MCFitter.cxx

Go to the documentation of this file.

// @(#)root/tmva $Id$
// Author: Andreas Hoecker, Peter Speckmayer, Joerg Stelzer, Helge Voss
 
/**********************************************************************************
 * Project: TMVA - a Root-integrated toolkit for multivariate data analysis       *
 * Package: TMVA                                                                  *
 * Class  : TMVA::MCFitter                                                        *
 *                                             *
 *                                                                                *
 * Description:                                                                   *
 *      Implementation                                                            *
 *                                                                                *
 * Authors (alphabetical):                                                        *
 *      Andreas Hoecker <Andreas.Hocker@cern.ch> - CERN, Switzerland              *
 *      Peter Speckmayer <speckmay@mail.cern.ch> - CERN, Switzerland              *
 *      Joerg Stelzer    <Joerg.Stelzer@cern.ch>  - CERN, Switzerland             *
 *      Helge Voss      <Helge.Voss@cern.ch>     - MPI-K Heidelberg, Germany      *
 *                                                                                *
 * Copyright (c) 2005:                                                            *
 *      CERN, Switzerland                                                         *
 *      MPI-K Heidelberg, Germany                                                 *
 *                                                                                *
 * Redistribution and use in source and binary forms, with or without             *
 * modification, are permitted according to the terms listed in LICENSE           *
 * (see tmva/doc/LICENSE)                                          *
 **********************************************************************************/
 
/*! \class TMVA::MCFitter
\ingroup TMVA
 
Fitter using Monte Carlo sampling of parameters.
 
*/
 
#include "TMVA/MCFitter.h"
 
#include "TMVA/Configurable.h"
#include "TMVA/FitterBase.h"
#include "TMVA/GeneticRange.h"
#include "TMVA/Interval.h"
#include "TMVA/MsgLogger.h"
#include "TMVA/Timer.h"
#include "TMVA/Types.h"
#include "TRandom3.h"
 
ClassImp(TMVA::MCFitter);
 
////////////////////////////////////////////////////////////////////////////////
/// constructor
 
TMVA::MCFitter::MCFitter( IFitterTarget& target,
                          const TString& name,
                          const std::vector<Interval*>& ranges,
                          const TString& theOption )
: TMVA::FitterBase( target, name, ranges, theOption ),
   fSamples( 0 ),
   fSigma  ( 1 ),
   fSeed   ( 0 )
{
   DeclareOptions();
   ParseOptions();
}
 
////////////////////////////////////////////////////////////////////////////////
/// Declare MCFitter options
 
void TMVA::MCFitter::DeclareOptions()
{
   DeclareOptionRef( fSamples = 100000, "SampleSize", "Number of Monte Carlo events in toy sample" );
   DeclareOptionRef( fSigma   = -1.0,   "Sigma",
                     "If > 0: new points are generated according to Gauss around best value and with \"Sigma\" in units of interval length" );
   DeclareOptionRef( fSeed    = 100,    "Seed",       "Seed for the random generator (0 takes random seeds)" );
}
 
////////////////////////////////////////////////////////////////////////////////
/// set MC fitter configuration parameters
 
void TMVA::MCFitter::SetParameters( Int_t samples )
{
   fSamples = samples;
}
 
////////////////////////////////////////////////////////////////////////////////
/// Execute fitting
 
Double_t TMVA::MCFitter::Run( std::vector<Double_t>& pars )
{
   Log() << kHEADER << "<MCFitter> Sampling, please be patient ..." << Endl;
 
   // sanity check
   if ((Int_t)pars.size() != GetNpars())
      Log() << kFATAL << "<Run> Mismatch in number of parameters: "
            << GetNpars() << " != " << pars.size() << Endl;
 
   // timing of MC
   Timer timer( fSamples, GetName() );
   if (fIPyMaxIter) *fIPyMaxIter = fSamples;
 
   std::vector<Double_t> parameters;
   std::vector<Double_t> bestParameters;
 
   TRandom3*rnd = new TRandom3( fSeed );
   rnd->Uniform(0.,1.);
 
   std::vector<TMVA::GeneticRange*> rndRanges;
 
   // initial parameters (given by argument) are ignored
   std::vector< TMVA::Interval* >::const_iterator rIt;
   Double_t val;
   for (rIt = fRanges.begin(); rIt<fRanges.end(); ++rIt) {
      rndRanges.push_back( new TMVA::GeneticRange( rnd, (*rIt) ) );
      val = rndRanges.back()->Random();
      parameters.push_back( val );
      bestParameters.push_back( val );
   }
 
   std::vector<Double_t>::iterator parIt;
   std::vector<Double_t>::iterator parBestIt;
 
   Double_t estimator = 0;
   Double_t bestFit   = 0;
 
   // loop over all MC samples
   for (Int_t sample = 0; sample < fSamples; sample++) {
     if (fIPyCurrentIter) *fIPyCurrentIter = sample;
     if (fExitFromTraining && *fExitFromTraining) break;
 
      // dice the parameters
      parIt = parameters.begin();
      if (fSigma > 0.0) {
         parBestIt = bestParameters.begin();
         for (std::vector<TMVA::GeneticRange*>::iterator rndIt = rndRanges.begin(); rndIt<rndRanges.end(); ++rndIt) {
            (*parIt) = (*rndIt)->Random( kTRUE, (*parBestIt), fSigma );
            ++parIt;
            ++parBestIt;
         }
      }
      else {
         for (std::vector<TMVA::GeneticRange*>::iterator rndIt = rndRanges.begin(); rndIt<rndRanges.end(); ++rndIt) {
            (*parIt) = (*rndIt)->Random();
            ++parIt;
         }
      }
 
      // test the estimator value for the parameters
      estimator = EstimatorFunction( parameters );
 
      // if the estimator ist better (=smaller), take the new parameters as the best ones
      if (estimator < bestFit || sample==0) {
         bestFit = estimator;
         bestParameters.swap( parameters );
      }
 
      // whats the time please?
      if ((fSamples<100) || sample%Int_t(fSamples/100.0) == 0) timer.DrawProgressBar( sample );
   }
   pars.swap( bestParameters ); // return best parameters found
 
   // get elapsed time
   Log() << kINFO << "Elapsed time: " << timer.GetElapsedTime()
         << "                           " << Endl;
 
   return bestFit;
}