root/TMVA/MethodFDA.h

0001 // @(#)root/tmva $Id$
0002 // Author: Andreas Hoecker, Peter Speckmayer
0003
0004 /**********************************************************************************
0005  * Project: TMVA - a Root-integrated toolkit for multivariate data analysis       *
0006  * Package: TMVA                                                                  *
0007  * Class  : MethodFDA                                                             *
0008  *                                             *
0009  *                                                                                *
0010  * Description:                                                                   *
0011  *      Function discriminant analysis (FDA). This simple classifier              *
0012  *      fits any user-defined TFormula (via option configuration string) to       *
0013  *      the training data by requiring a formula response of 1 (0) to signal      *
0014  *      (background) events. The parameter fitting is done via the abstract       *
0015  *      class FitterBase, featuring Monte Carlo sampling, Genetic                 *
0016  *      Algorithm, Simulated Annealing, MINUIT and combinations of these.         *
0017  *                                                                                *
0018  * Authors (alphabetical):                                                        *
0019  *      Andreas Hoecker  <Andreas.Hocker@cern.ch> - CERN, Switzerland             *
0020  *      Peter Speckmayer <speckmay@mail.cern.ch>  - CERN, Switzerland             *
0021  *                                                                                *
0022  * Copyright (c) 2005-2010:                                                       *
0023  *      CERN, Switzerland                                                         *
0024  *      MPI-K Heidelberg, Germany                                                 *
0025  *                                                                                *
0026  * Redistribution and use in source and binary forms, with or without             *
0027  * modification, are permitted according to the terms listed in LICENSE           *
0028  * (see tmva/doc/LICENSE)                                          *
0029  **********************************************************************************/
0030
0031 #ifndef ROOT_TMVA_MethodFDA
0032 #define ROOT_TMVA_MethodFDA
0033
0034 //////////////////////////////////////////////////////////////////////////
0035 //                                                                      //
0036 // MethodFDA                                                            //
0037 //                                                                      //
0038 // Function discriminant analysis (FDA). This simple classifier         //
0039 // fits any user-defined TFormula (via option configuration string) to  //
0040 // the training data by requiring a formula response of 1 (0) to signal //
0041 // (background) events. The parameter fitting is done via the abstract  //
0042 // class FitterBase, featuring Monte Carlo sampling, Genetic            //
0043 // Algorithm, Simulated Annealing, MINUIT and combinations of these.    //
0044 //                                                                      //
0045 // Can compute one-dimensional regression                               //
0046 //                                                                      //
0047 //////////////////////////////////////////////////////////////////////////
0048
0049 #include "TMVA/MethodBase.h"
0050 #include "TMVA/IFitterTarget.h"
0051 #include <vector>
0052
0053 class TFormula;
0054
0055 namespace TMVA {
0056
0057    class Interval;
0058    class Event;
0059    class FitterBase;
0060
0061    class MethodFDA : public MethodBase, public IFitterTarget {
0062
0063    public:
0064
0065       MethodFDA( const TString& jobName,
0066                  const TString& methodTitle,
0067                  DataSetInfo& theData,
0068                  const TString& theOption = "");
0069
0070       MethodFDA( DataSetInfo& theData,
0071                  const TString& theWeightFile);
0072
0073       virtual ~MethodFDA( void );
0074
0075       Bool_t HasAnalysisType( Types::EAnalysisType type, UInt_t numberClasses, UInt_t numberTargets ) override;
0076
0077       // training method
0078       void Train( void ) override;
0079
0080       using MethodBase::ReadWeightsFromStream;
0081
0082       void AddWeightsXMLTo      ( void* parent     ) const override;
0083
0084       void ReadWeightsFromStream( std::istream & i ) override;
0085       void ReadWeightsFromXML   ( void* wghtnode ) override;
0086
0087       // calculate the MVA value
0088       Double_t GetMvaValue( Double_t* err = nullptr, Double_t* errUpper = nullptr ) override;
0089
0090       const std::vector<Float_t>& GetRegressionValues() override;
0091       const std::vector<Float_t>& GetMulticlassValues() override;
0092
0093       void Init( void ) override;
0094
0095       // ranking of input variables
0096       const Ranking* CreateRanking() override { return nullptr; }
0097
0098       Double_t EstimatorFunction( std::vector<Double_t>& ) override;
0099
0100       // no check of options at this place
0101       void CheckSetup() override {}
0102
0103    protected:
0104
0105       // make ROOT-independent C++ class for classifier response (classifier-specific implementation)
0106       void MakeClassSpecific( std::ostream&, const TString& ) const override;
0107
0108       // get help message text
0109       void GetHelpMessage() const override;
0110
0111    private:
0112
0113       // compute multiclass values
0114       void CalculateMulticlassValues( const TMVA::Event*& evt, std::vector<Double_t>& parameters, std::vector<Float_t>& values);
0115
0116
0117       // create and interpret formula expression and compute estimator
0118       void     CreateFormula   ();
0119       Double_t InterpretFormula( const Event*, std::vector<Double_t>::iterator begin, std::vector<Double_t>::iterator end );
0120
0121       // clean up
0122       void ClearAll();
0123
0124       // print fit results
0125       void PrintResults( const TString&, std::vector<Double_t>&, const Double_t ) const;
0126
0127       // the option handling methods
0128       void DeclareOptions() override;
0129       void ProcessOptions() override;
0130
0131       TString                fFormulaStringP;     ///< string with function
0132       TString                fParRangeStringP;    ///< string with ranges of parameters
0133       TString                fFormulaStringT;     ///< string with function
0134       TString                fParRangeStringT;    ///< string with ranges of parameters
0135
0136       TFormula*              fFormula;            ///< the discrimination function
0137       UInt_t                 fNPars;              ///< number of parameters
0138       std::vector<Interval*> fParRange;           ///< ranges of parameters
0139       std::vector<Double_t>  fBestPars;           ///< the pars that optimise (minimise) the estimator
0140       TString                fFitMethod;          ///< estimator optimisation method
0141       TString                fConverger;          ///< fit method uses fConverger as intermediate step to converge into local minimas
0142       FitterBase*            fFitter;             ///< the fitter used in the training
0143       IFitterTarget*         fConvergerFitter;    ///< intermediate fitter
0144
0145
0146       // sum of weights (this should become centrally available through the dataset)
0147       Double_t               fSumOfWeightsSig;    ///< sum of weights (signal)
0148       Double_t               fSumOfWeightsBkg;    ///< sum of weights (background)
0149       Double_t               fSumOfWeights;       ///< sum of weights
0150
0151       //
0152       Int_t                  fOutputDimensions;   ///< number of output values
0153
0154       ClassDefOverride(MethodFDA,0);  // Function Discriminant Analysis
0155    };
0156
0157 } // namespace TMVA
0158
0159 #endif // MethodFDA_H