MethodBase.cxx

Go to the documentation of this file.

// @(#)root/tmva $Id$
// Author: Andreas Hoecker, Peter Speckmayer, Joerg Stelzer, Helge Voss, Kai Voss, Eckhard von Toerne, Jan Therhaag

/**********************************************************************************
 * Project: TMVA - a Root-integrated toolkit for multivariate data analysis       *
 * Package: TMVA                                                                  *
 * Class  : MethodBase                                                            *
 * Web    : http://tmva.sourceforge.net                                           *
 *                                                                                *
 * Description:                                                                   *
 *      Implementation (see header for description)                               *
 *                                                                                *
 * Authors (alphabetical):                                                        *
 *      Andreas Hoecker <Andreas.Hocker@cern.ch> - CERN, Switzerland              *
 *      Joerg Stelzer   <Joerg.Stelzer@cern.ch>  - CERN, Switzerland              *
 *      Peter Speckmayer  <Peter.Speckmayer@cern.ch>  - CERN, Switzerland         *
 *      Helge Voss      <Helge.Voss@cern.ch>     - MPI-K Heidelberg, Germany      *
 *      Kai Voss        <Kai.Voss@cern.ch>       - U. of Victoria, Canada         *
 *      Jan Therhaag       <Jan.Therhaag@cern.ch>     - U of Bonn, Germany        *
 *      Eckhard v. Toerne  <evt@uni-bonn.de>          - U of Bonn, Germany        *
 *                                                                                *
 * Copyright (c) 2005-2011:                                                       *
 *      CERN, Switzerland                                                         *
 *      U. of Victoria, Canada                                                    *
 *      MPI-K Heidelberg, Germany                                                 *
 *      U. of Bonn, Germany                                                       *
 *                                                                                *
 * Redistribution and use in source and binary forms, with or without             *
 * modification, are permitted according to the terms listed in LICENSE           *
 * (http://tmva.sourceforge.net/LICENSE)                                          *
 *                                                                                *
 **********************************************************************************/

/*! \class TMVA::MethodBase
\ingroup TMVA

   Virtual base Class for all MVA method

   MethodBase hosts several specific evaluation methods.

   The kind of MVA that provides optimal performance in an analysis strongly
   depends on the particular application. The evaluation factory provides a
   number of numerical benchmark results to directly assess the performance
   of the MVA training on the independent test sample. These are:

  - The _signal efficiency_ at three representative background efficiencies
    (which is 1 &minus; rejection).
  - The _significance_ of an MVA estimator, defined by the difference
    between the MVA mean values for signal and background, divided by the
    quadratic sum of their root mean squares.
  - The _separation_ of an MVA _x_, defined by the integral
    \f[
    \frac{1}{2} \int \frac{(S(x) - B(x))^2}{(S(x) + B(x))} dx
    \f]
    where
    \f$ S(x) \f$ and \f$ B(x) \f$ are the signal and background distributions,
    respectively. The separation is zero for identical signal and background MVA
    shapes, and it is one for disjunctive shapes.
  - The average, \f$ \int x \mu (S(x)) dx \f$, of the signal \f$ \mu_{transform} \f$.
    The \f$ \mu_{transform} \f$ of an MVA denotes the transformation that yields
    a uniform background distribution. In this way, the signal distributions
    \f$ S(x) \f$ can be directly compared among the various MVAs. The stronger
    \f$ S(x) \f$ peaks towards one, the better is the discrimination of the MVA.
    The \f$ \mu_{transform} \f$  is
   [documented here](http://tel.ccsd.cnrs.fr/documents/archives0/00/00/29/91/index_fr.html).

   The MVA standard output also prints the linear correlation coefficients between
   signal and background, which can be useful to eliminate variables that exhibit too
   strong correlations.
*/

#include "TMVA/MethodBase.h"

#include "TMVA/Config.h"
#include "TMVA/Configurable.h"
#include "TMVA/DataSetInfo.h"
#include "TMVA/DataSet.h"
#include "TMVA/Factory.h"
#include "TMVA/IMethod.h"
#include "TMVA/MsgLogger.h"
#include "TMVA/PDF.h"
#include "TMVA/Ranking.h"
#include "TMVA/DataLoader.h"
#include "TMVA/Tools.h"
#include "TMVA/Results.h"
#include "TMVA/ResultsClassification.h"
#include "TMVA/ResultsRegression.h"
#include "TMVA/ResultsMulticlass.h"
#include "TMVA/RootFinder.h"
#include "TMVA/Timer.h"
#include "TMVA/TSpline1.h"
#include "TMVA/Types.h"
#include "TMVA/VariableDecorrTransform.h"
#include "TMVA/VariableGaussTransform.h"
#include "TMVA/VariableIdentityTransform.h"
#include "TMVA/VariableInfo.h"
#include "TMVA/VariableNormalizeTransform.h"
#include "TMVA/VariablePCATransform.h"
#include "TMVA/VariableTransform.h"
#include "TMVA/Version.h"

#include "TROOT.h"
#include "TSystem.h"
#include "TObjString.h"
#include "TQObject.h"
#include "TSpline.h"
#include "TMatrix.h"
#include "TMath.h"
#include "TH1F.h"
#include "TH2F.h"
#include "TFile.h"
#include "TGraph.h"
#include "TXMLEngine.h"

#include <iomanip>
#include <iostream>
#include <fstream>
#include <sstream>
#include <cstdlib>
#include <algorithm>
#include <limits>


ClassImp(TMVA::MethodBase);

using std::endl;
using std::atof;

//const Int_t    MethodBase_MaxIterations_ = 200;
const Bool_t   Use_Splines_for_Eff_      = kTRUE;

//const Int_t    NBIN_HIST_PLOT = 100;
const Int_t    NBIN_HIST_HIGH = 10000;

#ifdef _WIN32
/* Disable warning C4355: 'this' : used in base member initializer list */
#pragma warning ( disable : 4355 )
#endif


#include "TMultiGraph.h"

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

TMVA::IPythonInteractive::IPythonInteractive() : fMultiGraph(new TMultiGraph())
{
   fNumGraphs = 0;
   fIndex = 0;
}

////////////////////////////////////////////////////////////////////////////////
/// standard destructor
TMVA::IPythonInteractive::~IPythonInteractive()
{
   if (fMultiGraph){
      delete fMultiGraph;
      fMultiGraph = nullptr;
   }
   return;
}

////////////////////////////////////////////////////////////////////////////////
/// This function gets some title and it creates a TGraph for every title.
/// It also sets up the style for every TGraph. All graphs are added to a single TMultiGraph.
///
///  \param[in] graphTitles vector of titles

void TMVA::IPythonInteractive::Init(std::vector<TString>& graphTitles)
{
  if (fNumGraphs!=0){
    std::cerr << kERROR << "IPythonInteractive::Init: already initialized..." << std::endl;
    return;
  }
  Int_t color = 2;
  for(auto& title : graphTitles){
    fGraphs.push_back( new TGraph() );
    fGraphs.back()->SetTitle(title);
    fGraphs.back()->SetName(title);
    fGraphs.back()->SetFillColor(color);
    fGraphs.back()->SetLineColor(color);
    fGraphs.back()->SetMarkerColor(color);
    fMultiGraph->Add(fGraphs.back());
    color      += 2;
    fNumGraphs += 1;
  }
  return;
}

////////////////////////////////////////////////////////////////////////////////
/// This function sets the point number to 0 for all graphs.

void TMVA::IPythonInteractive::ClearGraphs()
{
   for(Int_t i=0; i<fNumGraphs; i++){
      fGraphs[i]->Set(0);
   }
}

////////////////////////////////////////////////////////////////////////////////
/// This function is used only in 2 TGraph case, and it will add new data points to graphs.
///
/// \param[in] x the x coordinate
/// \param[in] y1 the y coordinate for the first TGraph
/// \param[in] y2 the y coordinate for the second TGraph

void TMVA::IPythonInteractive::AddPoint(Double_t x, Double_t y1, Double_t y2)
{
   fGraphs[0]->Set(fIndex+1);
   fGraphs[1]->Set(fIndex+1);
   fGraphs[0]->SetPoint(fIndex, x, y1);
   fGraphs[1]->SetPoint(fIndex, x, y2);
   fIndex++;
   return;
}

////////////////////////////////////////////////////////////////////////////////
/// This function can add data points to as many TGraphs as we have.
///
/// \param[in] dat vector of data points. The dat[0] contains the x coordinate,
///            dat[1] contains the y coordinate for first TGraph, dat[2] for second, ...

void TMVA::IPythonInteractive::AddPoint(std::vector<Double_t>& dat)
{
  for(Int_t i=0; i<fNumGraphs;i++){
    fGraphs[i]->Set(fIndex+1);
    fGraphs[i]->SetPoint(fIndex, dat[0], dat[i+1]);
  }
   fIndex++;
   return;
}


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

TMVA::MethodBase::MethodBase( const TString& jobName,
                              Types::EMVA methodType,
                              const TString& methodTitle,
                              DataSetInfo& dsi,
                              const TString& theOption) :
   IMethod(),
   Configurable               ( theOption ),
   fTmpEvent                  ( 0 ),
   fRanking                   ( 0 ),
   fInputVars                 ( 0 ),
   fAnalysisType              ( Types::kNoAnalysisType ),
   fRegressionReturnVal       ( 0 ),
   fMulticlassReturnVal       ( 0 ),
   fDataSetInfo               ( dsi ),
   fSignalReferenceCut        ( 0.5 ),
   fSignalReferenceCutOrientation( 1. ),
   fVariableTransformType     ( Types::kSignal ),
   fJobName                   ( jobName ),
   fMethodName                ( methodTitle ),
   fMethodType                ( methodType ),
   fTestvar                   ( "" ),
   fTMVATrainingVersion       ( TMVA_VERSION_CODE ),
   fROOTTrainingVersion       ( ROOT_VERSION_CODE ),
   fConstructedFromWeightFile ( kFALSE ),
   fBaseDir                   ( 0 ),
   fMethodBaseDir             ( 0 ),
   fFile                      ( 0 ),
   fSilentFile                (kFALSE),
   fModelPersistence          (kTRUE),
   fWeightFile                ( "" ),
   fEffS                      ( 0 ),
   fDefaultPDF                ( 0 ),
   fMVAPdfS                   ( 0 ),
   fMVAPdfB                   ( 0 ),
   fSplS                      ( 0 ),
   fSplB                      ( 0 ),
   fSpleffBvsS                ( 0 ),
   fSplTrainS                 ( 0 ),
   fSplTrainB                 ( 0 ),
   fSplTrainEffBvsS           ( 0 ),
   fVarTransformString        ( "None" ),
   fTransformationPointer     ( 0 ),
   fTransformation            ( dsi, methodTitle ),
   fVerbose                   ( kFALSE ),
   fVerbosityLevelString      ( "Default" ),
   fHelp                      ( kFALSE ),
   fHasMVAPdfs                ( kFALSE ),
   fIgnoreNegWeightsInTraining( kFALSE ),
   fSignalClass               ( 0 ),
   fBackgroundClass           ( 0 ),
   fSplRefS                   ( 0 ),
   fSplRefB                   ( 0 ),
   fSplTrainRefS              ( 0 ),
   fSplTrainRefB              ( 0 ),
   fSetupCompleted            (kFALSE)
{
   SetTestvarName();
   fLogger->SetSource(GetName());

//    // default extension for weight files
}

////////////////////////////////////////////////////////////////////////////////
/// constructor used for Testing + Application of the MVA,
/// only (no training), using given WeightFiles

TMVA::MethodBase::MethodBase( Types::EMVA methodType,
                              DataSetInfo& dsi,
                              const TString& weightFile ) :
   IMethod(),
   Configurable(""),
   fTmpEvent                  ( 0 ),
   fRanking                   ( 0 ),
   fInputVars                 ( 0 ),
   fAnalysisType              ( Types::kNoAnalysisType ),
   fRegressionReturnVal       ( 0 ),
   fMulticlassReturnVal       ( 0 ),
   fDataSetInfo               ( dsi ),
   fSignalReferenceCut        ( 0.5 ),
   fVariableTransformType     ( Types::kSignal ),
   fJobName                   ( "" ),
   fMethodName                ( "MethodBase"  ),
   fMethodType                ( methodType ),
   fTestvar                   ( "" ),
   fTMVATrainingVersion       ( 0 ),
   fROOTTrainingVersion       ( 0 ),
   fConstructedFromWeightFile ( kTRUE ),
   fBaseDir                   ( 0 ),
   fMethodBaseDir             ( 0 ),
   fFile                      ( 0 ),
   fSilentFile                (kFALSE),
   fModelPersistence          (kTRUE),
   fWeightFile                ( weightFile ),
   fEffS                      ( 0 ),
   fDefaultPDF                ( 0 ),
   fMVAPdfS                   ( 0 ),
   fMVAPdfB                   ( 0 ),
   fSplS                      ( 0 ),
   fSplB                      ( 0 ),
   fSpleffBvsS                ( 0 ),
   fSplTrainS                 ( 0 ),
   fSplTrainB                 ( 0 ),
   fSplTrainEffBvsS           ( 0 ),
   fVarTransformString        ( "None" ),
   fTransformationPointer     ( 0 ),
   fTransformation            ( dsi, "" ),
   fVerbose                   ( kFALSE ),
   fVerbosityLevelString      ( "Default" ),
   fHelp                      ( kFALSE ),
   fHasMVAPdfs                ( kFALSE ),
   fIgnoreNegWeightsInTraining( kFALSE ),
   fSignalClass               ( 0 ),
   fBackgroundClass           ( 0 ),
   fSplRefS                   ( 0 ),
   fSplRefB                   ( 0 ),
   fSplTrainRefS              ( 0 ),
   fSplTrainRefB              ( 0 ),
   fSetupCompleted            (kFALSE)
{
   fLogger->SetSource(GetName());
//    // constructor used for Testing + Application of the MVA,
//    // only (no training), using given WeightFiles
}

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

TMVA::MethodBase::~MethodBase( void )
{
   // destructor
   if (!fSetupCompleted) Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Calling destructor of method which got never setup" << Endl;

   // destructor
   if (fInputVars != 0)  { fInputVars->clear(); delete fInputVars; }
   if (fRanking   != 0)  delete fRanking;

   // PDFs
   if (fDefaultPDF!= 0)  { delete fDefaultPDF; fDefaultPDF = 0; }
   if (fMVAPdfS   != 0)  { delete fMVAPdfS; fMVAPdfS = 0; }
   if (fMVAPdfB   != 0)  { delete fMVAPdfB; fMVAPdfB = 0; }

   // Splines
   if (fSplS)            { delete fSplS; fSplS = 0; }
   if (fSplB)            { delete fSplB; fSplB = 0; }
   if (fSpleffBvsS)      { delete fSpleffBvsS; fSpleffBvsS = 0; }
   if (fSplRefS)         { delete fSplRefS; fSplRefS = 0; }
   if (fSplRefB)         { delete fSplRefB; fSplRefB = 0; }
   if (fSplTrainRefS)    { delete fSplTrainRefS; fSplTrainRefS = 0; }
   if (fSplTrainRefB)    { delete fSplTrainRefB; fSplTrainRefB = 0; }
   if (fSplTrainEffBvsS) { delete fSplTrainEffBvsS; fSplTrainEffBvsS = 0; }

   for (Int_t i = 0; i < 2; i++ ) {
      if (fEventCollections.at(i)) {
         for (std::vector<Event*>::const_iterator it = fEventCollections.at(i)->begin();
              it != fEventCollections.at(i)->end(); ++it) {
            delete (*it);
         }
         delete fEventCollections.at(i);
         fEventCollections.at(i) = 0;
      }
   }

   if (fRegressionReturnVal) delete fRegressionReturnVal;
   if (fMulticlassReturnVal) delete fMulticlassReturnVal;
}

////////////////////////////////////////////////////////////////////////////////
/// setup of methods

void TMVA::MethodBase::SetupMethod()
{
   // setup of methods

   if (fSetupCompleted) Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Calling SetupMethod for the second time" << Endl;
   InitBase();
   DeclareBaseOptions();
   Init();
   DeclareOptions();
   fSetupCompleted = kTRUE;
}

////////////////////////////////////////////////////////////////////////////////
/// process all options
/// the "CheckForUnusedOptions" is done in an independent call, since it may be overridden by derived class
/// (sometimes, eg, fitters are used which can only be implemented during training phase)

void TMVA::MethodBase::ProcessSetup()
{
   ProcessBaseOptions();
   ProcessOptions();
}

////////////////////////////////////////////////////////////////////////////////
/// check may be overridden by derived class
/// (sometimes, eg, fitters are used which can only be implemented during training phase)

void TMVA::MethodBase::CheckSetup()
{
   CheckForUnusedOptions();
}

////////////////////////////////////////////////////////////////////////////////
/// default initialization called by all constructors

void TMVA::MethodBase::InitBase()
{
   SetConfigDescription( "Configuration options for classifier architecture and tuning" );

   fNbins              = gConfig().fVariablePlotting.fNbinsXOfROCCurve;
   fNbinsMVAoutput     = gConfig().fVariablePlotting.fNbinsMVAoutput;
   fNbinsH             = NBIN_HIST_HIGH;

   fSplTrainS          = 0;
   fSplTrainB          = 0;
   fSplTrainEffBvsS    = 0;
   fMeanS              = -1;
   fMeanB              = -1;
   fRmsS               = -1;
   fRmsB               = -1;
   fXmin               = DBL_MAX;
   fXmax               = -DBL_MAX;
   fTxtWeightsOnly     = kTRUE;
   fSplRefS            = 0;
   fSplRefB            = 0;

   fTrainTime          = -1.;
   fTestTime           = -1.;

   fRanking            = 0;

   // temporary until the move to DataSet is complete
   fInputVars = new std::vector<TString>;
   for (UInt_t ivar=0; ivar<GetNvar(); ivar++) {
      fInputVars->push_back(DataInfo().GetVariableInfo(ivar).GetLabel());
   }
   fRegressionReturnVal = 0;
   fMulticlassReturnVal = 0;

   fEventCollections.resize( 2 );
   fEventCollections.at(0) = 0;
   fEventCollections.at(1) = 0;

   // retrieve signal and background class index
   if (DataInfo().GetClassInfo("Signal") != 0) {
      fSignalClass = DataInfo().GetClassInfo("Signal")->GetNumber();
   }
   if (DataInfo().GetClassInfo("Background") != 0) {
      fBackgroundClass = DataInfo().GetClassInfo("Background")->GetNumber();
   }

   SetConfigDescription( "Configuration options for MVA method" );
   SetConfigName( TString("Method") + GetMethodTypeName() );
}

////////////////////////////////////////////////////////////////////////////////
/// define the options (their key words) that can be set in the option string
/// here the options valid for ALL MVA methods are declared.
///
/// know options:
///
///  - VariableTransform=None,Decorrelated,PCA  to use transformed variables
///    instead of the original ones
///  - VariableTransformType=Signal,Background  which decorrelation matrix to use
///    in the method. Only the Likelihood
///    Method can make proper use of independent
///    transformations of signal and background
///  - fNbinsMVAPdf   = 50 Number of bins used to create a PDF of MVA
///  - fNsmoothMVAPdf =  2 Number of times a histogram is smoothed before creating the PDF
///  - fHasMVAPdfs         create PDFs for the MVA outputs
///  - V                   for Verbose output (!V) for non verbos
///  - H                   for Help message

void TMVA::MethodBase::DeclareBaseOptions()
{
   DeclareOptionRef( fVerbose, "V", "Verbose output (short form of \"VerbosityLevel\" below - overrides the latter one)" );

   DeclareOptionRef( fVerbosityLevelString="Default", "VerbosityLevel", "Verbosity level" );
   AddPreDefVal( TString("Default") ); // uses default defined in MsgLogger header
   AddPreDefVal( TString("Debug")   );
   AddPreDefVal( TString("Verbose") );
   AddPreDefVal( TString("Info")    );
   AddPreDefVal( TString("Warning") );
   AddPreDefVal( TString("Error")   );
   AddPreDefVal( TString("Fatal")   );

   // If True (default): write all training results (weights) as text files only;
   // if False: write also in ROOT format (not available for all methods - will abort if not
   fTxtWeightsOnly = kTRUE;  // OBSOLETE !!!
   fNormalise      = kFALSE; // OBSOLETE !!!

   DeclareOptionRef( fVarTransformString, "VarTransform", "List of variable transformations performed before training, e.g., \"D_Background,P_Signal,G,N_AllClasses\" for: \"Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)\"" );

   DeclareOptionRef( fHelp, "H", "Print method-specific help message" );

   DeclareOptionRef( fHasMVAPdfs, "CreateMVAPdfs", "Create PDFs for classifier outputs (signal and background)" );

   DeclareOptionRef( fIgnoreNegWeightsInTraining, "IgnoreNegWeightsInTraining",
                     "Events with negative weights are ignored in the training (but are included for testing and performance evaluation)" );
}

////////////////////////////////////////////////////////////////////////////////
/// the option string is decoded, for available options see "DeclareOptions"

void TMVA::MethodBase::ProcessBaseOptions()
{
   if (HasMVAPdfs()) {
      // setting the default bin num... maybe should be static ? ==> Please no static (JS)
      // You can't use the logger in the constructor!!! Log() << kINFO << "Create PDFs" << Endl;
      // reading every PDF's definition and passing the option string to the next one to be read and marked
      fDefaultPDF = new PDF( TString(GetName())+"_PDF", GetOptions(), "MVAPdf" );
      fDefaultPDF->DeclareOptions();
      fDefaultPDF->ParseOptions();
      fDefaultPDF->ProcessOptions();
      fMVAPdfB = new PDF( TString(GetName())+"_PDFBkg", fDefaultPDF->GetOptions(), "MVAPdfBkg", fDefaultPDF );
      fMVAPdfB->DeclareOptions();
      fMVAPdfB->ParseOptions();
      fMVAPdfB->ProcessOptions();
      fMVAPdfS = new PDF( TString(GetName())+"_PDFSig", fMVAPdfB->GetOptions(),    "MVAPdfSig", fDefaultPDF );
      fMVAPdfS->DeclareOptions();
      fMVAPdfS->ParseOptions();
      fMVAPdfS->ProcessOptions();

      // the final marked option string is written back to the original methodbase
      SetOptions( fMVAPdfS->GetOptions() );
   }

   TMVA::CreateVariableTransforms( fVarTransformString,
                                               DataInfo(),
                                               GetTransformationHandler(),
                                               Log() );

   if (!HasMVAPdfs()) {
      if (fDefaultPDF!= 0) { delete fDefaultPDF; fDefaultPDF = 0; }
      if (fMVAPdfS   != 0) { delete fMVAPdfS; fMVAPdfS = 0; }
      if (fMVAPdfB   != 0) { delete fMVAPdfB; fMVAPdfB = 0; }
   }

   if (fVerbose) { // overwrites other settings
      fVerbosityLevelString = TString("Verbose");
      Log().SetMinType( kVERBOSE );
   }
   else if (fVerbosityLevelString == "Debug"   ) Log().SetMinType( kDEBUG );
   else if (fVerbosityLevelString == "Verbose" ) Log().SetMinType( kVERBOSE );
   else if (fVerbosityLevelString == "Info"    ) Log().SetMinType( kINFO );
   else if (fVerbosityLevelString == "Warning" ) Log().SetMinType( kWARNING );
   else if (fVerbosityLevelString == "Error"   ) Log().SetMinType( kERROR );
   else if (fVerbosityLevelString == "Fatal"   ) Log().SetMinType( kFATAL );
   else if (fVerbosityLevelString != "Default" ) {
      Log() << kFATAL << "<ProcessOptions> Verbosity level type '"
            << fVerbosityLevelString << "' unknown." << Endl;
   }
   Event::SetIgnoreNegWeightsInTraining(fIgnoreNegWeightsInTraining);
}

////////////////////////////////////////////////////////////////////////////////
/// options that are used ONLY for the READER to ensure backward compatibility
/// they are hence without any effect (the reader is only reading the training
/// options that HAD been used at the training of the .xml weight file at hand

void TMVA::MethodBase::DeclareCompatibilityOptions()
{
   DeclareOptionRef( fNormalise=kFALSE, "Normalise", "Normalise input variables" ); // don't change the default !!!
   DeclareOptionRef( fUseDecorr=kFALSE, "D", "Use-decorrelated-variables flag" );
   DeclareOptionRef( fVariableTransformTypeString="Signal", "VarTransformType",
                     "Use signal or background events to derive for variable transformation (the transformation is applied on both types of, course)" );
   AddPreDefVal( TString("Signal") );
   AddPreDefVal( TString("Background") );
   DeclareOptionRef( fTxtWeightsOnly=kTRUE, "TxtWeightFilesOnly", "If True: write all training results (weights) as text files (False: some are written in ROOT format)" );
   // Why on earth ?? was this here? Was the verbosity level option  meant to 'disappear? Not a good idea i think..
   // DeclareOptionRef( fVerbosityLevelString="Default", "VerboseLevel", "Verbosity level" );
   // AddPreDefVal( TString("Default") ); // uses default defined in MsgLogger header
   // AddPreDefVal( TString("Debug")   );
   // AddPreDefVal( TString("Verbose") );
   // AddPreDefVal( TString("Info")    );
   // AddPreDefVal( TString("Warning") );
   // AddPreDefVal( TString("Error")   );
   // AddPreDefVal( TString("Fatal")   );
   DeclareOptionRef( fNbinsMVAPdf   = 60, "NbinsMVAPdf",   "Number of bins used for the PDFs of classifier outputs" );
   DeclareOptionRef( fNsmoothMVAPdf = 2,  "NsmoothMVAPdf", "Number of smoothing iterations for classifier PDFs" );
}


////////////////////////////////////////////////////////////////////////////////
/// call the Optimizer with the set of parameters and ranges that
/// are meant to be tuned.

std::map<TString,Double_t>  TMVA::MethodBase::OptimizeTuningParameters(TString /* fomType */ , TString /* fitType */)
{
   // this is just a dummy...  needs to be implemented for each method
   // individually (as long as we don't have it automatized via the
   // configuration string

   Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Parameter optimization is not yet implemented for method "
         << GetName() << Endl;
   Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Currently we need to set hardcoded which parameter is tuned in which ranges"<<Endl;

   std::map<TString,Double_t> tunedParameters;
   tunedParameters.size(); // just to get rid of "unused" warning
   return tunedParameters;

}

////////////////////////////////////////////////////////////////////////////////
/// set the tuning parameters according to the argument
/// This is just a dummy .. have a look at the MethodBDT how you could
/// perhaps implement the same thing for the other Classifiers..

void TMVA::MethodBase::SetTuneParameters(std::map<TString,Double_t> /* tuneParameters */)
{
}

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

void TMVA::MethodBase::TrainMethod()
{
   Data()->SetCurrentType(Types::kTraining);
   Event::SetIsTraining(kTRUE); // used to set negative event weights to zero if chosen to do so

   // train the MVA method
   if (Help()) PrintHelpMessage();

   // all histograms should be created in the method's subdirectory
   if(!IsSilentFile()) BaseDir()->cd();

   // once calculate all the transformation (e.g. the sequence of Decorr:Gauss:Decorr)
   //    needed for this classifier
   GetTransformationHandler().CalcTransformations(Data()->GetEventCollection());

   // call training of derived MVA
   Log() << kDEBUG //<<Form("\tDataset[%s] : ",DataInfo().GetName())
    << "Begin training" << Endl;
   Long64_t nEvents = Data()->GetNEvents();
   Timer traintimer( nEvents, GetName(), kTRUE );
   Train();
   Log() << kDEBUG //<<Form("Dataset[%s] : ",DataInfo().GetName()
    << "\tEnd of training                                              " << Endl;
   SetTrainTime(traintimer.ElapsedSeconds());
   Log() << kINFO //<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Elapsed time for training with " << nEvents <<  " events: "
         << traintimer.GetElapsedTime() << "         " << Endl;

   Log() << kDEBUG //<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "\tCreate MVA output for ";

   // create PDFs for the signal and background MVA distributions (if required)
   if (DoMulticlass()) {
      Log() <<Form("[%s] : ",DataInfo().GetName())<< "Multiclass classification on training sample" << Endl;
      AddMulticlassOutput(Types::kTraining);
   }
   else if (!DoRegression()) {

      Log() <<Form("[%s] : ",DataInfo().GetName())<< "classification on training sample" << Endl;
      AddClassifierOutput(Types::kTraining);
      if (HasMVAPdfs()) {
         CreateMVAPdfs();
         AddClassifierOutputProb(Types::kTraining);
      }

   } else {

      Log() <<Form("Dataset[%s] : ",DataInfo().GetName())<< "regression on training sample" << Endl;
      AddRegressionOutput( Types::kTraining );

      if (HasMVAPdfs() ) {
         Log() <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Create PDFs" << Endl;
         CreateMVAPdfs();
      }
   }

   // write the current MVA state into stream
   // produced are one text file and one ROOT file
   if (fModelPersistence ) WriteStateToFile();

   // produce standalone make class (presently only supported for classification)
   if ((!DoRegression()) && (fModelPersistence)) MakeClass();

   // write additional monitoring histograms to main target file (not the weight file)
   // again, make sure the histograms go into the method's subdirectory
   if(!IsSilentFile())
   {
       BaseDir()->cd();
       WriteMonitoringHistosToFile();
   }
}

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

void TMVA::MethodBase::GetRegressionDeviation(UInt_t tgtNum, Types::ETreeType type, Double_t& stddev, Double_t& stddev90Percent ) const
{
   if (!DoRegression()) Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Trying to use GetRegressionDeviation() with a classification job" << Endl;
   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Create results for " << (type==Types::kTraining?"training":"testing") << Endl;
   ResultsRegression* regRes = (ResultsRegression*)Data()->GetResults(GetMethodName(), Types::kTesting, Types::kRegression);
   bool truncate = false;
   TH1F* h1 = regRes->QuadraticDeviation( tgtNum , truncate, 1.);
   stddev = sqrt(h1->GetMean());
   truncate = true;
   Double_t yq[1], xq[]={0.9};
   h1->GetQuantiles(1,yq,xq);
   TH1F* h2 = regRes->QuadraticDeviation( tgtNum , truncate, yq[0]);
   stddev90Percent = sqrt(h2->GetMean());
   delete h1;
   delete h2;
}

////////////////////////////////////////////////////////////////////////////////
/// prepare tree branch with the method's discriminating variable

void TMVA::MethodBase::AddRegressionOutput(Types::ETreeType type)
{
   Data()->SetCurrentType(type);

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Create results for " << (type==Types::kTraining?"training":"testing") << Endl;

   ResultsRegression* regRes = (ResultsRegression*)Data()->GetResults(GetMethodName(), type, Types::kRegression);

   Long64_t nEvents = Data()->GetNEvents();

   // use timer
   Timer timer( nEvents, GetName(), kTRUE );
   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName()) << "Evaluation of " << GetMethodName() << " on "
         << (type==Types::kTraining?"training":"testing") << " sample" << Endl;

   regRes->Resize( nEvents );

   // Drawing the progress bar every event was causing a huge slowdown in the evaluation time
   // So we set some parameters to draw the progress bar a total of totalProgressDraws, i.e. only draw every 1 in 100

   Int_t totalProgressDraws = 100; // total number of times to update the progress bar
   Int_t drawProgressEvery = 1;    // draw every nth event such that we have a total of totalProgressDraws
   if(nEvents >= totalProgressDraws) drawProgressEvery = nEvents/totalProgressDraws;

   for (Int_t ievt=0; ievt<nEvents; ievt++) {

      Data()->SetCurrentEvent(ievt);
      std::vector< Float_t > vals = GetRegressionValues();
      regRes->SetValue( vals, ievt );

      // Only draw the progress bar once in a while, doing this every event causes the evaluation to be ridiculously slow
      if(ievt % drawProgressEvery == 0 || ievt==nEvents-1) timer.DrawProgressBar( ievt );
   }

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Elapsed time for evaluation of " << nEvents <<  " events: "
         << timer.GetElapsedTime() << "       " << Endl;

   // store time used for testing
   if (type==Types::kTesting)
      SetTestTime(timer.ElapsedSeconds());

   TString histNamePrefix(GetTestvarName());
   histNamePrefix += (type==Types::kTraining?"train":"test");
   regRes->CreateDeviationHistograms( histNamePrefix );
}

////////////////////////////////////////////////////////////////////////////////
/// prepare tree branch with the method's discriminating variable

void TMVA::MethodBase::AddMulticlassOutput(Types::ETreeType type)
{
   Data()->SetCurrentType(type);

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Create results for " << (type==Types::kTraining?"training":"testing") << Endl;

   ResultsMulticlass* resMulticlass = dynamic_cast<ResultsMulticlass*>(Data()->GetResults(GetMethodName(), type, Types::kMulticlass));
   if (!resMulticlass) Log() << kFATAL<<Form("Dataset[%s] : ",DataInfo().GetName())<< "unable to create pointer in AddMulticlassOutput, exiting."<<Endl;

   Long64_t nEvents = Data()->GetNEvents();

   // use timer
   Timer timer( nEvents, GetName(), kTRUE );

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Multiclass evaluation of " << GetMethodName() << " on "
         << (type==Types::kTraining?"training":"testing") << " sample" << Endl;

   resMulticlass->Resize( nEvents );
   for (Int_t ievt=0; ievt<nEvents; ievt++) {
      Data()->SetCurrentEvent(ievt);
      std::vector< Float_t > vals = GetMulticlassValues();
      resMulticlass->SetValue( vals, ievt );
      timer.DrawProgressBar( ievt );
   }

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Elapsed time for evaluation of " << nEvents <<  " events: "
         << timer.GetElapsedTime() << "       " << Endl;

   // store time used for testing
   if (type==Types::kTesting)
      SetTestTime(timer.ElapsedSeconds());

   TString histNamePrefix(GetTestvarName());
   histNamePrefix += (type==Types::kTraining?"_Train":"_Test");

   resMulticlass->CreateMulticlassHistos( histNamePrefix, fNbinsMVAoutput, fNbinsH );
   resMulticlass->CreateMulticlassPerformanceHistos(histNamePrefix);
}

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

void TMVA::MethodBase::NoErrorCalc(Double_t* const err, Double_t* const errUpper) {
   if (err) *err=-1;
   if (errUpper) *errUpper=-1;
}

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

Double_t TMVA::MethodBase::GetMvaValue( const Event* const ev, Double_t* err, Double_t* errUpper ) {
   fTmpEvent = ev;
   Double_t val = GetMvaValue(err, errUpper);
   fTmpEvent = 0;
   return val;
}

////////////////////////////////////////////////////////////////////////////////
/// uses a pre-set cut on the MVA output (SetSignalReferenceCut and SetSignalReferenceCutOrientation)
/// for a quick determination if an event would be selected as signal or background

Bool_t TMVA::MethodBase::IsSignalLike() {
   return GetMvaValue()*GetSignalReferenceCutOrientation() > GetSignalReferenceCut()*GetSignalReferenceCutOrientation() ? kTRUE : kFALSE;
}
////////////////////////////////////////////////////////////////////////////////
/// uses a pre-set cut on the MVA output (SetSignalReferenceCut and SetSignalReferenceCutOrientation)
/// for a quick determination if an event with this mva output value would be selected as signal or background

Bool_t TMVA::MethodBase::IsSignalLike(Double_t mvaVal) {
   return mvaVal*GetSignalReferenceCutOrientation() > GetSignalReferenceCut()*GetSignalReferenceCutOrientation() ? kTRUE : kFALSE;
}

////////////////////////////////////////////////////////////////////////////////
/// prepare tree branch with the method's discriminating variable

void TMVA::MethodBase::AddClassifierOutput( Types::ETreeType type )
{
   Data()->SetCurrentType(type);

   ResultsClassification* clRes =
      (ResultsClassification*)Data()->GetResults(GetMethodName(), type, Types::kClassification );

   Long64_t nEvents =  Data()->GetNEvents();
   clRes->Resize( nEvents );

   // use timer
   Timer timer( nEvents, GetName(), kTRUE );
   std::vector<Double_t> mvaValues = GetMvaValues(0, nEvents, true);

   // store time used for testing
   if (type==Types::kTesting)
      SetTestTime(timer.ElapsedSeconds());

   // load mva values to results object
   for (Int_t ievt=0; ievt<nEvents; ievt++) {
      clRes->SetValue( mvaValues[ievt], ievt );
   }
}

////////////////////////////////////////////////////////////////////////////////
/// get all the MVA values for the events of the current Data type
std::vector<Double_t> TMVA::MethodBase::GetMvaValues(Long64_t firstEvt, Long64_t lastEvt, Bool_t logProgress)
{

   Long64_t nEvents = Data()->GetNEvents();
   if (firstEvt > lastEvt || lastEvt > nEvents) lastEvt = nEvents;
   if (firstEvt < 0) firstEvt = 0;
   std::vector<Double_t> values(lastEvt-firstEvt);
   // log in case of looping on all the events
   nEvents = values.size();

   // use timer
   Timer timer( nEvents, GetName(), kTRUE );

   if (logProgress)
      Log() << kHEADER << Form("[%s] : ",DataInfo().GetName())
            << "Evaluation of " << GetMethodName() << " on "
            << (Data()->GetCurrentType() == Types::kTraining ? "training" : "testing")
            << " sample (" << nEvents << " events)" << Endl;

   for (Int_t ievt=firstEvt; ievt<lastEvt; ievt++) {
      Data()->SetCurrentEvent(ievt);
      values[ievt] = GetMvaValue();

      // print progress
      if (logProgress) {
         Int_t modulo = Int_t(nEvents/100);
         if (modulo <= 0 ) modulo = 1;
         if (ievt%modulo == 0) timer.DrawProgressBar( ievt );
      }
   }
   if (logProgress) {
     Log() << kINFO //<<Form("Dataset[%s] : ",DataInfo().GetName())
        << "Elapsed time for evaluation of " << nEvents <<  " events: "
         << timer.GetElapsedTime() << "       " << Endl;
   }

   return values;
}

////////////////////////////////////////////////////////////////////////////////
/// prepare tree branch with the method's discriminating variable

void TMVA::MethodBase::AddClassifierOutputProb( Types::ETreeType type )
{
   Data()->SetCurrentType(type);

   ResultsClassification* mvaProb =
      (ResultsClassification*)Data()->GetResults(TString("prob_")+GetMethodName(), type, Types::kClassification );

   Long64_t nEvents = Data()->GetNEvents();

   // use timer
   Timer timer( nEvents, GetName(), kTRUE );

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName()) << "Evaluation of " << GetMethodName() << " on "
         << (type==Types::kTraining?"training":"testing") << " sample" << Endl;

   mvaProb->Resize( nEvents );
   for (Int_t ievt=0; ievt<nEvents; ievt++) {

      Data()->SetCurrentEvent(ievt);
      Float_t proba = ((Float_t)GetProba( GetMvaValue(), 0.5 ));
      if (proba < 0) break;
      mvaProb->SetValue( proba, ievt );

      // print progress
      Int_t modulo = Int_t(nEvents/100);
      if (modulo <= 0 ) modulo = 1;
      if (ievt%modulo == 0) timer.DrawProgressBar( ievt );
   }

   Log() << kDEBUG <<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Elapsed time for evaluation of " << nEvents <<  " events: "
         << timer.GetElapsedTime() << "       " << Endl;
}

////////////////////////////////////////////////////////////////////////////////
/// calculate <sum-of-deviation-squared> of regression output versus "true" value from test sample
///
///  - bias = average deviation
///  - dev  = average absolute deviation
///  - rms  = rms of deviation

void TMVA::MethodBase::TestRegression( Double_t& bias, Double_t& biasT,
                                       Double_t& dev,  Double_t& devT,
                                       Double_t& rms,  Double_t& rmsT,
                                       Double_t& mInf, Double_t& mInfT,
                                       Double_t& corr,
                                       Types::ETreeType type )
{
   Types::ETreeType savedType = Data()->GetCurrentType();
   Data()->SetCurrentType(type);

   bias = 0; biasT = 0; dev = 0; devT = 0; rms = 0; rmsT = 0;
   Double_t sumw = 0;
   Double_t m1 = 0, m2 = 0, s1 = 0, s2 = 0, s12 = 0; // for correlation
   const Int_t nevt = GetNEvents();
   Float_t* rV = new Float_t[nevt];
   Float_t* tV = new Float_t[nevt];
   Float_t* wV = new Float_t[nevt];
   Float_t  xmin = 1e30, xmax = -1e30;
   Log() << kINFO << "Calculate regression for all events" << Endl;
   Timer timer( nevt, GetName(), kTRUE );
   for (Long64_t ievt=0; ievt<nevt; ievt++) {

      const Event* ev = Data()->GetEvent(ievt); // NOTE: need untransformed event here !
      Float_t t = ev->GetTarget(0);
      Float_t w = ev->GetWeight();
      Float_t r = GetRegressionValues()[0];
      Float_t d = (r-t);

      // find min/max
      xmin = TMath::Min(xmin, TMath::Min(t, r));
      xmax = TMath::Max(xmax, TMath::Max(t, r));

      // store for truncated RMS computation
      rV[ievt] = r;
      tV[ievt] = t;
      wV[ievt] = w;

      // compute deviation-squared
      sumw += w;
      bias += w * d;
      dev  += w * TMath::Abs(d);
      rms  += w * d * d;

      // compute correlation between target and regression estimate
      m1  += t*w; s1 += t*t*w;
      m2  += r*w; s2 += r*r*w;
      s12 += t*r;
      if ((ievt & 0xFF) == 0) timer.DrawProgressBar(ievt);
   }
   timer.DrawProgressBar(nevt - 1);
   Log() << kINFO << "Elapsed time for evaluation of " << nevt <<  " events: "
         << timer.GetElapsedTime() << "       " << Endl;

   // standard quantities
   bias /= sumw;
   dev  /= sumw;
   rms  /= sumw;
   rms  = TMath::Sqrt(rms - bias*bias);

   // correlation
   m1   /= sumw;
   m2   /= sumw;
   corr  = s12/sumw - m1*m2;
   corr /= TMath::Sqrt( (s1/sumw - m1*m1) * (s2/sumw - m2*m2) );

   // create histogram required for computation of mutual information
   TH2F* hist  = new TH2F( "hist",  "hist",  150, xmin, xmax, 100, xmin, xmax );
   TH2F* histT = new TH2F( "histT", "histT", 150, xmin, xmax, 100, xmin, xmax );

   // compute truncated RMS and fill histogram
   Double_t devMax = bias + 2*rms;
   Double_t devMin = bias - 2*rms;
   sumw = 0;
   int ic=0;
   for (Long64_t ievt=0; ievt<nevt; ievt++) {
      Float_t d = (rV[ievt] - tV[ievt]);
      hist->Fill( rV[ievt], tV[ievt], wV[ievt] );
      if (d >= devMin && d <= devMax) {
         sumw  += wV[ievt];
         biasT += wV[ievt] * d;
         devT  += wV[ievt] * TMath::Abs(d);
         rmsT  += wV[ievt] * d * d;
         histT->Fill( rV[ievt], tV[ievt], wV[ievt] );
         ic++;
      }
   }
   biasT /= sumw;
   devT  /= sumw;
   rmsT  /= sumw;
   rmsT  = TMath::Sqrt(rmsT - biasT*biasT);
   mInf  = gTools().GetMutualInformation( *hist );
   mInfT = gTools().GetMutualInformation( *histT );

   delete hist;
   delete histT;

   delete [] rV;
   delete [] tV;
   delete [] wV;

   Data()->SetCurrentType(savedType);
}


////////////////////////////////////////////////////////////////////////////////
/// test multiclass classification

void TMVA::MethodBase::TestMulticlass()
{
   ResultsMulticlass* resMulticlass = dynamic_cast<ResultsMulticlass*>(Data()->GetResults(GetMethodName(), Types::kTesting, Types::kMulticlass));
   if (!resMulticlass) Log() << kFATAL<<Form("Dataset[%s] : ",DataInfo().GetName())<< "unable to create pointer in TestMulticlass, exiting."<<Endl;

   // GA evaluation of best cut for sig eff * sig pur. Slow, disabled for now.
   // Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Determine optimal multiclass cuts for test
   // data..." << Endl; for (UInt_t icls = 0; icls<DataInfo().GetNClasses(); ++icls) {
   //    resMulticlass->GetBestMultiClassCuts(icls);
   // }

   // Create histograms for use in TMVA GUI
   TString histNamePrefix(GetTestvarName());
   TString histNamePrefixTest{histNamePrefix + "_Test"};
   TString histNamePrefixTrain{histNamePrefix + "_Train"};

   resMulticlass->CreateMulticlassHistos(histNamePrefixTest, fNbinsMVAoutput, fNbinsH);
   resMulticlass->CreateMulticlassPerformanceHistos(histNamePrefixTest);

   resMulticlass->CreateMulticlassHistos(histNamePrefixTrain, fNbinsMVAoutput, fNbinsH);
   resMulticlass->CreateMulticlassPerformanceHistos(histNamePrefixTrain);
}


////////////////////////////////////////////////////////////////////////////////
/// initialization

void TMVA::MethodBase::TestClassification()
{
   Data()->SetCurrentType(Types::kTesting);

   ResultsClassification* mvaRes = dynamic_cast<ResultsClassification*>
      ( Data()->GetResults(GetMethodName(),Types::kTesting, Types::kClassification) );

   // sanity checks: tree must exist, and theVar must be in tree
   if (0==mvaRes && !(GetMethodTypeName().Contains("Cuts"))) {
      Log()<<Form("Dataset[%s] : ",DataInfo().GetName()) << "mvaRes " << mvaRes << " GetMethodTypeName " << GetMethodTypeName()
           << " contains " << !(GetMethodTypeName().Contains("Cuts")) << Endl;
      Log() << kFATAL<<Form("Dataset[%s] : ",DataInfo().GetName()) << "<TestInit> Test variable " << GetTestvarName()
            << " not found in tree" << Endl;
   }

   // basic statistics operations are made in base class
   gTools().ComputeStat( GetEventCollection(Types::kTesting), mvaRes->GetValueVector(),
                         fMeanS, fMeanB, fRmsS, fRmsB, fXmin, fXmax, fSignalClass );

   // choose reasonable histogram ranges, by removing outliers
   Double_t nrms = 10;
   fXmin = TMath::Max( TMath::Min( fMeanS - nrms*fRmsS, fMeanB - nrms*fRmsB ), fXmin );
   fXmax = TMath::Min( TMath::Max( fMeanS + nrms*fRmsS, fMeanB + nrms*fRmsB ), fXmax );

   // determine cut orientation
   fCutOrientation = (fMeanS > fMeanB) ? kPositive : kNegative;

   // fill 2 types of histograms for the various analyses
   // this one is for actual plotting

   Double_t sxmax = fXmax+0.00001;

   // classifier response distributions for training sample
   // MVA plots used for graphics representation (signal)
   TString TestvarName;
   if(IsSilentFile())
      {
         TestvarName=Form("[%s]%s",DataInfo().GetName(),GetTestvarName().Data());
      }else
      {
         TestvarName=GetTestvarName();
      }
   TH1* mva_s = new TH1D( TestvarName + "_S",TestvarName + "_S", fNbinsMVAoutput, fXmin, sxmax );
   TH1* mva_b = new TH1D( TestvarName + "_B",TestvarName + "_B", fNbinsMVAoutput, fXmin, sxmax );
   mvaRes->Store(mva_s, "MVA_S");
   mvaRes->Store(mva_b, "MVA_B");
   mva_s->Sumw2();
   mva_b->Sumw2();

   TH1* proba_s = 0;
   TH1* proba_b = 0;
   TH1* rarity_s = 0;
   TH1* rarity_b = 0;
   if (HasMVAPdfs()) {
      // P(MVA) plots used for graphics representation
      proba_s = new TH1D( TestvarName + "_Proba_S", TestvarName + "_Proba_S", fNbinsMVAoutput, 0.0, 1.0 );
      proba_b = new TH1D( TestvarName + "_Proba_B", TestvarName + "_Proba_B", fNbinsMVAoutput, 0.0, 1.0 );
      mvaRes->Store(proba_s, "Prob_S");
      mvaRes->Store(proba_b, "Prob_B");
      proba_s->Sumw2();
      proba_b->Sumw2();

      // R(MVA) plots used for graphics representation
      rarity_s = new TH1D( TestvarName + "_Rarity_S", TestvarName + "_Rarity_S", fNbinsMVAoutput, 0.0, 1.0 );
      rarity_b = new TH1D( TestvarName + "_Rarity_B", TestvarName + "_Rarity_B", fNbinsMVAoutput, 0.0, 1.0 );
      mvaRes->Store(rarity_s, "Rar_S");
      mvaRes->Store(rarity_b, "Rar_B");
      rarity_s->Sumw2();
      rarity_b->Sumw2();
   }

   // MVA plots used for efficiency calculations (large number of bins)
   TH1* mva_eff_s = new TH1D( TestvarName + "_S_high", TestvarName + "_S_high", fNbinsH, fXmin, sxmax );
   TH1* mva_eff_b = new TH1D( TestvarName + "_B_high", TestvarName + "_B_high", fNbinsH, fXmin, sxmax );
   mvaRes->Store(mva_eff_s, "MVA_HIGHBIN_S");
   mvaRes->Store(mva_eff_b, "MVA_HIGHBIN_B");
   mva_eff_s->Sumw2();
   mva_eff_b->Sumw2();

   // fill the histograms

   ResultsClassification* mvaProb = dynamic_cast<ResultsClassification*>
      (Data()->GetResults( TString("prob_")+GetMethodName(), Types::kTesting, Types::kMaxAnalysisType ) );

   Log() << kHEADER <<Form("[%s] : ",DataInfo().GetName())<< "Loop over test events and fill histograms with classifier response..." << Endl << Endl;
   if (mvaProb) Log() << kINFO << "Also filling probability and rarity histograms (on request)..." << Endl;
   std::vector<Bool_t>* mvaResTypes = mvaRes->GetValueVectorTypes();

   //LM: this is needed to avoid crashes in ROOCCURVE
   if ( mvaRes->GetSize() != GetNEvents() ) {
      Log() << kFATAL << TString::Format("Inconsistent result size  %lld with number of events %u ",    mvaRes->GetSize() ,  GetNEvents() ) << Endl;
      assert(mvaRes->GetSize() == GetNEvents());
   }

   for (Long64_t ievt=0; ievt<GetNEvents(); ievt++) {

      const Event* ev = GetEvent(ievt);
      Float_t v = (*mvaRes)[ievt][0];
      Float_t w = ev->GetWeight();

      if (DataInfo().IsSignal(ev)) {
         mvaResTypes->push_back(kTRUE);
         mva_s ->Fill( v, w );
         if (mvaProb) {
            proba_s->Fill( (*mvaProb)[ievt][0], w );
            rarity_s->Fill( GetRarity( v ), w );
         }

         mva_eff_s ->Fill( v, w );
      }
      else {
         mvaResTypes->push_back(kFALSE);
         mva_b ->Fill( v, w );
         if (mvaProb) {
            proba_b->Fill( (*mvaProb)[ievt][0], w );
            rarity_b->Fill( GetRarity( v ), w );
         }
         mva_eff_b ->Fill( v, w );
      }
   }

   // uncomment those (and several others if you want unnormalized output
   gTools().NormHist( mva_s  );
   gTools().NormHist( mva_b  );
   gTools().NormHist( proba_s );
   gTools().NormHist( proba_b );
   gTools().NormHist( rarity_s );
   gTools().NormHist( rarity_b );
   gTools().NormHist( mva_eff_s  );
   gTools().NormHist( mva_eff_b  );

   // create PDFs from histograms, using default splines, and no additional smoothing
   if (fSplS) { delete fSplS; fSplS = 0; }
   if (fSplB) { delete fSplB; fSplB = 0; }
   fSplS = new PDF( TString(GetName()) + " PDF Sig", mva_s, PDF::kSpline2 );
   fSplB = new PDF( TString(GetName()) + " PDF Bkg", mva_b, PDF::kSpline2 );
}

////////////////////////////////////////////////////////////////////////////////
/// general method used in writing the header of the weight files where
/// the used variables, variable transformation type etc. is specified

void TMVA::MethodBase::WriteStateToStream( std::ostream& tf ) const
{
   TString prefix = "";
   UserGroup_t * userInfo = gSystem->GetUserInfo();

   tf << prefix << "#GEN -*-*-*-*-*-*-*-*-*-*-*- general info -*-*-*-*-*-*-*-*-*-*-*-" << std::endl << prefix << std::endl;
   tf << prefix << "Method         : " << GetMethodTypeName() << "::" << GetMethodName() << std::endl;
   tf.setf(std::ios::left);
   tf << prefix << "TMVA Release   : " << std::setw(10) << GetTrainingTMVAVersionString() << "    ["
      << GetTrainingTMVAVersionCode() << "]" << std::endl;
   tf << prefix << "ROOT Release   : " << std::setw(10) << GetTrainingROOTVersionString() << "    ["
      << GetTrainingROOTVersionCode() << "]" << std::endl;
   tf << prefix << "Creator        : " << userInfo->fUser << std::endl;
   tf << prefix << "Date           : "; TDatime *d = new TDatime; tf << d->AsString() << std::endl; delete d;
   tf << prefix << "Host           : " << gSystem->GetBuildNode() << std::endl;
   tf << prefix << "Dir            : " << gSystem->WorkingDirectory() << std::endl;
   tf << prefix << "Training events: " << Data()->GetNTrainingEvents() << std::endl;

   TString analysisType(((const_cast<TMVA::MethodBase*>(this)->GetAnalysisType()==Types::kRegression) ? "Regression" : "Classification"));

   tf << prefix << "Analysis type  : " << "[" << ((GetAnalysisType()==Types::kRegression) ? "Regression" : "Classification") << "]" << std::endl;
   tf << prefix << std::endl;

   delete userInfo;

   // First write all options
   tf << prefix << std::endl << prefix << "#OPT -*-*-*-*-*-*-*-*-*-*-*-*- options -*-*-*-*-*-*-*-*-*-*-*-*-" << std::endl << prefix << std::endl;
   WriteOptionsToStream( tf, prefix );
   tf << prefix << std::endl;

   // Second write variable info
   tf << prefix << std::endl << prefix << "#VAR -*-*-*-*-*-*-*-*-*-*-*-* variables *-*-*-*-*-*-*-*-*-*-*-*-" << std::endl << prefix << std::endl;
   WriteVarsToStream( tf, prefix );
   tf << prefix << std::endl;
}

////////////////////////////////////////////////////////////////////////////////
/// xml writing

void TMVA::MethodBase::AddInfoItem( void* gi, const TString& name, const TString& value) const
{
   void* it = gTools().AddChild(gi,"Info");
   gTools().AddAttr(it,"name", name);
   gTools().AddAttr(it,"value", value);
}

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

void TMVA::MethodBase::AddOutput( Types::ETreeType type, Types::EAnalysisType analysisType ) {
   if (analysisType == Types::kRegression) {
      AddRegressionOutput( type );
   } else if (analysisType == Types::kMulticlass) {
      AddMulticlassOutput( type );
   } else {
      AddClassifierOutput( type );
      if (HasMVAPdfs())
         AddClassifierOutputProb( type );
   }
}

////////////////////////////////////////////////////////////////////////////////
/// general method used in writing the header of the weight files where
/// the used variables, variable transformation type etc. is specified

void TMVA::MethodBase::WriteStateToXML( void* parent ) const
{
   if (!parent) return;

   UserGroup_t* userInfo = gSystem->GetUserInfo();

   void* gi = gTools().AddChild(parent, "GeneralInfo");
   AddInfoItem( gi, "TMVA Release", GetTrainingTMVAVersionString() + " [" + gTools().StringFromInt(GetTrainingTMVAVersionCode()) + "]" );
   AddInfoItem( gi, "ROOT Release", GetTrainingROOTVersionString() + " [" + gTools().StringFromInt(GetTrainingROOTVersionCode()) + "]");
   AddInfoItem( gi, "Creator", userInfo->fUser);
   TDatime dt; AddInfoItem( gi, "Date", dt.AsString());
   AddInfoItem( gi, "Host", gSystem->GetBuildNode() );
   AddInfoItem( gi, "Dir", gSystem->WorkingDirectory());
   AddInfoItem( gi, "Training events", gTools().StringFromInt(Data()->GetNTrainingEvents()));
   AddInfoItem( gi, "TrainingTime", gTools().StringFromDouble(const_cast<TMVA::MethodBase*>(this)->GetTrainTime()));

   Types::EAnalysisType aType = const_cast<TMVA::MethodBase*>(this)->GetAnalysisType();
   TString analysisType((aType==Types::kRegression) ? "Regression" :
                        (aType==Types::kMulticlass ? "Multiclass" : "Classification"));
   AddInfoItem( gi, "AnalysisType", analysisType );
   delete userInfo;

   // write options
   AddOptionsXMLTo( parent );

   // write variable info
   AddVarsXMLTo( parent );

   // write spectator info
   if (fModelPersistence)
      AddSpectatorsXMLTo( parent );

   // write class info if in multiclass mode
   AddClassesXMLTo(parent);

   // write target info if in regression mode
   if (DoRegression()) AddTargetsXMLTo(parent);

   // write transformations
   GetTransformationHandler(false).AddXMLTo( parent );

   // write MVA variable distributions
   void* pdfs = gTools().AddChild(parent, "MVAPdfs");
   if (fMVAPdfS) fMVAPdfS->AddXMLTo(pdfs);
   if (fMVAPdfB) fMVAPdfB->AddXMLTo(pdfs);

   // write weights
   AddWeightsXMLTo( parent );
}

////////////////////////////////////////////////////////////////////////////////
/// write reference MVA distributions (and other information)
/// to a ROOT type weight file

void TMVA::MethodBase::ReadStateFromStream( TFile& rf )
{
   Bool_t addDirStatus = TH1::AddDirectoryStatus();
   TH1::AddDirectory( 0 ); // this avoids the binding of the hists in PDF to the current ROOT file
   fMVAPdfS = (TMVA::PDF*)rf.Get( "MVA_PDF_Signal" );
   fMVAPdfB = (TMVA::PDF*)rf.Get( "MVA_PDF_Background" );

   TH1::AddDirectory( addDirStatus );

   ReadWeightsFromStream( rf );

   SetTestvarName();
}

////////////////////////////////////////////////////////////////////////////////
/// write options and weights to file
/// note that each one text file for the main configuration information
/// and one ROOT file for ROOT objects are created

void TMVA::MethodBase::WriteStateToFile() const
{
   // ---- create the text file
   TString tfname( GetWeightFileName() );

   // writing xml file
   TString xmlfname( tfname ); xmlfname.ReplaceAll( ".txt", ".xml" );
   Log() << kINFO //<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Creating xml weight file: "
         << gTools().Color("lightblue") << xmlfname << gTools().Color("reset") << Endl;
   void* doc      = gTools().xmlengine().NewDoc();
   void* rootnode = gTools().AddChild(0,"MethodSetup", "", true);
   gTools().xmlengine().DocSetRootElement(doc,rootnode);
   gTools().AddAttr(rootnode,"Method", GetMethodTypeName() + "::" + GetMethodName());
   WriteStateToXML(rootnode);
   gTools().xmlengine().SaveDoc(doc,xmlfname);
   gTools().xmlengine().FreeDoc(doc);
}

////////////////////////////////////////////////////////////////////////////////
/// Function to write options and weights to file

void TMVA::MethodBase::ReadStateFromFile()
{
   // get the filename

   TString tfname(GetWeightFileName());

   Log() << kINFO //<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Reading weight file: "
         << gTools().Color("lightblue") << tfname << gTools().Color("reset") << Endl;

   if (tfname.EndsWith(".xml") ) {
      void* doc = gTools().xmlengine().ParseFile(tfname,gTools().xmlenginebuffersize()); // the default buffer size in TXMLEngine::ParseFile is 100k. Starting with ROOT 5.29 one can set the buffer size, see: http://savannah.cern.ch/bugs/?78864. This might be necessary for large XML files
      if (!doc) {
         Log() << kFATAL << "Error parsing XML file " << tfname << Endl;
      }
      void* rootnode = gTools().xmlengine().DocGetRootElement(doc); // node "MethodSetup"
      ReadStateFromXML(rootnode);
      gTools().xmlengine().FreeDoc(doc);
   }
   else {
      std::filebuf fb;
      fb.open(tfname.Data(),std::ios::in);
      if (!fb.is_open()) { // file not found --> Error
         Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<ReadStateFromFile> "
               << "Unable to open input weight file: " << tfname << Endl;
      }
      std::istream fin(&fb);
      ReadStateFromStream(fin);
      fb.close();
   }
   if (!fTxtWeightsOnly) {
      // ---- read the ROOT file
      TString rfname( tfname ); rfname.ReplaceAll( ".txt", ".root" );
      Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Reading root weight file: "
            << gTools().Color("lightblue") << rfname << gTools().Color("reset") << Endl;
      TFile* rfile = TFile::Open( rfname, "READ" );
      ReadStateFromStream( *rfile );
      rfile->Close();
   }
}
////////////////////////////////////////////////////////////////////////////////
/// for reading from memory

void TMVA::MethodBase::ReadStateFromXMLString( const char* xmlstr ) {
   void* doc = gTools().xmlengine().ParseString(xmlstr);
   void* rootnode = gTools().xmlengine().DocGetRootElement(doc); // node "MethodSetup"
   ReadStateFromXML(rootnode);
   gTools().xmlengine().FreeDoc(doc);

   return;
}

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

void TMVA::MethodBase::ReadStateFromXML( void* methodNode )
{

   TString fullMethodName;
   gTools().ReadAttr( methodNode, "Method", fullMethodName );

   fMethodName = fullMethodName(fullMethodName.Index("::")+2,fullMethodName.Length());

   // update logger
   Log().SetSource( GetName() );
   Log() << kDEBUG//<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Read method \"" << GetMethodName() << "\" of type \"" << GetMethodTypeName() << "\"" << Endl;

   // after the method name is read, the testvar can be set
   SetTestvarName();

   TString nodeName("");
   void* ch = gTools().GetChild(methodNode);
   while (ch!=0) {
      nodeName = TString( gTools().GetName(ch) );

      if (nodeName=="GeneralInfo") {
         // read analysis type

         TString name(""),val("");
         void* antypeNode = gTools().GetChild(ch);
         while (antypeNode) {
            gTools().ReadAttr( antypeNode, "name",   name );

            if (name == "TrainingTime")
               gTools().ReadAttr( antypeNode, "value",  fTrainTime );

            if (name == "AnalysisType") {
               gTools().ReadAttr( antypeNode, "value",  val );
               val.ToLower();
               if      (val == "regression" )     SetAnalysisType( Types::kRegression );
               else if (val == "classification" ) SetAnalysisType( Types::kClassification );
               else if (val == "multiclass" )     SetAnalysisType( Types::kMulticlass );
               else Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Analysis type " << val << " is not known." << Endl;
            }

            if (name == "TMVA Release" || name == "TMVA") {
               TString s;
               gTools().ReadAttr( antypeNode, "value", s);
               fTMVATrainingVersion = TString(s(s.Index("[")+1,s.Index("]")-s.Index("[")-1)).Atoi();
               Log() << kDEBUG <<Form("[%s] : ",DataInfo().GetName()) << "MVA method was trained with TMVA Version: " << GetTrainingTMVAVersionString() << Endl;
            }

            if (name == "ROOT Release" || name == "ROOT") {
               TString s;
               gTools().ReadAttr( antypeNode, "value", s);
               fROOTTrainingVersion = TString(s(s.Index("[")+1,s.Index("]")-s.Index("[")-1)).Atoi();
               Log() << kDEBUG //<<Form("Dataset[%s] : ",DataInfo().GetName())
           << "MVA method was trained with ROOT Version: " << GetTrainingROOTVersionString() << Endl;
            }
            antypeNode = gTools().GetNextChild(antypeNode);
         }
      }
      else if (nodeName=="Options") {
         ReadOptionsFromXML(ch);
         ParseOptions();

      }
      else if (nodeName=="Variables") {
         ReadVariablesFromXML(ch);
      }
      else if (nodeName=="Spectators") {
         ReadSpectatorsFromXML(ch);
      }
      else if (nodeName=="Classes") {
         if (DataInfo().GetNClasses()==0) ReadClassesFromXML(ch);
      }
      else if (nodeName=="Targets") {
         if (DataInfo().GetNTargets()==0 && DoRegression()) ReadTargetsFromXML(ch);
      }
      else if (nodeName=="Transformations") {
         GetTransformationHandler().ReadFromXML(ch);
      }
      else if (nodeName=="MVAPdfs") {
         TString pdfname;
         if (fMVAPdfS) { delete fMVAPdfS; fMVAPdfS=0; }
         if (fMVAPdfB) { delete fMVAPdfB; fMVAPdfB=0; }
         void* pdfnode = gTools().GetChild(ch);
         if (pdfnode) {
            gTools().ReadAttr(pdfnode, "Name", pdfname);
            fMVAPdfS = new PDF(pdfname);
            fMVAPdfS->ReadXML(pdfnode);
            pdfnode = gTools().GetNextChild(pdfnode);
            gTools().ReadAttr(pdfnode, "Name", pdfname);
            fMVAPdfB = new PDF(pdfname);
            fMVAPdfB->ReadXML(pdfnode);
         }
      }
      else if (nodeName=="Weights") {
         ReadWeightsFromXML(ch);
      }
      else {
         Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Unparsed XML node: '" << nodeName << "'" << Endl;
      }
      ch = gTools().GetNextChild(ch);

   }

   // update transformation handler
   if (GetTransformationHandler().GetCallerName() == "") GetTransformationHandler().SetCallerName( GetName() );
}

////////////////////////////////////////////////////////////////////////////////
/// read the header from the weight files of the different MVA methods

void TMVA::MethodBase::ReadStateFromStream( std::istream& fin )
{
   char buf[512];

   // when reading from stream, we assume the files are produced with TMVA<=397
   SetAnalysisType(Types::kClassification);


   // first read the method name
   GetLine(fin,buf);
   while (!TString(buf).BeginsWith("Method")) GetLine(fin,buf);
   TString namestr(buf);

   TString methodType = namestr(0,namestr.Index("::"));
   methodType = methodType(methodType.Last(' '),methodType.Length());
   methodType = methodType.Strip(TString::kLeading);

   TString methodName = namestr(namestr.Index("::")+2,namestr.Length());
   methodName = methodName.Strip(TString::kLeading);
   if (methodName == "") methodName = methodType;
   fMethodName  = methodName;

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Read method \"" << GetMethodName() << "\" of type \"" << GetMethodTypeName() << "\"" << Endl;

   // update logger
   Log().SetSource( GetName() );

   // now the question is whether to read the variables first or the options (well, of course the order
   // of writing them needs to agree)
   //
   // the option "Decorrelation" is needed to decide if the variables we
   // read are decorrelated or not
   //
   // the variables are needed by some methods (TMLP) to build the NN
   // which is done in ProcessOptions so for the time being we first Read and Parse the options then
   // we read the variables, and then we process the options

   // now read all options
   GetLine(fin,buf);
   while (!TString(buf).BeginsWith("#OPT")) GetLine(fin,buf);
   ReadOptionsFromStream(fin);
   ParseOptions();

   // Now read variable info
   fin.getline(buf,512);
   while (!TString(buf).BeginsWith("#VAR")) fin.getline(buf,512);
   ReadVarsFromStream(fin);

   // now we process the options (of the derived class)
   ProcessOptions();

   if (IsNormalised()) {
      VariableNormalizeTransform* norm = (VariableNormalizeTransform*)
         GetTransformationHandler().AddTransformation( new VariableNormalizeTransform(DataInfo()), -1 );
      norm->BuildTransformationFromVarInfo( DataInfo().GetVariableInfos() );
   }
   VariableTransformBase *varTrafo(0), *varTrafo2(0);
   if ( fVarTransformString == "None") {
      if (fUseDecorr)
         varTrafo = GetTransformationHandler().AddTransformation( new VariableDecorrTransform(DataInfo()), -1 );
   } else if ( fVarTransformString == "Decorrelate" ) {
      varTrafo = GetTransformationHandler().AddTransformation( new VariableDecorrTransform(DataInfo()), -1 );
   } else if ( fVarTransformString == "PCA"  ) {
      varTrafo = GetTransformationHandler().AddTransformation( new VariablePCATransform(DataInfo()), -1 );
   } else if ( fVarTransformString == "Uniform" ) {
      varTrafo  = GetTransformationHandler().AddTransformation( new VariableGaussTransform(DataInfo(),"Uniform"), -1 );
   } else if ( fVarTransformString == "Gauss" ) {
      varTrafo  = GetTransformationHandler().AddTransformation( new VariableGaussTransform(DataInfo()), -1 );
   } else if ( fVarTransformString == "GaussDecorr" ) {
      varTrafo  = GetTransformationHandler().AddTransformation( new VariableGaussTransform(DataInfo()), -1 );
      varTrafo2 = GetTransformationHandler().AddTransformation( new VariableDecorrTransform(DataInfo()), -1 );
   } else {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<ProcessOptions> Variable transform '"
            << fVarTransformString << "' unknown." << Endl;
   }
   // Now read decorrelation matrix if available
   if (GetTransformationHandler().GetTransformationList().GetSize() > 0) {
      fin.getline(buf,512);
      while (!TString(buf).BeginsWith("#MAT")) fin.getline(buf,512);
      if (varTrafo) {
         TString trafo(fVariableTransformTypeString); trafo.ToLower();
         varTrafo->ReadTransformationFromStream(fin, trafo );
      }
      if (varTrafo2) {
         TString trafo(fVariableTransformTypeString); trafo.ToLower();
         varTrafo2->ReadTransformationFromStream(fin, trafo );
      }
   }


   if (HasMVAPdfs()) {
      // Now read the MVA PDFs
      fin.getline(buf,512);
      while (!TString(buf).BeginsWith("#MVAPDFS")) fin.getline(buf,512);
      if (fMVAPdfS != 0) { delete fMVAPdfS; fMVAPdfS = 0; }
      if (fMVAPdfB != 0) { delete fMVAPdfB; fMVAPdfB = 0; }
      fMVAPdfS = new PDF(TString(GetName()) + " MVA PDF Sig");
      fMVAPdfB = new PDF(TString(GetName()) + " MVA PDF Bkg");
      fMVAPdfS->SetReadingVersion( GetTrainingTMVAVersionCode() );
      fMVAPdfB->SetReadingVersion( GetTrainingTMVAVersionCode() );

      fin >> *fMVAPdfS;
      fin >> *fMVAPdfB;
   }

   // Now read weights
   fin.getline(buf,512);
   while (!TString(buf).BeginsWith("#WGT")) fin.getline(buf,512);
   fin.getline(buf,512);
   ReadWeightsFromStream( fin );;

   // update transformation handler
   if (GetTransformationHandler().GetCallerName() == "") GetTransformationHandler().SetCallerName( GetName() );

}

////////////////////////////////////////////////////////////////////////////////
/// write the list of variables (name, min, max) for a given data
/// transformation method to the stream

void TMVA::MethodBase::WriteVarsToStream( std::ostream& o, const TString& prefix ) const
{
   o << prefix << "NVar " << DataInfo().GetNVariables() << std::endl;
   std::vector<VariableInfo>::const_iterator varIt = DataInfo().GetVariableInfos().begin();
   for (; varIt!=DataInfo().GetVariableInfos().end(); ++varIt) { o << prefix; varIt->WriteToStream(o); }
   o << prefix << "NSpec " << DataInfo().GetNSpectators() << std::endl;
   varIt = DataInfo().GetSpectatorInfos().begin();
   for (; varIt!=DataInfo().GetSpectatorInfos().end(); ++varIt) { o << prefix; varIt->WriteToStream(o); }
}

////////////////////////////////////////////////////////////////////////////////
/// Read the variables (name, min, max) for a given data
/// transformation method from the stream. In the stream we only
/// expect the limits which will be set

void TMVA::MethodBase::ReadVarsFromStream( std::istream& istr )
{
   TString dummy;
   UInt_t readNVar;
   istr >> dummy >> readNVar;

   if (readNVar!=DataInfo().GetNVariables()) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "You declared "<< DataInfo().GetNVariables() << " variables in the Reader"
            << " while there are " << readNVar << " variables declared in the file"
            << Endl;
   }

   // we want to make sure all variables are read in the order they are defined
   VariableInfo varInfo;
   std::vector<VariableInfo>::iterator varIt = DataInfo().GetVariableInfos().begin();
   int varIdx = 0;
   for (; varIt!=DataInfo().GetVariableInfos().end(); ++varIt, ++varIdx) {
      varInfo.ReadFromStream(istr);
      if (varIt->GetExpression() == varInfo.GetExpression()) {
         varInfo.SetExternalLink((*varIt).GetExternalLink());
         (*varIt) = varInfo;
      }
      else {
         Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "ERROR in <ReadVarsFromStream>" << Endl;
         Log() << kINFO << "The definition (or the order) of the variables found in the input file is"  << Endl;
         Log() << kINFO << "is not the same as the one declared in the Reader (which is necessary for" << Endl;
         Log() << kINFO << "the correct working of the method):" << Endl;
         Log() << kINFO << "   var #" << varIdx <<" declared in Reader: " << varIt->GetExpression() << Endl;
         Log() << kINFO << "   var #" << varIdx <<" declared in file  : " << varInfo.GetExpression() << Endl;
         Log() << kFATAL << "The expression declared to the Reader needs to be checked (name or order are wrong)" << Endl;
      }
   }
}

////////////////////////////////////////////////////////////////////////////////
/// write variable info to XML

void TMVA::MethodBase::AddVarsXMLTo( void* parent ) const
{
   void* vars = gTools().AddChild(parent, "Variables");
   gTools().AddAttr( vars, "NVar", gTools().StringFromInt(DataInfo().GetNVariables()) );

   for (UInt_t idx=0; idx<DataInfo().GetVariableInfos().size(); idx++) {
      VariableInfo& vi = DataInfo().GetVariableInfos()[idx];
      void* var = gTools().AddChild( vars, "Variable" );
      gTools().AddAttr( var, "VarIndex", idx );
      vi.AddToXML( var );
   }
}

////////////////////////////////////////////////////////////////////////////////
/// write spectator info to XML

void TMVA::MethodBase::AddSpectatorsXMLTo( void* parent ) const
{
   void* specs = gTools().AddChild(parent, "Spectators");

   UInt_t writeIdx=0;
   for (UInt_t idx=0; idx<DataInfo().GetSpectatorInfos().size(); idx++) {

      VariableInfo& vi = DataInfo().GetSpectatorInfos()[idx];

      // we do not want to write spectators that are category-cuts,
      // except if the method is the category method and the spectators belong to it
      if (vi.GetVarType()=='C') continue;

      void* spec = gTools().AddChild( specs, "Spectator" );
      gTools().AddAttr( spec, "SpecIndex", writeIdx++ );
      vi.AddToXML( spec );
   }
   gTools().AddAttr( specs, "NSpec", gTools().StringFromInt(writeIdx) );
}

////////////////////////////////////////////////////////////////////////////////
/// write class info to XML

void TMVA::MethodBase::AddClassesXMLTo( void* parent ) const
{
   UInt_t nClasses=DataInfo().GetNClasses();

   void* classes = gTools().AddChild(parent, "Classes");
   gTools().AddAttr( classes, "NClass", nClasses );

   for (UInt_t iCls=0; iCls<nClasses; ++iCls) {
      ClassInfo *classInfo=DataInfo().GetClassInfo (iCls);
      TString  className  =classInfo->GetName();
      UInt_t   classNumber=classInfo->GetNumber();

      void* classNode=gTools().AddChild(classes, "Class");
      gTools().AddAttr( classNode, "Name",  className   );
      gTools().AddAttr( classNode, "Index", classNumber );
   }
}
////////////////////////////////////////////////////////////////////////////////
/// write target info to XML

void TMVA::MethodBase::AddTargetsXMLTo( void* parent ) const
{
   void* targets = gTools().AddChild(parent, "Targets");
   gTools().AddAttr( targets, "NTrgt", gTools().StringFromInt(DataInfo().GetNTargets()) );

   for (UInt_t idx=0; idx<DataInfo().GetTargetInfos().size(); idx++) {
      VariableInfo& vi = DataInfo().GetTargetInfos()[idx];
      void* tar = gTools().AddChild( targets, "Target" );
      gTools().AddAttr( tar, "TargetIndex", idx );
      vi.AddToXML( tar );
   }
}

////////////////////////////////////////////////////////////////////////////////
/// read variable info from XML

void TMVA::MethodBase::ReadVariablesFromXML( void* varnode )
{
   UInt_t readNVar;
   gTools().ReadAttr( varnode, "NVar", readNVar);

   if (readNVar!=DataInfo().GetNVariables()) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "You declared "<< DataInfo().GetNVariables() << " variables in the Reader"
            << " while there are " << readNVar << " variables declared in the file"
            << Endl;
   }

   // we want to make sure all variables are read in the order they are defined
   VariableInfo readVarInfo, existingVarInfo;
   int varIdx = 0;
   void* ch = gTools().GetChild(varnode);
   while (ch) {
      gTools().ReadAttr( ch, "VarIndex", varIdx);
      existingVarInfo = DataInfo().GetVariableInfos()[varIdx];
      readVarInfo.ReadFromXML(ch);

      if (existingVarInfo.GetExpression() == readVarInfo.GetExpression()) {
         readVarInfo.SetExternalLink(existingVarInfo.GetExternalLink());
         existingVarInfo = readVarInfo;
      }
      else {
         Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "ERROR in <ReadVariablesFromXML>" << Endl;
         Log() << kINFO << "The definition (or the order) of the variables found in the input file is"  << Endl;
         Log() << kINFO << "not the same as the one declared in the Reader (which is necessary for the" << Endl;
         Log() << kINFO << "correct working of the method):" << Endl;
         Log() << kINFO << "   var #" << varIdx <<" declared in Reader: " << existingVarInfo.GetExpression() << Endl;
         Log() << kINFO << "   var #" << varIdx <<" declared in file  : " << readVarInfo.GetExpression() << Endl;
         Log() << kFATAL << "The expression declared to the Reader needs to be checked (name or order are wrong)" << Endl;
      }
      ch = gTools().GetNextChild(ch);
   }
}

////////////////////////////////////////////////////////////////////////////////
/// read spectator info from XML

void TMVA::MethodBase::ReadSpectatorsFromXML( void* specnode )
{
   UInt_t readNSpec;
   gTools().ReadAttr( specnode, "NSpec", readNSpec);

   if (readNSpec!=DataInfo().GetNSpectators(kFALSE)) {
      Log() << kFATAL<<Form("Dataset[%s] : ",DataInfo().GetName()) << "You declared "<< DataInfo().GetNSpectators(kFALSE) << " spectators in the Reader"
            << " while there are " << readNSpec << " spectators declared in the file"
            << Endl;
   }

   // we want to make sure all variables are read in the order they are defined
   VariableInfo readSpecInfo, existingSpecInfo;
   int specIdx = 0;
   void* ch = gTools().GetChild(specnode);
   while (ch) {
      gTools().ReadAttr( ch, "SpecIndex", specIdx);
      existingSpecInfo = DataInfo().GetSpectatorInfos()[specIdx];
      readSpecInfo.ReadFromXML(ch);

      if (existingSpecInfo.GetExpression() == readSpecInfo.GetExpression()) {
         readSpecInfo.SetExternalLink(existingSpecInfo.GetExternalLink());
         existingSpecInfo = readSpecInfo;
      }
      else {
         Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "ERROR in <ReadSpectatorsFromXML>" << Endl;
         Log() << kINFO << "The definition (or the order) of the spectators found in the input file is"  << Endl;
         Log() << kINFO << "not the same as the one declared in the Reader (which is necessary for the" << Endl;
         Log() << kINFO << "correct working of the method):" << Endl;
         Log() << kINFO << "  spec #" << specIdx <<" declared in Reader: " << existingSpecInfo.GetExpression() << Endl;
         Log() << kINFO << "  spec #" << specIdx <<" declared in file  : " << readSpecInfo.GetExpression() << Endl;
         Log() << kFATAL << "The expression declared to the Reader needs to be checked (name or order are wrong)" << Endl;
      }
      ch = gTools().GetNextChild(ch);
   }
}

////////////////////////////////////////////////////////////////////////////////
/// read number of classes from XML

void TMVA::MethodBase::ReadClassesFromXML( void* clsnode )
{
   UInt_t readNCls;
   // coverity[tainted_data_argument]
   gTools().ReadAttr( clsnode, "NClass", readNCls);

   TString className="";
   UInt_t  classIndex=0;
   void* ch = gTools().GetChild(clsnode);
   if (!ch) {
      for (UInt_t icls = 0; icls<readNCls;++icls) {
         TString classname = Form("class%i",icls);
         DataInfo().AddClass(classname);

      }
   }
   else{
      while (ch) {
         gTools().ReadAttr( ch, "Index", classIndex);
         gTools().ReadAttr( ch, "Name",  className );
         DataInfo().AddClass(className);

         ch = gTools().GetNextChild(ch);
      }
   }

   // retrieve signal and background class index
   if (DataInfo().GetClassInfo("Signal") != 0) {
      fSignalClass = DataInfo().GetClassInfo("Signal")->GetNumber();
   }
   else
      fSignalClass=0;
   if (DataInfo().GetClassInfo("Background") != 0) {
      fBackgroundClass = DataInfo().GetClassInfo("Background")->GetNumber();
   }
   else
      fBackgroundClass=1;
}

////////////////////////////////////////////////////////////////////////////////
/// read target info from XML

void TMVA::MethodBase::ReadTargetsFromXML( void* tarnode )
{
   UInt_t readNTar;
   gTools().ReadAttr( tarnode, "NTrgt", readNTar);

   int tarIdx = 0;
   TString expression;
   void* ch = gTools().GetChild(tarnode);
   while (ch) {
      gTools().ReadAttr( ch, "TargetIndex", tarIdx);
      gTools().ReadAttr( ch, "Expression", expression);
      DataInfo().AddTarget(expression,"","",0,0);

      ch = gTools().GetNextChild(ch);
   }
}

////////////////////////////////////////////////////////////////////////////////
/// returns the ROOT directory where info/histograms etc of the
/// corresponding MVA method instance are stored

TDirectory* TMVA::MethodBase::BaseDir() const
{
   if (fBaseDir != 0) return fBaseDir;
   Log()<<kDEBUG<<Form("Dataset[%s] : ",DataInfo().GetName())<<" Base Directory for " << GetMethodName() << " not set yet --> check if already there.." <<Endl;

   if (IsSilentFile()) {
      Log() << kFATAL << Form("Dataset[%s] : ", DataInfo().GetName())
            << "MethodBase::BaseDir() - No directory exists when running a Method without output file. Enable the "
               "output when creating the factory"
            << Endl;
   }

   TDirectory* methodDir = MethodBaseDir();
   if (methodDir==0)
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "MethodBase::BaseDir() - MethodBaseDir() return a NULL pointer!" << Endl;

   TString defaultDir = GetMethodName();
   TDirectory *sdir = methodDir->GetDirectory(defaultDir.Data());
   if(!sdir)
      {
         Log()<<kDEBUG<<Form("Dataset[%s] : ",DataInfo().GetName())<<" Base Directory for " << GetMethodTypeName() << " does not exist yet--> created it" <<Endl;
         sdir = methodDir->mkdir(defaultDir);
         sdir->cd();
         // write weight file name into target file
         if (fModelPersistence) {
            TObjString wfilePath( gSystem->WorkingDirectory() );
            TObjString wfileName( GetWeightFileName() );
            wfilePath.Write( "TrainingPath" );
            wfileName.Write( "WeightFileName" );
         }
      }

   Log()<<kDEBUG<<Form("Dataset[%s] : ",DataInfo().GetName())<<" Base Directory for " << GetMethodTypeName() << " existed, return it.." <<Endl;
   return sdir;
}

////////////////////////////////////////////////////////////////////////////////
/// returns the ROOT directory where all instances of the
/// corresponding MVA method are stored

TDirectory *TMVA::MethodBase::MethodBaseDir() const
{
   if (fMethodBaseDir != 0) {
      return fMethodBaseDir;
   }

   const char *datasetName = DataInfo().GetName();

   Log() << kDEBUG << Form("Dataset[%s] : ", datasetName) << " Base Directory for " << GetMethodTypeName()
         << " not set yet --> check if already there.." << Endl;

   TDirectory *factoryBaseDir = GetFile();
   if (!factoryBaseDir) return nullptr;
   fMethodBaseDir = factoryBaseDir->GetDirectory(datasetName);
   if (!fMethodBaseDir) {
      fMethodBaseDir = factoryBaseDir->mkdir(datasetName, Form("Base directory for dataset %s", datasetName));
      if (!fMethodBaseDir) {
         Log() << kFATAL << "Can not create dir " << datasetName;
      }
   }
   TString methodTypeDir = Form("Method_%s", GetMethodTypeName().Data());
   fMethodBaseDir = fMethodBaseDir->GetDirectory(methodTypeDir.Data());

   if (!fMethodBaseDir) {
      TDirectory *datasetDir = factoryBaseDir->GetDirectory(datasetName);
      TString methodTypeDirHelpStr = Form("Directory for all %s methods", GetMethodTypeName().Data());
      fMethodBaseDir = datasetDir->mkdir(methodTypeDir.Data(), methodTypeDirHelpStr);
      Log() << kDEBUG << Form("Dataset[%s] : ", datasetName) << " Base Directory for " << GetMethodName()
            << " does not exist yet--> created it" << Endl;
   }

   Log() << kDEBUG << Form("Dataset[%s] : ", datasetName)
         << "Return from MethodBaseDir() after creating base directory " << Endl;
   return fMethodBaseDir;
}

////////////////////////////////////////////////////////////////////////////////
/// set directory of weight file

void TMVA::MethodBase::SetWeightFileDir( TString fileDir )
{
   fFileDir = fileDir;
   gSystem->mkdir( fFileDir, kTRUE );
}

////////////////////////////////////////////////////////////////////////////////
/// set the weight file name (depreciated)

void TMVA::MethodBase::SetWeightFileName( TString theWeightFile)
{
   fWeightFile = theWeightFile;
}

////////////////////////////////////////////////////////////////////////////////
/// retrieve weight file name

TString TMVA::MethodBase::GetWeightFileName() const
{
   if (fWeightFile!="") return fWeightFile;

   // the default consists of
   // directory/jobname_methodname_suffix.extension.{root/txt}
   TString suffix = "";
   TString wFileDir(GetWeightFileDir());
   TString wFileName = GetJobName() + "_" + GetMethodName() +
      suffix + "." + gConfig().GetIONames().fWeightFileExtension + ".xml";
   if (wFileDir.IsNull() )  return wFileName;
   // add weight file directory of it is not null
   return ( wFileDir + (wFileDir[wFileDir.Length()-1]=='/' ? "" : "/")
            + wFileName );
}
////////////////////////////////////////////////////////////////////////////////
/// writes all MVA evaluation histograms to file

void TMVA::MethodBase::WriteEvaluationHistosToFile(Types::ETreeType treetype)
{
   BaseDir()->cd();


   // write MVA PDFs to file - if exist
   if (0 != fMVAPdfS) {
      fMVAPdfS->GetOriginalHist()->Write();
      fMVAPdfS->GetSmoothedHist()->Write();
      fMVAPdfS->GetPDFHist()->Write();
   }
   if (0 != fMVAPdfB) {
      fMVAPdfB->GetOriginalHist()->Write();
      fMVAPdfB->GetSmoothedHist()->Write();
      fMVAPdfB->GetPDFHist()->Write();
   }

   // write result-histograms
   Results* results = Data()->GetResults( GetMethodName(), treetype, Types::kMaxAnalysisType );
   if (!results)
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<WriteEvaluationHistosToFile> Unknown result: "
            << GetMethodName() << (treetype==Types::kTraining?"/kTraining":"/kTesting")
            << "/kMaxAnalysisType" << Endl;
   results->GetStorage()->Write();
   if (treetype==Types::kTesting) {
      // skipping plotting of variables if too many (default is 200)
      if ((int) DataInfo().GetNVariables()< gConfig().GetVariablePlotting().fMaxNumOfAllowedVariables)
         GetTransformationHandler().PlotVariables (GetEventCollection( Types::kTesting ), BaseDir() );
      else
         Log() << kINFO << TString::Format("Dataset[%s] : ",DataInfo().GetName())
               << " variable plots are not produces ! The number of variables is " << DataInfo().GetNVariables()
               << " , it is larger than " << gConfig().GetVariablePlotting().fMaxNumOfAllowedVariables << Endl;
   }
}

////////////////////////////////////////////////////////////////////////////////
/// write special monitoring histograms to file
/// dummy implementation here -----------------

void  TMVA::MethodBase::WriteMonitoringHistosToFile( void ) const
{
}

////////////////////////////////////////////////////////////////////////////////
/// reads one line from the input stream
/// checks for certain keywords and interprets
/// the line if keywords are found

Bool_t TMVA::MethodBase::GetLine(std::istream& fin, char* buf )
{
   fin.getline(buf,512);
   TString line(buf);
   if (line.BeginsWith("TMVA Release")) {
      Ssiz_t start  = line.First('[')+1;
      Ssiz_t length = line.Index("]",start)-start;
      TString code  = line(start,length);
      std::stringstream s(code.Data());
      s >> fTMVATrainingVersion;
      Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "MVA method was trained with TMVA Version: " << GetTrainingTMVAVersionString() << Endl;
   }
   if (line.BeginsWith("ROOT Release")) {
      Ssiz_t start  = line.First('[')+1;
      Ssiz_t length = line.Index("]",start)-start;
      TString code  = line(start,length);
      std::stringstream s(code.Data());
      s >> fROOTTrainingVersion;
      Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "MVA method was trained with ROOT Version: " << GetTrainingROOTVersionString() << Endl;
   }
   if (line.BeginsWith("Analysis type")) {
      Ssiz_t start  = line.First('[')+1;
      Ssiz_t length = line.Index("]",start)-start;
      TString code  = line(start,length);
      std::stringstream s(code.Data());
      std::string analysisType;
      s >> analysisType;
      if      (analysisType == "regression"     || analysisType == "Regression")     SetAnalysisType( Types::kRegression );
      else if (analysisType == "classification" || analysisType == "Classification") SetAnalysisType( Types::kClassification );
      else if (analysisType == "multiclass"     || analysisType == "Multiclass")     SetAnalysisType( Types::kMulticlass );
      else Log() << kFATAL << "Analysis type " << analysisType << " from weight-file not known!" << std::endl;

      Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Method was trained for "
            << (GetAnalysisType() == Types::kRegression ? "Regression" :
                (GetAnalysisType() == Types::kMulticlass ? "Multiclass" : "Classification")) << Endl;
   }

   return true;
}

////////////////////////////////////////////////////////////////////////////////
/// Create PDFs of the MVA output variables

void TMVA::MethodBase::CreateMVAPdfs()
{
   Data()->SetCurrentType(Types::kTraining);

   // the PDF's are stored as results ONLY if the corresponding "results" are booked,
   // otherwise they will be only used 'online'
   ResultsClassification * mvaRes = dynamic_cast<ResultsClassification*>
      ( Data()->GetResults(GetMethodName(), Types::kTraining, Types::kClassification) );

   if (mvaRes==0 || mvaRes->GetSize()==0) {
      Log() << kERROR<<Form("Dataset[%s] : ",DataInfo().GetName())<< "<CreateMVAPdfs> No result of classifier testing available" << Endl;
   }

   Double_t minVal = *std::min_element(mvaRes->GetValueVector()->begin(),mvaRes->GetValueVector()->end());
   Double_t maxVal = *std::max_element(mvaRes->GetValueVector()->begin(),mvaRes->GetValueVector()->end());

   // create histograms that serve as basis to create the MVA Pdfs
   TH1* histMVAPdfS = new TH1D( GetMethodTypeName() + "_tr_S", GetMethodTypeName() + "_tr_S",
                                fMVAPdfS->GetHistNBins( mvaRes->GetSize() ), minVal, maxVal );
   TH1* histMVAPdfB = new TH1D( GetMethodTypeName() + "_tr_B", GetMethodTypeName() + "_tr_B",
                                fMVAPdfB->GetHistNBins( mvaRes->GetSize() ), minVal, maxVal );


   // compute sum of weights properly
   histMVAPdfS->Sumw2();
   histMVAPdfB->Sumw2();

   // fill histograms
   for (UInt_t ievt=0; ievt<mvaRes->GetSize(); ievt++) {
      Double_t theVal    = mvaRes->GetValueVector()->at(ievt);
      Double_t theWeight = Data()->GetEvent(ievt)->GetWeight();

      if (DataInfo().IsSignal(Data()->GetEvent(ievt))) histMVAPdfS->Fill( theVal, theWeight );
      else                                             histMVAPdfB->Fill( theVal, theWeight );
   }

   gTools().NormHist( histMVAPdfS );
   gTools().NormHist( histMVAPdfB );

   // momentary hack for ROOT problem
   if(!IsSilentFile())
   {
        histMVAPdfS->Write();
        histMVAPdfB->Write();
   }
   // create PDFs
   fMVAPdfS->BuildPDF   ( histMVAPdfS );
   fMVAPdfB->BuildPDF   ( histMVAPdfB );
   fMVAPdfS->ValidatePDF( histMVAPdfS );
   fMVAPdfB->ValidatePDF( histMVAPdfB );

   if (DataInfo().GetNClasses() == 2) { // TODO: this is an ugly hack.. adapt this to new framework
      Log() << kINFO<<Form("Dataset[%s] : ",DataInfo().GetName())
            << Form( "<CreateMVAPdfs> Separation from histogram (PDF): %1.3f (%1.3f)",
                     GetSeparation( histMVAPdfS, histMVAPdfB ), GetSeparation( fMVAPdfS, fMVAPdfB ) )
            << Endl;
   }

   delete histMVAPdfS;
   delete histMVAPdfB;
}

Double_t TMVA::MethodBase::GetProba(const Event *ev){
   // the simple one, automatically calculates the mvaVal and uses the
   // SAME sig/bkg ratio as given in the training sample (typically 50/50
   // .. (NormMode=EqualNumEvents) but can be different)
   if (!fMVAPdfS || !fMVAPdfB) {
      Log() << kINFO<<Form("Dataset[%s] : ",DataInfo().GetName()) << "<GetProba> MVA PDFs for Signal and Background don't exist yet, we'll create them on demand" << Endl;
      CreateMVAPdfs();
   }
   Double_t sigFraction = DataInfo().GetTrainingSumSignalWeights() / (DataInfo().GetTrainingSumSignalWeights() + DataInfo().GetTrainingSumBackgrWeights() );
   Double_t mvaVal = GetMvaValue(ev);

   return GetProba(mvaVal,sigFraction);

}
////////////////////////////////////////////////////////////////////////////////
/// compute likelihood ratio

Double_t TMVA::MethodBase::GetProba( Double_t mvaVal, Double_t ap_sig )
{
   if (!fMVAPdfS || !fMVAPdfB) {
      Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetProba> MVA PDFs for Signal and Background don't exist" << Endl;
      return -1.0;
   }
   Double_t p_s = fMVAPdfS->GetVal( mvaVal );
   Double_t p_b = fMVAPdfB->GetVal( mvaVal );

   Double_t denom = p_s*ap_sig + p_b*(1 - ap_sig);

   return (denom > 0) ? (p_s*ap_sig) / denom : -1;
}

////////////////////////////////////////////////////////////////////////////////
/// compute rarity:
/// \f[
/// R(x) = \int_{[-\infty..x]} { PDF(x') dx' }
/// \f]
/// where PDF(x) is the PDF of the classifier's signal or background distribution

Double_t TMVA::MethodBase::GetRarity( Double_t mvaVal, Types::ESBType reftype ) const
{
   if ((reftype == Types::kSignal && !fMVAPdfS) || (reftype == Types::kBackground && !fMVAPdfB)) {
      Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetRarity> Required MVA PDF for Signal or Background does not exist: "
            << "select option \"CreateMVAPdfs\"" << Endl;
      return 0.0;
   }

   PDF* thePdf = ((reftype == Types::kSignal) ? fMVAPdfS : fMVAPdfB);

   return thePdf->GetIntegral( thePdf->GetXmin(), mvaVal );
}

////////////////////////////////////////////////////////////////////////////////
/// fill background efficiency (resp. rejection) versus signal efficiency plots
/// returns signal efficiency at background efficiency indicated in theString

Double_t TMVA::MethodBase::GetEfficiency( const TString& theString, Types::ETreeType type,Double_t& effSerr )
{
   Data()->SetCurrentType(type);
   Results* results = Data()->GetResults( GetMethodName(), type, Types::kClassification );
   std::vector<Float_t>* mvaRes = dynamic_cast<ResultsClassification*>(results)->GetValueVector();

   // parse input string for required background efficiency
   TList* list  = gTools().ParseFormatLine( theString );

   // sanity check
   Bool_t computeArea = kFALSE;
   if      (!list || list->GetSize() < 2) computeArea = kTRUE; // the area is computed
   else if (list->GetSize() > 2) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetEfficiency> Wrong number of arguments"
            << " in string: " << theString
            << " | required format, e.g., Efficiency:0.05, or empty string" << Endl;
      delete list;
      return -1;
   }

   // sanity check
   if ( results->GetHist("MVA_S")->GetNbinsX() != results->GetHist("MVA_B")->GetNbinsX() ||
        results->GetHist("MVA_HIGHBIN_S")->GetNbinsX() != results->GetHist("MVA_HIGHBIN_B")->GetNbinsX() ) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetEfficiency> Binning mismatch between signal and background histos" << Endl;
      delete list;
      return -1.0;
   }

   // create histograms

   // first, get efficiency histograms for signal and background
   TH1 * effhist = results->GetHist("MVA_HIGHBIN_S");
   Double_t xmin = effhist->GetXaxis()->GetXmin();
   Double_t xmax = effhist->GetXaxis()->GetXmax();

   TTHREAD_TLS(Double_t) nevtS;

   // first round ? --> create histograms
   if (results->DoesExist("MVA_EFF_S")==0) {

      // for efficiency plot
      TH1* eff_s = new TH1D( GetTestvarName() + "_effS", GetTestvarName() + " (signal)",     fNbinsH, xmin, xmax );
      TH1* eff_b = new TH1D( GetTestvarName() + "_effB", GetTestvarName() + " (background)", fNbinsH, xmin, xmax );
      results->Store(eff_s, "MVA_EFF_S");
      results->Store(eff_b, "MVA_EFF_B");

      // sign if cut
      Int_t sign = (fCutOrientation == kPositive) ? +1 : -1;

      // this method is unbinned
      nevtS = 0;
      for (UInt_t ievt=0; ievt<Data()->GetNEvents(); ievt++) {

         // read the tree
         Bool_t  isSignal  = DataInfo().IsSignal(GetEvent(ievt));
         Float_t theWeight = GetEvent(ievt)->GetWeight();
         Float_t theVal    = (*mvaRes)[ievt];

         // select histogram depending on if sig or bgd
         TH1* theHist = isSignal ? eff_s : eff_b;

         // count signal and background events in tree
         if (isSignal) nevtS+=theWeight;

         TAxis* axis   = theHist->GetXaxis();
         Int_t  maxbin = Int_t((theVal - axis->GetXmin())/(axis->GetXmax() - axis->GetXmin())*fNbinsH) + 1;
         if (sign > 0 && maxbin > fNbinsH) continue; // can happen... event doesn't count
         if (sign < 0 && maxbin < 1      ) continue; // can happen... event doesn't count
         if (sign > 0 && maxbin < 1      ) maxbin = 1;
         if (sign < 0 && maxbin > fNbinsH) maxbin = fNbinsH;

         if (sign > 0)
            for (Int_t ibin=1; ibin<=maxbin; ibin++) theHist->AddBinContent( ibin , theWeight);
         else if (sign < 0)
            for (Int_t ibin=maxbin+1; ibin<=fNbinsH; ibin++) theHist->AddBinContent( ibin , theWeight );
         else
            Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetEfficiency> Mismatch in sign" << Endl;
      }

      // renormalise maximum to <=1
      // eff_s->Scale( 1.0/TMath::Max(1.,eff_s->GetMaximum()) );
      // eff_b->Scale( 1.0/TMath::Max(1.,eff_b->GetMaximum()) );

      eff_s->Scale( 1.0/TMath::Max(std::numeric_limits<double>::epsilon(),eff_s->GetMaximum()) );
      eff_b->Scale( 1.0/TMath::Max(std::numeric_limits<double>::epsilon(),eff_b->GetMaximum()) );

      // background efficiency versus signal efficiency
      TH1* eff_BvsS = new TH1D( GetTestvarName() + "_effBvsS", GetTestvarName() + "", fNbins, 0, 1 );
      results->Store(eff_BvsS, "MVA_EFF_BvsS");
      eff_BvsS->SetXTitle( "Signal eff" );
      eff_BvsS->SetYTitle( "Backgr eff" );

      // background rejection (=1-eff.) versus signal efficiency
      TH1* rej_BvsS = new TH1D( GetTestvarName() + "_rejBvsS", GetTestvarName() + "", fNbins, 0, 1 );
      results->Store(rej_BvsS);
      rej_BvsS->SetXTitle( "Signal eff" );
      rej_BvsS->SetYTitle( "Backgr rejection (1-eff)" );

      // inverse background eff (1/eff.) versus signal efficiency
      TH1* inveff_BvsS = new TH1D( GetTestvarName() + "_invBeffvsSeff",
                                   GetTestvarName(), fNbins, 0, 1 );
      results->Store(inveff_BvsS);
      inveff_BvsS->SetXTitle( "Signal eff" );
      inveff_BvsS->SetYTitle( "Inverse backgr. eff (1/eff)" );

      // use root finder
      // spline background efficiency plot
      // note that there is a bin shift when going from a TH1D object to a TGraph :-(
      if (Use_Splines_for_Eff_) {
         fSplRefS  = new TSpline1( "spline2_signal",     new TGraph( eff_s ) );
         fSplRefB  = new TSpline1( "spline2_background", new TGraph( eff_b ) );

         // verify spline sanity
         gTools().CheckSplines( eff_s, fSplRefS );
         gTools().CheckSplines( eff_b, fSplRefB );
      }

      // make the background-vs-signal efficiency plot

      // create root finder
      RootFinder rootFinder( this, fXmin, fXmax );

      Double_t effB = 0;
      fEffS = eff_s; // to be set for the root finder
      for (Int_t bini=1; bini<=fNbins; bini++) {

         // find cut value corresponding to a given signal efficiency
         Double_t effS = eff_BvsS->GetBinCenter( bini );
         Double_t cut  = rootFinder.Root( effS );

         // retrieve background efficiency for given cut
         if (Use_Splines_for_Eff_) effB = fSplRefB->Eval( cut );
         else                      effB = eff_b->GetBinContent( eff_b->FindBin( cut ) );

         // and fill histograms
         eff_BvsS->SetBinContent( bini, effB     );
         rej_BvsS->SetBinContent( bini, 1.0-effB );
         if (effB>std::numeric_limits<double>::epsilon())
            inveff_BvsS->SetBinContent( bini, 1.0/effB );
      }

      // create splines for histogram
      fSpleffBvsS = new TSpline1( "effBvsS", new TGraph( eff_BvsS ) );

      // search for overlap point where, when cutting on it,
      // one would obtain: eff_S = rej_B = 1 - eff_B
      Double_t effS = 0., rejB, effS_ = 0., rejB_ = 0.;
      Int_t    nbins_ = 5000;
      for (Int_t bini=1; bini<=nbins_; bini++) {

         // get corresponding signal and background efficiencies
         effS = (bini - 0.5)/Float_t(nbins_);
         rejB = 1.0 - fSpleffBvsS->Eval( effS );

         // find signal efficiency that corresponds to required background efficiency
         if ((effS - rejB)*(effS_ - rejB_) < 0) break;
         effS_ = effS;
         rejB_ = rejB;
      }

      // find cut that corresponds to signal efficiency and update signal-like criterion
      Double_t cut = rootFinder.Root( 0.5*(effS + effS_) );
      SetSignalReferenceCut( cut );
      fEffS = 0;
   }

   // must exist...
   if (0 == fSpleffBvsS) {
      delete list;
      return 0.0;
   }

   // now find signal efficiency that corresponds to required background efficiency
   Double_t effS = 0, effB = 0, effS_ = 0, effB_ = 0;
   Int_t    nbins_ = 1000;

   if (computeArea) {

      // compute area of rej-vs-eff plot
      Double_t integral = 0;
      for (Int_t bini=1; bini<=nbins_; bini++) {

         // get corresponding signal and background efficiencies
         effS = (bini - 0.5)/Float_t(nbins_);
         effB = fSpleffBvsS->Eval( effS );
         integral += (1.0 - effB);
      }
      integral /= nbins_;

      delete list;
      return integral;
   }
   else {

      // that will be the value of the efficiency retured (does not affect
      // the efficiency-vs-bkg plot which is done anyway.
      Float_t effBref  = atof( ((TObjString*)list->At(1))->GetString() );

      // find precise efficiency value
      for (Int_t bini=1; bini<=nbins_; bini++) {

         // get corresponding signal and background efficiencies
         effS = (bini - 0.5)/Float_t(nbins_);
         effB = fSpleffBvsS->Eval( effS );

         // find signal efficiency that corresponds to required background efficiency
         if ((effB - effBref)*(effB_ - effBref) <= 0) break;
         effS_ = effS;
         effB_ = effB;
      }

      // take mean between bin above and bin below
      effS = 0.5*(effS + effS_);

      effSerr = 0;
      if (nevtS > 0) effSerr = TMath::Sqrt( effS*(1.0 - effS)/nevtS );

      delete list;
      return effS;
   }

   return -1;
}

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

Double_t TMVA::MethodBase::GetTrainingEfficiency(const TString& theString)
{
   Data()->SetCurrentType(Types::kTraining);

   Results* results = Data()->GetResults(GetMethodName(), Types::kTesting, Types::kNoAnalysisType);

   // fill background efficiency (resp. rejection) versus signal efficiency plots
   // returns signal efficiency at background efficiency indicated in theString

   // parse input string for required background efficiency
   TList*  list  = gTools().ParseFormatLine( theString );
   // sanity check

   if (list->GetSize() != 2) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetTrainingEfficiency> Wrong number of arguments"
            << " in string: " << theString
            << " | required format, e.g., Efficiency:0.05" << Endl;
      delete list;
      return -1;
   }
   // that will be the value of the efficiency retured (does not affect
   // the efficiency-vs-bkg plot which is done anyway.
   Float_t effBref  = atof( ((TObjString*)list->At(1))->GetString() );

   delete list;

   // sanity check
   if (results->GetHist("MVA_S")->GetNbinsX() != results->GetHist("MVA_B")->GetNbinsX() ||
       results->GetHist("MVA_HIGHBIN_S")->GetNbinsX() != results->GetHist("MVA_HIGHBIN_B")->GetNbinsX() ) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetTrainingEfficiency> Binning mismatch between signal and background histos"
            << Endl;
      return -1.0;
   }

   // create histogram

   // first, get efficiency histograms for signal and background
   TH1 * effhist = results->GetHist("MVA_HIGHBIN_S");
   Double_t xmin = effhist->GetXaxis()->GetXmin();
   Double_t xmax = effhist->GetXaxis()->GetXmax();

   // first round ? --> create and fill histograms
   if (results->DoesExist("MVA_TRAIN_S")==0) {

      // classifier response distributions for test sample
      Double_t sxmax = fXmax+0.00001;

      // MVA plots on the training sample (check for overtraining)
      TH1* mva_s_tr = new TH1D( GetTestvarName() + "_Train_S",GetTestvarName() + "_Train_S", fNbinsMVAoutput, fXmin, sxmax );
      TH1* mva_b_tr = new TH1D( GetTestvarName() + "_Train_B",GetTestvarName() + "_Train_B", fNbinsMVAoutput, fXmin, sxmax );
      results->Store(mva_s_tr, "MVA_TRAIN_S");
      results->Store(mva_b_tr, "MVA_TRAIN_B");
      mva_s_tr->Sumw2();
      mva_b_tr->Sumw2();

      // Training efficiency plots
      TH1* mva_eff_tr_s = new TH1D( GetTestvarName() + "_trainingEffS", GetTestvarName() + " (signal)",
                                    fNbinsH, xmin, xmax );
      TH1* mva_eff_tr_b = new TH1D( GetTestvarName() + "_trainingEffB", GetTestvarName() + " (background)",
                                    fNbinsH, xmin, xmax );
      results->Store(mva_eff_tr_s, "MVA_TRAINEFF_S");
      results->Store(mva_eff_tr_b, "MVA_TRAINEFF_B");

      // sign if cut
      Int_t sign = (fCutOrientation == kPositive) ? +1 : -1;

      std::vector<Double_t> mvaValues = GetMvaValues(0,Data()->GetNEvents());
      assert( (Long64_t) mvaValues.size() == Data()->GetNEvents());

      // this method is unbinned
      for (Int_t ievt=0; ievt<Data()->GetNEvents(); ievt++) {

         Data()->SetCurrentEvent(ievt);
         const Event* ev = GetEvent();

         Double_t theVal    = mvaValues[ievt];
         Double_t theWeight = ev->GetWeight();

         TH1* theEffHist = DataInfo().IsSignal(ev) ? mva_eff_tr_s : mva_eff_tr_b;
         TH1* theClsHist = DataInfo().IsSignal(ev) ? mva_s_tr : mva_b_tr;

         theClsHist->Fill( theVal, theWeight );

         TAxis* axis   = theEffHist->GetXaxis();
         Int_t  maxbin = Int_t((theVal - axis->GetXmin())/(axis->GetXmax() - axis->GetXmin())*fNbinsH) + 1;
         if (sign > 0 && maxbin > fNbinsH) continue; // can happen... event doesn't count
         if (sign < 0 && maxbin < 1      ) continue; // can happen... event doesn't count
         if (sign > 0 && maxbin < 1      ) maxbin = 1;
         if (sign < 0 && maxbin > fNbinsH) maxbin = fNbinsH;

         if (sign > 0) for (Int_t ibin=1;        ibin<=maxbin;  ibin++) theEffHist->AddBinContent( ibin , theWeight );
         else          for (Int_t ibin=maxbin+1; ibin<=fNbinsH; ibin++) theEffHist->AddBinContent( ibin , theWeight );
      }

      // normalise output distributions
      // uncomment those (and several others if you want unnormalized output
      gTools().NormHist( mva_s_tr  );
      gTools().NormHist( mva_b_tr  );

      // renormalise to maximum
      mva_eff_tr_s->Scale( 1.0/TMath::Max(std::numeric_limits<double>::epsilon(), mva_eff_tr_s->GetMaximum()) );
      mva_eff_tr_b->Scale( 1.0/TMath::Max(std::numeric_limits<double>::epsilon(), mva_eff_tr_b->GetMaximum()) );

      // Training background efficiency versus signal efficiency
      TH1* eff_bvss = new TH1D( GetTestvarName() + "_trainingEffBvsS", GetTestvarName() + "", fNbins, 0, 1 );
      // Training background rejection (=1-eff.) versus signal efficiency
      TH1* rej_bvss = new TH1D( GetTestvarName() + "_trainingRejBvsS", GetTestvarName() + "", fNbins, 0, 1 );
      results->Store(eff_bvss, "EFF_BVSS_TR");
      results->Store(rej_bvss, "REJ_BVSS_TR");

      // use root finder
      // spline background efficiency plot
      // note that there is a bin shift when going from a TH1D object to a TGraph :-(
      if (Use_Splines_for_Eff_) {
         if (fSplTrainRefS) delete fSplTrainRefS;
         if (fSplTrainRefB) delete fSplTrainRefB;
         fSplTrainRefS  = new TSpline1( "spline2_signal",     new TGraph( mva_eff_tr_s ) );
         fSplTrainRefB  = new TSpline1( "spline2_background", new TGraph( mva_eff_tr_b ) );

         // verify spline sanity
         gTools().CheckSplines( mva_eff_tr_s, fSplTrainRefS );
         gTools().CheckSplines( mva_eff_tr_b, fSplTrainRefB );
      }

      // make the background-vs-signal efficiency plot

      // create root finder
      RootFinder rootFinder(this, fXmin, fXmax );

      Double_t effB = 0;
      fEffS = results->GetHist("MVA_TRAINEFF_S");
      for (Int_t bini=1; bini<=fNbins; bini++) {

         // find cut value corresponding to a given signal efficiency
         Double_t effS = eff_bvss->GetBinCenter( bini );

         Double_t cut  = rootFinder.Root( effS );

         // retrieve background efficiency for given cut
         if (Use_Splines_for_Eff_) effB = fSplTrainRefB->Eval( cut );
         else                      effB = mva_eff_tr_b->GetBinContent( mva_eff_tr_b->FindBin( cut ) );

         // and fill histograms
         eff_bvss->SetBinContent( bini, effB     );
         rej_bvss->SetBinContent( bini, 1.0-effB );
      }
      fEffS = 0;

      // create splines for histogram
      fSplTrainEffBvsS = new TSpline1( "effBvsS", new TGraph( eff_bvss ) );
   }

   // must exist...
   if (0 == fSplTrainEffBvsS) return 0.0;

   // now find signal efficiency that corresponds to required background efficiency
   Double_t effS = 0., effB, effS_ = 0., effB_ = 0.;
   Int_t    nbins_ = 1000;
   for (Int_t bini=1; bini<=nbins_; bini++) {

      // get corresponding signal and background efficiencies
      effS = (bini - 0.5)/Float_t(nbins_);
      effB = fSplTrainEffBvsS->Eval( effS );

      // find signal efficiency that corresponds to required background efficiency
      if ((effB - effBref)*(effB_ - effBref) <= 0) break;
      effS_ = effS;
      effB_ = effB;
   }

   return 0.5*(effS + effS_); // the mean between bin above and bin below
}

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

std::vector<Float_t> TMVA::MethodBase::GetMulticlassEfficiency(std::vector<std::vector<Float_t> >& purity)
{
   Data()->SetCurrentType(Types::kTesting);
   ResultsMulticlass* resMulticlass = dynamic_cast<ResultsMulticlass*>(Data()->GetResults(GetMethodName(), Types::kTesting, Types::kMulticlass));
   if (!resMulticlass) Log() << kFATAL<<Form("Dataset[%s] : ",DataInfo().GetName())<< "unable to create pointer in GetMulticlassEfficiency, exiting."<<Endl;

   purity.push_back(resMulticlass->GetAchievablePur());
   return resMulticlass->GetAchievableEff();
}

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

std::vector<Float_t> TMVA::MethodBase::GetMulticlassTrainingEfficiency(std::vector<std::vector<Float_t> >& purity)
{
   Data()->SetCurrentType(Types::kTraining);
   ResultsMulticlass* resMulticlass = dynamic_cast<ResultsMulticlass*>(Data()->GetResults(GetMethodName(), Types::kTraining, Types::kMulticlass));
   if (!resMulticlass) Log() << kFATAL<< "unable to create pointer in GetMulticlassTrainingEfficiency, exiting."<<Endl;

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Determine optimal multiclass cuts for training data..." << Endl;
   for (UInt_t icls = 0; icls<DataInfo().GetNClasses(); ++icls) {
      resMulticlass->GetBestMultiClassCuts(icls);
   }

   purity.push_back(resMulticlass->GetAchievablePur());
   return resMulticlass->GetAchievableEff();
}

////////////////////////////////////////////////////////////////////////////////
/// Construct a confusion matrix for a multiclass classifier. The confusion
/// matrix compares, in turn, each class agaist all other classes in a pair-wise
/// fashion. In rows with index \f$ k_r = 0 ... K \f$, \f$ k_r \f$ is
/// considered signal for the sake of comparison and for each column
/// \f$ k_c = 0 ... K \f$ the corresponding class is considered background.
///
/// Note that the diagonal elements will be returned as NaN since this will
/// compare a class against itself.
///
/// \see TMVA::ResultsMulticlass::GetConfusionMatrix
///
/// \param[in] effB The background efficiency for which to evaluate.
/// \param[in] type The data set on which to evaluate (training, testing ...).
///
/// \return A matrix containing signal efficiencies for the given background
///         efficiency. The diagonal elements are NaN since this measure is
///         meaningless (comparing a class against itself).
///

TMatrixD TMVA::MethodBase::GetMulticlassConfusionMatrix(Double_t effB, Types::ETreeType type)
{
   if (GetAnalysisType() != Types::kMulticlass) {
      Log() << kFATAL << "Cannot get confusion matrix for non-multiclass analysis." << std::endl;
      return TMatrixD(0, 0);
   }

   Data()->SetCurrentType(type);
   ResultsMulticlass *resMulticlass =
      dynamic_cast<ResultsMulticlass *>(Data()->GetResults(GetMethodName(), type, Types::kMulticlass));

   if (resMulticlass == nullptr) {
      Log() << kFATAL << Form("Dataset[%s] : ", DataInfo().GetName())
            << "unable to create pointer in GetMulticlassEfficiency, exiting." << Endl;
      return TMatrixD(0, 0);
   }

   return resMulticlass->GetConfusionMatrix(effB);
}

////////////////////////////////////////////////////////////////////////////////
/// compute significance of mean difference
/// \f[
/// significance = \frac{|<S> - <B>|}{\sqrt{RMS_{S2} + RMS_{B2}}}
/// \f]

Double_t TMVA::MethodBase::GetSignificance( void ) const
{
   Double_t rms = sqrt( fRmsS*fRmsS + fRmsB*fRmsB );

   return (rms > 0) ? TMath::Abs(fMeanS - fMeanB)/rms : 0;
}

////////////////////////////////////////////////////////////////////////////////
/// compute "separation" defined as
/// \f[
/// <s2> = \frac{1}{2} \int_{-\infty}^{+\infty} { \frac{(S(x) - B(x))^2}{(S(x) + B(x))} dx }
/// \f]

Double_t TMVA::MethodBase::GetSeparation( TH1* histoS, TH1* histoB ) const
{
   return gTools().GetSeparation( histoS, histoB );
}

////////////////////////////////////////////////////////////////////////////////
/// compute "separation" defined as
/// \f[
/// <s2> = \frac{1}{2} \int_{-\infty}^{+\infty} { \frac{(S(x) - B(x))^2}{(S(x) + B(x))} dx }
/// \f]

Double_t TMVA::MethodBase::GetSeparation( PDF* pdfS, PDF* pdfB ) const
{
   // note, if zero pointers given, use internal pdf
   // sanity check first
   if ((!pdfS && pdfB) || (pdfS && !pdfB))
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetSeparation> Mismatch in pdfs" << Endl;
   if (!pdfS) pdfS = fSplS;
   if (!pdfB) pdfB = fSplB;

   if (!fSplS || !fSplB) {
      Log()<<kDEBUG<<Form("[%s] : ",DataInfo().GetName())<< "could not calculate the separation, distributions"
           << " fSplS or fSplB are not yet filled" << Endl;
      return 0;
   }else{
      return gTools().GetSeparation( *pdfS, *pdfB );
   }
}

////////////////////////////////////////////////////////////////////////////////
/// calculate the area (integral) under the ROC curve as a
/// overall quality measure of the classification

Double_t TMVA::MethodBase::GetROCIntegral(TH1D *histS, TH1D *histB) const
{
   // note, if zero pointers given, use internal pdf
   // sanity check first
   if ((!histS && histB) || (histS && !histB))
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetROCIntegral(TH1D*, TH1D*)> Mismatch in hists" << Endl;

   if (histS==0 || histB==0) return 0.;

   TMVA::PDF *pdfS = new TMVA::PDF( " PDF Sig", histS, TMVA::PDF::kSpline3 );
   TMVA::PDF *pdfB = new TMVA::PDF( " PDF Bkg", histB, TMVA::PDF::kSpline3 );


   Double_t xmin = TMath::Min(pdfS->GetXmin(), pdfB->GetXmin());
   Double_t xmax = TMath::Max(pdfS->GetXmax(), pdfB->GetXmax());

   Double_t integral = 0;
   UInt_t   nsteps = 1000;
   Double_t step = (xmax-xmin)/Double_t(nsteps);
   Double_t cut = xmin;
   for (UInt_t i=0; i<nsteps; i++) {
      integral += (1-pdfB->GetIntegral(cut,xmax)) * pdfS->GetVal(cut);
      cut+=step;
   }
   delete pdfS;
   delete pdfB;
   return integral*step;
}


////////////////////////////////////////////////////////////////////////////////
/// calculate the area (integral) under the ROC curve as a
/// overall quality measure of the classification

Double_t TMVA::MethodBase::GetROCIntegral(PDF *pdfS, PDF *pdfB) const
{
   // note, if zero pointers given, use internal pdf
   // sanity check first
   if ((!pdfS && pdfB) || (pdfS && !pdfB))
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetSeparation> Mismatch in pdfs" << Endl;
   if (!pdfS) pdfS = fSplS;
   if (!pdfB) pdfB = fSplB;

   if (pdfS==0 || pdfB==0) return 0.;

   Double_t xmin = TMath::Min(pdfS->GetXmin(), pdfB->GetXmin());
   Double_t xmax = TMath::Max(pdfS->GetXmax(), pdfB->GetXmax());

   Double_t integral = 0;
   UInt_t   nsteps = 1000;
   Double_t step = (xmax-xmin)/Double_t(nsteps);
   Double_t cut = xmin;
   for (UInt_t i=0; i<nsteps; i++) {
      integral += (1-pdfB->GetIntegral(cut,xmax)) * pdfS->GetVal(cut);
      cut+=step;
   }
   return integral*step;
}

////////////////////////////////////////////////////////////////////////////////
/// plot significance, \f$ \frac{S}{\sqrt{S^2 + B^2}} \f$, curve for given number
/// of signal and background events; returns cut for maximum significance
/// also returned via reference is the maximum significance

Double_t TMVA::MethodBase::GetMaximumSignificance( Double_t SignalEvents,
                                                   Double_t BackgroundEvents,
                                                   Double_t& max_significance_value ) const
{
   Results* results = Data()->GetResults( GetMethodName(), Types::kTesting, Types::kMaxAnalysisType );

   Double_t max_significance(0);
   Double_t effS(0),effB(0),significance(0);
   TH1D *temp_histogram = new TH1D("temp", "temp", fNbinsH, fXmin, fXmax );

   if (SignalEvents <= 0 || BackgroundEvents <= 0) {
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<GetMaximumSignificance> "
            << "Number of signal or background events is <= 0 ==> abort"
            << Endl;
   }

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Using ratio SignalEvents/BackgroundEvents = "
         << SignalEvents/BackgroundEvents << Endl;

   TH1* eff_s = results->GetHist("MVA_EFF_S");
   TH1* eff_b = results->GetHist("MVA_EFF_B");

   if ( (eff_s==0) || (eff_b==0) ) {
      Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Efficiency histograms empty !" << Endl;
      Log() << kWARNING <<Form("Dataset[%s] : ",DataInfo().GetName())<< "no maximum cut found, return 0" << Endl;
      return 0;
   }

   for (Int_t bin=1; bin<=fNbinsH; bin++) {
      effS = eff_s->GetBinContent( bin );
      effB = eff_b->GetBinContent( bin );

      // put significance into a histogram
      significance = sqrt(SignalEvents)*( effS )/sqrt( effS + ( BackgroundEvents / SignalEvents) * effB  );

      temp_histogram->SetBinContent(bin,significance);
   }

   // find maximum in histogram
   max_significance = temp_histogram->GetBinCenter( temp_histogram->GetMaximumBin() );
   max_significance_value = temp_histogram->GetBinContent( temp_histogram->GetMaximumBin() );

   // delete
   delete temp_histogram;

   Log() << kINFO <<Form("Dataset[%s] : ",DataInfo().GetName())<< "Optimal cut at      : " << max_significance << Endl;
   Log() << kINFO<<Form("Dataset[%s] : ",DataInfo().GetName()) << "Maximum significance: " << max_significance_value << Endl;

   return max_significance;
}

////////////////////////////////////////////////////////////////////////////////
/// calculates rms,mean, xmin, xmax of the event variable
/// this can be either done for the variables as they are or for
/// normalised variables (in the range of 0-1) if "norm" is set to kTRUE

void TMVA::MethodBase::Statistics( Types::ETreeType treeType, const TString& theVarName,
                                   Double_t& meanS, Double_t& meanB,
                                   Double_t& rmsS,  Double_t& rmsB,
                                   Double_t& xmin,  Double_t& xmax )
{
   Types::ETreeType previousTreeType = Data()->GetCurrentType();
   Data()->SetCurrentType(treeType);

   Long64_t entries = Data()->GetNEvents();

   // sanity check
   if (entries <=0)
      Log() << kFATAL <<Form("Dataset[%s] : ",DataInfo().GetName())<< "<CalculateEstimator> Wrong tree type: " << treeType << Endl;

   // index of the wanted variable
   UInt_t varIndex = DataInfo().FindVarIndex( theVarName );

   // first fill signal and background in arrays before analysis
   xmin               = +DBL_MAX;
   xmax               = -DBL_MAX;
   Long64_t nEventsS  = -1;
   Long64_t nEventsB  = -1;

   // take into account event weights
   meanS = 0;
   meanB = 0;
   rmsS  = 0;
   rmsB  = 0;
   Double_t sumwS = 0, sumwB = 0;

   // loop over all training events
   for (Int_t ievt = 0; ievt < entries; ievt++) {

      const Event* ev = GetEvent(ievt);

      Double_t theVar = ev->GetValue(varIndex);
      Double_t weight = ev->GetWeight();

      if (DataInfo().IsSignal(ev)) {
         sumwS               += weight;
         meanS               += weight*theVar;
         rmsS                += weight*theVar*theVar;
      }
      else {
         sumwB               += weight;
         meanB               += weight*theVar;
         rmsB                += weight*theVar*theVar;
      }
      xmin = TMath::Min( xmin, theVar );
      xmax = TMath::Max( xmax, theVar );
   }
   ++nEventsS;
   ++nEventsB;

   meanS = meanS/sumwS;
   meanB = meanB/sumwB;
   rmsS  = TMath::Sqrt( rmsS/sumwS - meanS*meanS );
   rmsB  = TMath::Sqrt( rmsB/sumwB - meanB*meanB );

   Data()->SetCurrentType(previousTreeType);
}

////////////////////////////////////////////////////////////////////////////////
/// create reader class for method (classification only at present)

void TMVA::MethodBase::MakeClass( const TString& theClassFileName ) const
{
   // the default consists of
   TString classFileName = "";
   if (theClassFileName == "")
      classFileName = GetWeightFileDir() + "/" + GetJobName() + "_" + GetMethodName() + ".class.C";
   else
      classFileName = theClassFileName;

   TString className = TString("Read") + GetMethodName();

   TString tfname( classFileName );
   Log() << kINFO //<<Form("Dataset[%s] : ",DataInfo().GetName())
    << "Creating standalone class: "
         << gTools().Color("lightblue") << classFileName << gTools().Color("reset") << Endl;

   std::ofstream fout( classFileName );
   if (!fout.good()) { // file could not be opened --> Error
      Log() << kFATAL << "<MakeClass> Unable to open file: " << classFileName << Endl;
   }

   // now create the class
   // preamble
   fout << "// Class: " << className << std::endl;
   fout << "// Automatically generated by MethodBase::MakeClass" << std::endl << "//" << std::endl;

   // print general information and configuration state
   fout << std::endl;
   fout << "/* configuration options =====================================================" << std::endl << std::endl;
   WriteStateToStream( fout );
   fout << std::endl;
   fout << "============================================================================ */" << std::endl;

   // generate the class
   fout << "" << std::endl;
   fout << "#include <array>" << std::endl;
   fout << "#include <vector>" << std::endl;
   fout << "#include <cmath>" << std::endl;
   fout << "#include <string>" << std::endl;
   fout << "#include <iostream>" << std::endl;
   fout << "" << std::endl;
   // now if the classifier needs to write some additional classes for its response implementation
   // this code goes here: (at least the header declarations need to come before the main class
   this->MakeClassSpecificHeader( fout, className );

   fout << "#ifndef IClassifierReader__def" << std::endl;
   fout << "#define IClassifierReader__def" << std::endl;
   fout << std::endl;
   fout << "class IClassifierReader {" << std::endl;
   fout << std::endl;
   fout << " public:" << std::endl;
   fout << std::endl;
   fout << "   // constructor" << std::endl;
   fout << "   IClassifierReader() : fStatusIsClean( true ) {}" << std::endl;
   fout << "   virtual ~IClassifierReader() {}" << std::endl;
   fout << std::endl;
   fout << "   // return classifier response" << std::endl;
   if(GetAnalysisType() == Types::kMulticlass) {
      fout << "   virtual std::vector<double> GetMulticlassValues( const std::vector<double>& inputValues ) const = 0;" << std::endl;
   } else {
      fout << "   virtual double GetMvaValue( const std::vector<double>& inputValues ) const = 0;" << std::endl;
   }
   fout << std::endl;
   fout << "   // returns classifier status" << std::endl;
   fout << "   bool IsStatusClean() const { return fStatusIsClean; }" << std::endl;
   fout << std::endl;
   fout << " protected:" << std::endl;
   fout << std::endl;
   fout << "   bool fStatusIsClean;" << std::endl;
   fout << "};" << std::endl;
   fout << std::endl;
   fout << "#endif" << std::endl;
   fout << std::endl;
   fout << "class " << className << " : public IClassifierReader {" << std::endl;
   fout << std::endl;
   fout << " public:" << std::endl;
   fout << std::endl;
   fout << "   // constructor" << std::endl;
   fout << "   " << className << "( std::vector<std::string>& theInputVars )" << std::endl;
   fout << "      : IClassifierReader()," << std::endl;
   fout << "        fClassName( \"" << className << "\" )," << std::endl;
   fout << "        fNvars( " << GetNvar() << " )" << std::endl;
   fout << "   {" << std::endl;
   fout << "      // the training input variables" << std::endl;
   fout << "      const char* inputVars[] = { ";
   for (UInt_t ivar=0; ivar<GetNvar(); ivar++) {
      fout << "\"" << GetOriginalVarName(ivar) << "\"";
      if (ivar<GetNvar()-1) fout << ", ";
   }
   fout << " };" << std::endl;
   fout << std::endl;
   fout << "      // sanity checks" << std::endl;
   fout << "      if (theInputVars.size() <= 0) {" << std::endl;
   fout << "         std::cout << \"Problem in class \\\"\" << fClassName << \"\\\": empty input vector\" << std::endl;" << std::endl;
   fout << "         fStatusIsClean = false;" << std::endl;
   fout << "      }" << std::endl;
   fout << std::endl;
   fout << "      if (theInputVars.size() != fNvars) {" << std::endl;
   fout << "         std::cout << \"Problem in class \\\"\" << fClassName << \"\\\": mismatch in number of input values: \"" << std::endl;
   fout << "                   << theInputVars.size() << \" != \" << fNvars << std::endl;" << std::endl;
   fout << "         fStatusIsClean = false;" << std::endl;
   fout << "      }" << std::endl;
   fout << std::endl;
   fout << "      // validate input variables" << std::endl;
   fout << "      for (size_t ivar = 0; ivar < theInputVars.size(); ivar++) {" << std::endl;
   fout << "         if (theInputVars[ivar] != inputVars[ivar]) {" << std::endl;
   fout << "            std::cout << \"Problem in class \\\"\" << fClassName << \"\\\": mismatch in input variable names\" << std::endl" << std::endl;
   fout << "                      << \" for variable [\" << ivar << \"]: \" << theInputVars[ivar].c_str() << \" != \" << inputVars[ivar] << std::endl;" << std::endl;
   fout << "            fStatusIsClean = false;" << std::endl;
   fout << "         }" << std::endl;
   fout << "      }" << std::endl;
   fout << std::endl;
   fout << "      // initialize min and max vectors (for normalisation)" << std::endl;
   for (UInt_t ivar = 0; ivar < GetNvar(); ivar++) {
      fout << "      fVmin[" << ivar << "] = " << std::setprecision(15) << GetXmin( ivar ) << ";" << std::endl;
      fout << "      fVmax[" << ivar << "] = " << std::setprecision(15) << GetXmax( ivar ) << ";" << std::endl;
   }
   fout << std::endl;
   fout << "      // initialize input variable types" << std::endl;
   for (UInt_t ivar=0; ivar<GetNvar(); ivar++) {
      fout << "      fType[" << ivar << "] = \'" << DataInfo().GetVariableInfo(ivar).GetVarType() << "\';" << std::endl;
   }
   fout << std::endl;
   fout << "      // initialize constants" << std::endl;
   fout << "      Initialize();" << std::endl;
   fout << std::endl;
   if (GetTransformationHandler().GetTransformationList().GetSize() != 0) {
      fout << "      // initialize transformation" << std::endl;
      fout << "      InitTransform();" << std::endl;
   }
   fout << "   }" << std::endl;
   fout << std::endl;
   fout << "   // destructor" << std::endl;
   fout << "   virtual ~" << className << "() {" << std::endl;
   fout << "      Clear(); // method-specific" << std::endl;
   fout << "   }" << std::endl;
   fout << std::endl;
   fout << "   // the classifier response" << std::endl;
   fout << "   // \"inputValues\" is a vector of input values in the same order as the" << std::endl;
   fout << "   // variables given to the constructor" << std::endl;
   if(GetAnalysisType() == Types::kMulticlass) {
      fout << "   std::vector<double> GetMulticlassValues( const std::vector<double>& inputValues ) const override;" << std::endl;
   } else {
      fout << "   double GetMvaValue( const std::vector<double>& inputValues ) const override;" << std::endl;
   }
   fout << std::endl;
   fout << " private:" << std::endl;
   fout << std::endl;
   fout << "   // method-specific destructor" << std::endl;
   fout << "   void Clear();" << std::endl;
   fout << std::endl;
   if (GetTransformationHandler().GetTransformationList().GetSize()!=0) {
      fout << "   // input variable transformation" << std::endl;
      GetTransformationHandler().MakeFunction(fout, className,1);
      fout << "   void InitTransform();" << std::endl;
      fout << "   void Transform( std::vector<double> & iv, int sigOrBgd ) const;" << std::endl;
      fout << std::endl;
   }
   fout << "   // common member variables" << std::endl;
   fout << "   const char* fClassName;" << std::endl;
   fout << std::endl;
   fout << "   const size_t fNvars;" << std::endl;
   fout << "   size_t GetNvar()           const { return fNvars; }" << std::endl;
   fout << "   char   GetType( int ivar ) const { return fType[ivar]; }" << std::endl;
   fout << std::endl;
   fout << "   // normalisation of input variables" << std::endl;
   fout << "   double fVmin[" << GetNvar() << "];" << std::endl;
   fout << "   double fVmax[" << GetNvar() << "];" << std::endl;
   fout << "   double NormVariable( double x, double xmin, double xmax ) const {" << std::endl;
   fout << "      // normalise to output range: [-1, 1]" << std::endl;
   fout << "      return 2*(x - xmin)/(xmax - xmin) - 1.0;" << std::endl;
   fout << "   }" << std::endl;
   fout << std::endl;
   fout << "   // type of input variable: 'F' or 'I'" << std::endl;
   fout << "   char   fType[" << GetNvar() << "];" << std::endl;
   fout << std::endl;
   fout << "   // initialize internal variables" << std::endl;
   fout << "   void Initialize();" << std::endl;
   if(GetAnalysisType() == Types::kMulticlass) {
      fout << "   std::vector<double> GetMulticlassValues__( const std::vector<double>& inputValues ) const;" << std::endl;
   } else {
      fout << "   double GetMvaValue__( const std::vector<double>& inputValues ) const;" << std::endl;
   }
   fout << "" << std::endl;
   fout << "   // private members (method specific)" << std::endl;

   // call the classifier specific output (the classifier must close the class !)
   MakeClassSpecific( fout, className );

   if(GetAnalysisType() == Types::kMulticlass) {
      fout << "inline std::vector<double> " << className <<  "::GetMulticlassValues( const std::vector<double>& inputValues ) const" << std::endl;
   } else {
      fout << "inline double " << className << "::GetMvaValue( const std::vector<double>& inputValues ) const" << std::endl;
   }
   fout << "{" << std::endl;
   fout << "   // classifier response value" << std::endl;
   if(GetAnalysisType() == Types::kMulticlass) {
      fout << "   std::vector<double> retval;" << std::endl;
   } else {
      fout << "   double retval = 0;" << std::endl;
   }
   fout << std::endl;
   fout << "   // classifier response, sanity check first" << std::endl;
   fout << "   if (!IsStatusClean()) {" << std::endl;
   fout << "      std::cout << \"Problem in class \\\"\" << fClassName << \"\\\": cannot return classifier response\"" << std::endl;
   fout << "                << \" because status is dirty\" << std::endl;" << std::endl;
   fout << "   }" << std::endl;
   fout << "   else {" << std::endl;
   if (IsNormalised()) {
      fout << "         // normalise variables" << std::endl;
      fout << "         std::vector<double> iV;" << std::endl;
      fout << "         iV.reserve(inputValues.size());" << std::endl;
      fout << "         int ivar = 0;" << std::endl;
      fout << "         for (std::vector<double>::const_iterator varIt = inputValues.begin();" << std::endl;
      fout << "              varIt != inputValues.end(); varIt++, ivar++) {" << std::endl;
      fout << "            iV.push_back(NormVariable( *varIt, fVmin[ivar], fVmax[ivar] ));" << std::endl;
      fout << "         }" << std::endl;
      if (GetTransformationHandler().GetTransformationList().GetSize() != 0 && GetMethodType() != Types::kLikelihood &&
          GetMethodType() != Types::kHMatrix) {
         fout << "         Transform( iV, -1 );" << std::endl;
      }

      if(GetAnalysisType() == Types::kMulticlass) {
         fout << "         retval = GetMulticlassValues__( iV );" << std::endl;
      } else {
         fout << "         retval = GetMvaValue__( iV );" << std::endl;
      }
   } else {
      if (GetTransformationHandler().GetTransformationList().GetSize() != 0 && GetMethodType() != Types::kLikelihood &&
          GetMethodType() != Types::kHMatrix) {
         fout << "         std::vector<double> iV(inputValues);" << std::endl;
         fout << "         Transform( iV, -1 );" << std::endl;
         if(GetAnalysisType() == Types::kMulticlass) {
            fout << "         retval = GetMulticlassValues__( iV );" << std::endl;
         } else {
            fout << "         retval = GetMvaValue__( iV );" << std::endl;
         }
      } else {
         if(GetAnalysisType() == Types::kMulticlass) {
            fout << "         retval = GetMulticlassValues__( inputValues );" << std::endl;
         } else {
            fout << "         retval = GetMvaValue__( inputValues );" << std::endl;
         }
      }
   }
   fout << "   }" << std::endl;
   fout << std::endl;
   fout << "   return retval;" << std::endl;
   fout << "}" << std::endl;

   // create output for transformation - if any
   if (GetTransformationHandler().GetTransformationList().GetSize()!=0)
      GetTransformationHandler().MakeFunction(fout, className,2);

   // close the file
   fout.close();
}

////////////////////////////////////////////////////////////////////////////////
/// prints out method-specific help method

void TMVA::MethodBase::PrintHelpMessage() const
{
   // if options are written to reference file, also append help info
   std::streambuf* cout_sbuf = std::cout.rdbuf(); // save original sbuf
   std::ofstream* o = 0;
   if (gConfig().WriteOptionsReference()) {
      Log() << kINFO << "Print Help message for class " << GetName() << " into file: " << GetReferenceFile() << Endl;
      o = new std::ofstream( GetReferenceFile(), std::ios::app );
      if (!o->good()) { // file could not be opened --> Error
         Log() << kFATAL << "<PrintHelpMessage> Unable to append to output file: " << GetReferenceFile() << Endl;
      }
      std::cout.rdbuf( o->rdbuf() ); // redirect 'std::cout' to file
   }

   //         "|--------------------------------------------------------------|"
   if (!o) {
      Log() << kINFO << Endl;
      Log() << gTools().Color("bold")
            << "================================================================"
            << gTools().Color( "reset" )
            << Endl;
      Log() << gTools().Color("bold")
            << "H e l p   f o r   M V A   m e t h o d   [ " << GetName() << " ] :"
            << gTools().Color( "reset" )
            << Endl;
   }
   else {
      Log() << "Help for MVA method [ " << GetName() << " ] :" << Endl;
   }

   // print method-specific help message
   GetHelpMessage();

   if (!o) {
      Log() << Endl;
      Log() << "<Suppress this message by specifying \"!H\" in the booking option>" << Endl;
      Log() << gTools().Color("bold")
            << "================================================================"
            << gTools().Color( "reset" )
            << Endl;
      Log() << Endl;
   }
   else {
      // indicate END
      Log() << "# End of Message___" << Endl;
   }

   std::cout.rdbuf( cout_sbuf ); // restore the original stream buffer
   if (o) o->close();
}

// ----------------------- r o o t   f i n d i n g ----------------------------

////////////////////////////////////////////////////////////////////////////////
/// returns efficiency as function of cut

Double_t TMVA::MethodBase::GetValueForRoot( Double_t theCut )
{
   Double_t retval=0;

   // retrieve the class object
   if (Use_Splines_for_Eff_) {
      retval = fSplRefS->Eval( theCut );
   }
   else retval = fEffS->GetBinContent( fEffS->FindBin( theCut ) );

   // caution: here we take some "forbidden" action to hide a problem:
   // in some cases, in particular for likelihood, the binned efficiency distributions
   // do not equal 1, at xmin, and 0 at xmax; of course, in principle we have the
   // unbinned information available in the trees, but the unbinned minimization is
   // too slow, and we don't need to do a precision measurement here. Hence, we force
   // this property.
   Double_t eps = 1.0e-5;
   if      (theCut-fXmin < eps) retval = (GetCutOrientation() == kPositive) ? 1.0 : 0.0;
   else if (fXmax-theCut < eps) retval = (GetCutOrientation() == kPositive) ? 0.0 : 1.0;

   return retval;
}

////////////////////////////////////////////////////////////////////////////////
/// returns the event collection (i.e. the dataset) TRANSFORMED using the
/// classifiers specific Variable Transformation (e.g. Decorr or Decorr:Gauss:Decorr)

const std::vector<TMVA::Event*>& TMVA::MethodBase::GetEventCollection( Types::ETreeType type)
{
   // if there's no variable transformation for this classifier, just hand back the
   //  event collection of the data set
   if (GetTransformationHandler().GetTransformationList().GetEntries() <= 0) {
      return (Data()->GetEventCollection(type));
   }

   // otherwise, transform ALL the events and hand back the vector of the pointers to the
   // transformed events. If the pointer is already != 0, i.e. the whole thing has been
   // done before, I don't need to do it again, but just "hand over" the pointer to those events.
   Int_t idx = Data()->TreeIndex(type);  //index indicating Training,Testing,...  events/datasets
   if (fEventCollections.at(idx) == 0) {
      fEventCollections.at(idx) = &(Data()->GetEventCollection(type));
      fEventCollections.at(idx) = GetTransformationHandler().CalcTransformations(*(fEventCollections.at(idx)),kTRUE);
   }
   return *(fEventCollections.at(idx));
}

////////////////////////////////////////////////////////////////////////////////
/// calculates the TMVA version string from the training version code on the fly

TString TMVA::MethodBase::GetTrainingTMVAVersionString() const
{
   UInt_t a = GetTrainingTMVAVersionCode() & 0xff0000; a>>=16;
   UInt_t b = GetTrainingTMVAVersionCode() & 0x00ff00; b>>=8;
   UInt_t c = GetTrainingTMVAVersionCode() & 0x0000ff;

   return TString(Form("%i.%i.%i",a,b,c));
}

////////////////////////////////////////////////////////////////////////////////
/// calculates the ROOT version string from the training version code on the fly

TString TMVA::MethodBase::GetTrainingROOTVersionString() const
{
   UInt_t a = GetTrainingROOTVersionCode() & 0xff0000; a>>=16;
   UInt_t b = GetTrainingROOTVersionCode() & 0x00ff00; b>>=8;
   UInt_t c = GetTrainingROOTVersionCode() & 0x0000ff;

   return TString(Form("%i.%02i/%02i",a,b,c));
}

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

Double_t TMVA::MethodBase::GetKSTrainingVsTest(Char_t SorB, TString opt){
   ResultsClassification* mvaRes = dynamic_cast<ResultsClassification*>
      ( Data()->GetResults(GetMethodName(),Types::kTesting, Types::kClassification) );

   if (mvaRes != NULL) {
      TH1D *mva_s = dynamic_cast<TH1D*> (mvaRes->GetHist("MVA_S"));
      TH1D *mva_b = dynamic_cast<TH1D*> (mvaRes->GetHist("MVA_B"));
      TH1D *mva_s_tr = dynamic_cast<TH1D*> (mvaRes->GetHist("MVA_TRAIN_S"));
      TH1D *mva_b_tr = dynamic_cast<TH1D*> (mvaRes->GetHist("MVA_TRAIN_B"));

      if ( !mva_s || !mva_b || !mva_s_tr || !mva_b_tr) return -1;

      if (SorB == 's' || SorB == 'S')
         return mva_s->KolmogorovTest( mva_s_tr, opt.Data() );
      else
         return mva_b->KolmogorovTest( mva_b_tr, opt.Data() );
   }
   return -1;
}