MethodPyTorch.h

Go to the documentation of this file.

// @(#)root/tmva/pymva $Id$
// Author: Anirudh Dagar, 2020
 
/**********************************************************************************
 * Project: TMVA - a Root-integrated toolkit for multivariate data analysis       *
 * Package: TMVA                                                                  *
 * Class  : MethodPyTorch                                                         *
 *                                             *
 *                                                                                *
 * Description:                                                                   *
 *      Interface for PyTorch python based scientific package supporting          *
 *      automatic differentiation for machine learning.                           *
 *                                                                                *
 * Authors (alphabetical):                                                        *
 *      Anirudh Dagar <anirudhdagar6@gmail.com> - IIT, Roorkee                    *
 *                                                                                *
 * Copyright (c) 2020:                                                            *
 *      CERN, Switzerland                                                         *
 *      IIT, Roorkee                                                              *
 *                                                                                *
 * Redistribution and use in source and binary forms, with or without             *
 * modification, are permitted according to the terms listed in LICENSE           *
 * (see tmva/doc/LICENSE)                                          *
 **********************************************************************************/
 
#ifndef ROOT_TMVA_MethodPyTorch
#define ROOT_TMVA_MethodPyTorch
 
#include "TMVA/PyMethodBase.h"
#include <vector>
 
namespace TMVA {
 
   class MethodPyTorch : public PyMethodBase {
 
   public :
 
      // constructors
      MethodPyTorch(const TString &jobName,
            const TString &methodTitle,
            DataSetInfo &dsi,
            const TString &theOption = "");
      MethodPyTorch(DataSetInfo &dsi,
            const TString &theWeightFile);
      ~MethodPyTorch();
 
      void Train() override;
      void Init() override;
      void DeclareOptions() override;
      void ProcessOptions() override;
 
      // Check whether the given analysis type (regression, classification, ...)
      // is supported by this method
      Bool_t HasAnalysisType(Types::EAnalysisType type, UInt_t numberClasses, UInt_t) override;
      // Get signal probability of given event
      Double_t GetMvaValue(Double_t *errLower, Double_t *errUpper) override;
      std::vector<Double_t> GetMvaValues(Long64_t firstEvt, Long64_t lastEvt, Bool_t logProgress) override;
      // Get regression values of given event
      std::vector<Float_t>& GetRegressionValues() override;
      // Get all regression values for all the events in the data set
      std::vector<Float_t> GetAllRegressionValues() override;
      // Get class probabilities of given event
      std::vector<Float_t>& GetMulticlassValues() override;
      // Get all multiclass values for all the events in the data set
      std::vector<Float_t> GetAllMulticlassValues() override;
 
      const Ranking *CreateRanking() override { return nullptr; }
      void TestClassification() override;
      void AddWeightsXMLTo(void*) const override{}
      void ReadWeightsFromXML(void*) override{}
      void ReadWeightsFromStream(std::istream&) override {} // backward compatibility
      void ReadWeightsFromStream(TFile&) override{} // backward compatibility
      void ReadModelFromFile() override;
 
      void GetHelpMessage() const override;
 
 
    private:
 
      TString fFilenameModel;                         // Filename of the previously exported PyTorch model
      UInt_t fBatchSize {0};                          // Training batch size
      UInt_t fNumEpochs {0};                          // Number of training epochs
      Int_t fNumThreads {0};                          // Number of CPU threads (if 0 uses default values)
 
      Bool_t fContinueTraining;                       // Load weights from previous training
      Bool_t fSaveBestOnly;                           // Store only weights with smallest validation loss
      TString fLearningRateSchedule;                  // Set new learning rate at specific epochs
 
      TString fNumValidationString;                   // option string defining the number of validation events
 
      TString fUserCodeName;                          // filename of the user script that will be executed before loading the PyTorch model
 
      bool fModelIsSetup = false;                     // flag whether model is loaded, needed for getMvaValue during evaluation
      std::vector<float> fVals;                       // input variables array
      std::vector<float> fOutput;                     // model output array
      UInt_t fNVars {0};                              // number of variables
      UInt_t fNOutputs {0};                           // number of outputs (classes or targets)
      TString fFilenameTrainedModel;                  // output filename for trained model
      PyObject * fPyVals = nullptr;                             // Python array object for input data
      PyObject * fPyOutput = nullptr;                           // Python array object for output data
 
      void SetupPyTorchModel(Bool_t loadTrainedModel);  // setups the needed variables, loads the model
      void InitEvaluation(size_t nEvents);              // allocate arrays for evaluation
      UInt_t  GetNumValidationSamples();                // get number of validation events according to given option
 
      ClassDefOverride(MethodPyTorch, 0);
   };
 
} // namespace TMVA
 
#endif // ROOT_TMVA_MethodPyTorch