalFunc.h>
#include <RooMsgService.h>
#include <RooProdPdf.h>
#include <RooUniform.h>
#include <RooWorkspace.h>
#include <TFile.h>

//for testing
#include "RooRealVar.h"
#include "RooAbsReal.h"
#include "RooRandom.h"

#include <iostream>

// ---------------------------------------------------------
void BCRooInterface::Initialize( RooAbsData& data,
         RooAbsPdf& model,
         RooAbsPdf& prior_trans,
         const RooArgSet* params,
         const RooArgSet& listPOI )
{

   fData = &data;
   fModel = &model;

   // make the product of both priors to get the full prior probability function
   RooAbsPdf* prior_total = &prior_trans;
   if (prior_total!=0 ) {
      fPrior = prior_total;
   }
   else {
      std::cout << "No prior PDF: without taking action the program would crash\n";
      std::cout << "No prior PDF: adding dummy uniform pdf on the interval [0..1]\n";
      priorhelpvar = new RooRealVar("priorhelpvar","",0.0, 1.0 );
      _addeddummyprior = true;
      RooUniform* priorhelpdist = new RooUniform("priorhelpdist","", *priorhelpvar);
      fPrior = priorhelpdist;
   }

   std::cout << "Imported parameters:\n";
   fParams = new RooArgList(listPOI);
   const RooArgSet* paramsTmp = params;
   if (paramsTmp!=0)
      fParams->add(*paramsTmp);
   fParams->Print("v");

   fParamsPOI = new RooArgList(listPOI);
   // create the log-likelihood function
   //    fNll = new RooNLLVar("fNll","",*fModel,*fData,true/*extended*/);

   RooArgSet* constrainedParams = fModel->getParameters(*fData);
   fNll = fModel->createNLL(*fData, RooFit::Constrain(*constrainedParams) );

   DefineParameters();

   if(_fillChain) {
      SetupRooStatsMarkovChain();
   }
}

// ---------------------------------------------------------

void BCRooInterface::Initialize( const char* rootFile,
                   const char* wsName,
                   const char* dataName,
                   const char* modelName,
                   const char* priorName,
                   const char* priorNuisanceName,
                   const char* paramsName,
                   const char* listPOIName )
{
   // retrieve the RooFit inputs from the ROOT file

   /*
   // hard coded names in the workspace
   char* rootFile = "bat_workspace.root";
   char* wsName= "batWS";
   char* dataName= "data";
   char* modelName= "model";
   char* priorName= "priorPOI";
   char* priorNuisanceName= "priorNuisance";
   char* paramsName= "parameters";
   char* listPOIName= "POI";
   */

   std::cout << "Opening " << rootFile << std::endl;
   TFile* file = TFile::Open(rootFile);
   std::cout << "content :\n";
   file->ls();

   RooWorkspace* bat_ws = (RooWorkspace*) file->Get(wsName);
   bat_ws->Print("v");

   fData = (RooAbsData*) bat_ws->data(dataName);
   fModel = (RooAbsPdf*) bat_ws->function(modelName);

   // make the product of both priors to get the full prior probability function
   RooAbsPdf* priorPOI = (RooAbsPdf*) bat_ws->function(priorName);
   RooAbsPdf* priorNuisance = (RooAbsPdf*) bat_ws->pdf(priorNuisanceName);
   if (priorNuisance!=0 && priorPOI!=0) {
      fPrior = new RooProdPdf("fPrior","complete prior",*priorPOI,*priorNuisance);
   }
   else {
      if ( priorNuisance!=0 )
         fPrior=priorNuisance;
      else if ( priorPOI!=0 )
         fPrior = priorPOI;
      else{
         std::cout << "No prior PDF: without taking action the program would crash\n";
         std::cout << "No prior PDF: adding dummy uniform pdf on the interval [0..1]\n";
         priorhelpvar = new RooRealVar("priorhelpvar","",0.0, 1.0 );
         _addeddummyprior = true;
         RooUniform* priorhelpdist = new RooUniform("priorhelpdist","", *priorhelpvar);
         fPrior = priorhelpdist;
      }
   }

   std::cout << "Imported parameters:\n";
   fParams   = new RooArgList(*(bat_ws->set(listPOIName)));
   RooArgSet* paramsTmp = (RooArgSet*) bat_ws->set(paramsName);
   if (paramsTmp!=0) {
      fParams->add(*paramsTmp);
   }
   if (_addeddummyprior == true ) {
      fParams->add(*priorhelpvar);
   }
   fParams->Print("v");

   // create the log-likelihood function
   //fNll = new RooNLLVar("fNll","",*fModel,*fData,true/*extended*/);
   RooArgSet* constrainedParams = fModel->getParameters(*fData);
   fNll = fModel->createNLL(*fData, RooFit::Constrain(*constrainedParams) );

   file->Close();

   DefineParameters();
}

// ---------------------------------------------------------
BCRooInterface::BCRooInterface(const char* name, bool fillChain) :
      BCModel(name),
      fData(NULL),
      fModel(NULL),
      fNll(NULL),
      fObservables(NULL),
      fParams(NULL),
      fParamsPOI(NULL),
      fPrior(NULL),
      _default_nbins(150),
      priorhelpvar(NULL),
      _addeddummyprior(false),
      _fillChain(fillChain),
      fFirstComparison(false),
      _roostatsMarkovChain(NULL)
{
   // todo this interface not ready for grid marginalization yet
   SetMarginalizationMethod(BCIntegrate::kMargMetropolis);
}

// ---------------------------------------------------------
BCRooInterface::~BCRooInterface()
{  // default destructor
   if(_fillChain) {
      delete _roostatsMarkovChain;
   }
}

// ---------------------------------------------------------
void BCRooInterface::DefineParameters()
{  // define for BAT the list of parameters, range and plot binning

   int nParams = fParams->getSize();
   for (int iParam=0; iParam<nParams; iParam++) {
      RooRealVar* ipar = (RooRealVar*) fParams->at(iParam);
      BCParameter * bcpar = new BCParameter(ipar->GetName(),ipar->getMin(),ipar->getMax());
      bcpar->SetNbins(_default_nbins);
      this->AddParameter(bcpar);
      std::cout << "added parameter: " << bcpar->GetName() << " defined in range [ " << bcpar->GetLowerLimit() << " - "
            << bcpar->GetUpperLimit() << " ] with number of bins: " << bcpar->GetNbins() << " \n";
   }

   for(std::list< std::pair<const char*,int> >::iterator listiter = _nbins_list.begin(); listiter != _nbins_list.end(); listiter++) {
      GetParameter((*listiter).first)->SetNbins((*listiter).second);
      std::cout << "adjusted parameter: " << (*listiter).first << " to number of bins: " << (*listiter).second << " \n";
   }
}

// ---------------------------------------------------------
double BCRooInterface::LogLikelihood(const std::vector<double> & parameters)
{  // this methods returns the logarithm of the conditional probability: p(data|parameters)

   // retrieve the values of the parameters to be tested
   int nParams = fParams->getSize();
   for (int iParam=0; iParam<nParams; iParam++) {
      RooRealVar* ipar = (RooRealVar*) fParams->at(iParam);
      ipar->setVal(parameters.at(iParam));
   }

   // compute the log of the likelihood function
   double logprob = -fNll->getVal();
   return logprob;
}

// ---------------------------------------------------------
double BCRooInterface::LogAPrioriProbability(const std::vector<double> & parameters)
{  // this method returs the logarithm of the prior probability for the parameters: p(parameters).
   // retrieve the values of the parameters to be tested
   int nParams = fParams->getSize();
   for (int iParam=0; iParam<nParams; iParam++) {
      RooRealVar* ipar = (RooRealVar*) fParams->at(iParam);
      ipar->setVal(parameters.at(iParam));
   }
   // compute the log of the prior function
   RooArgSet* tmpArgSet = new RooArgSet(*fParams);
   double prob = fPrior->getVal(tmpArgSet);
   delete tmpArgSet;
   if (prob<1e-300)
      prob = 1e-300;
   return log(prob);
}

void BCRooInterface::SetNumBins(const char * parname, int nbins)
{
   for(std::list< std::pair<const char*,int> >::iterator listiter = _nbins_list.begin(); listiter != _nbins_list.end(); listiter++) {
      if(!strcmp((*listiter).first, parname)) {
         (*listiter).second = nbins;
         return;
      }
   }
   _nbins_list.push_back( std::make_pair(parname,nbins) );
}

void BCRooInterface::SetNumBins(int nbins)
{
   _default_nbins = nbins;
}

void BCRooInterface::SetupRooStatsMarkovChain()
{
   //RooArgSet * tempRooArgSetPointer = new RooArgSet(* fParams, "paramsMarkovChainPointer");
   //_parametersForMarkovChain = RooArgSet(* fParams, "paramsMarkovChainPointer");
   //fParams->snapshot(_parametersForMarkovChain);

   //store only POI in RooStats MarkovChain object
   //_parametersForMarkovChainPrevious.add(*fParamsPOI);
   //_parametersForMarkovChainCurrent.add(*fParamsPOI);

   _parametersForMarkovChainPrevious.add(*fParams);
   _parametersForMarkovChainCurrent.add(*fParams);

   std::cout << "size of _parametersForMarkovChain: " << _parametersForMarkovChainCurrent.getSize() << std::endl;
   std::cout << "size of fParamsPOI: " << fParamsPOI->getSize() << std::endl;
   //std::cout << "size of tempRooArgSetPointer: " << tempRooArgSetPointer->getSize() << std::endl;

   _roostatsMarkovChain = new RooStats::MarkovChain();
   //test stuff begin
   //the following way of creating the MArkovChain object does not work!:
   //_roostatsMarkovChain = new RooStats::MarkovChain(_parametersForMarkovChain);
   //test stuff end
   std::cout << "setting up parameters for RooStats markov chain" << std::endl;
   _parametersForMarkovChainPrevious.writeToStream(std::cout, false);

   //initialize fPreviousStep, fCurrentStep, fVecWeights
   int nchains = MCMCGetNChains();
   for(int countChains = 1; countChains<=nchains ; countChains++ ) {
      double tempweight = 1.0;
      fVecWeights.push_back(tempweight);
         std::vector<double> tempvec;
      TIterator* setiter = fParamsPOI->createIterator();
      double tempval = 0.;
      while(setiter->Next()){
         tempvec.push_back(tempval);
      }
      fPreviousStep.push_back(tempvec);
      fCurrentStep.push_back(tempvec);
   }

   fFirstComparison = true;

   //test stuff begin
   //var1 = new RooRealVar("var1","var1", 10., 0., 100.);
   //fParamsTest   = new RooArgList();
   //fParamsTest->add(*var1);
   //_parametersForMarkovChain_test.add(*fParamsTest);
   //fIterationInterfacecount = 0;
   //test stuff end
}

// to be added: add elements with higher weights if elements repeat to avoid unneccessarily long chains
void BCRooInterface::MCMCIterationInterface()
{
   //fIterationInterfacecount+=1;

   if(_fillChain) {
      //std::cout << "Hei-ho running with fillChain!" << std::endl;
      // get number of chains
      int nchains = MCMCGetNChains();

      // get number of parameters
      int npar = GetNParameters();
      //std::cout << "number of parameters is: " << npar << std::endl;

      // loop over all chains and fill histogram
      for (int i = 0; i < nchains; ++i) {
         // get the current values of the parameters. These are
         // stored in fMCMCx.

         // std::cout << "log(likelihood*prior)" << *GetMarkovChainValue() << std::endl; //does not work this way?!
         TIterator* setiter = fParams->createIterator();
         int j = 0;

         //store only POI in RooStats MarkovChain object
         //TIterator* setiter = fParamsPOI->createIterator();
         //int j = 0;

         while(setiter->Next()){

            //check parameter names
            const BCParameter * tempBCparam = GetParameter(j);

            //_parametersForMarkovChainCurrent->Print();

            const char * paramnamepointer = (tempBCparam->GetName()).c_str();
            double xij = fMCMCx.at(i * npar + j);
            AddToCurrentChainElement(xij, i, j);
            RooRealVar* parampointer = (RooRealVar*) &(_parametersForMarkovChainCurrent[paramnamepointer]);
            parampointer->setVal(xij);
            //std::cout << "Chain " << i << " param: " << tempBCparam->GetName() << " Value: " << xij << std::endl;
            j++;
         }


         // will only work if _parametersForMarkovChain had correct info!

         //test stuff begin
         //var1->setVal( RooRandom::randomGenerator()->Uniform(100.) );
         //
         //_roostatsMarkovChain->Add(_parametersForMarkovChain_test, 0.001, 1.0);
         //
         //test stuff end

         if( !(EqualsLastChainElement(i)) ) {
            double weight = GetWeightForChain(i);
            _roostatsMarkovChain->Add(_parametersForMarkovChainPrevious, -1.* MCMCGetLogProbx(j), weight);
            _parametersForMarkovChainPrevious = _parametersForMarkovChainCurrent;
         }
      }
   }
}

void BCRooInterface::AddToCurrentChainElement(double xij, int chainNum, int poiNum)
{
   fCurrentStep[chainNum][poiNum] = xij;
}

bool BCRooInterface::EqualsLastChainElement(int chainNum)
{
   bool equals = true;
    std::vector<double>::iterator itPrevious = fPreviousStep[chainNum].begin();

   if(fFirstComparison == true) {
      fFirstComparison = false;
      _parametersForMarkovChainPrevious = _parametersForMarkovChainCurrent;
      return true;
   }


   for (std::vector<double>::iterator itCurrent = fCurrentStep[chainNum].begin(); itCurrent != fCurrentStep[chainNum].end(); ++itCurrent) {
      if(*itCurrent != *itPrevious) {
         equals = false;
         fPreviousStep[chainNum] = fCurrentStep[chainNum];
         break;
      }
   ++itPrevious;
   }

   if(equals == true) {
      fVecWeights[chainNum] += 1.0;
   }

   return equals;

}

double BCRooInterface::GetWeightForChain(int chainNum)
{
   double retval = fVecWeights[chainNum];
   fVecWeights[chainNum]= 1.0 ;
   return retval;
}