/******************************************************************************
*  $Id: CbmRichRingSelectNeuralNet.cxx,v 1.1 2006/09/13 14:53:31 hoehne Exp $
*
*  Class  : CbmRichRingSelect2DCuts
*  Description : Implementation for concrete RICH ring selection algorithm:
*                reject rings using a trained neural net (input file with weights needed!)
*                store resulting value (0-1) in "SelectionNN":
*                0 = good rings
*                1 = rings to be rejected
*                --> choose a value in between depending on required purity/ efficiency
*
*  Author : Simeon Lebedev
*  E-mail : salebedev@jinr.ru
*
*******************************************************************************
*  $Log: CbmRichRingSelectNeuralNet.cxx,v $
*  Revision 1.1  2006/09/13 14:53:31  hoehne
*  initial version
*
*  
*
*******************************************************************************/


#include "CbmRichRingSelectNeuralNet.h"
#include "CbmRichHit.h"
#include "CbmRichRing.h"
#include "TMultiLayerPerceptron.h"
#include "TTree.h"
using namespace std;

//--------------------------------------------------------------------------
// -----   Default constructor   -------------------------------------------
CbmRichRingSelectNeuralNet::CbmRichRingSelectNeuralNet()
{
	fVerbose = 1;
}

// -----   Standard constructor   ------------------------------------------
CbmRichRingSelectNeuralNet::CbmRichRingSelectNeuralNet ( Int_t verbose, const char* NNFile )
{
	fVerbose = verbose;
	fNeuralNetWeights = NNFile;
}

// -----   Destructor   ----------------------------------------------------
CbmRichRingSelectNeuralNet::~CbmRichRingSelectNeuralNet()
{}

// -----   Initialization   ----------------------------------------------------
void CbmRichRingSelectNeuralNet::Init ()
{
	CbmRichRingSelect::Init();
	
    TTree *simu = new TTree ("MonteCarlo","MontecarloData");
	Float_t x1,x2,x3,x4,x5,x6,x7,x8;

	simu->Branch("x1", &x1,"x1/F");
	simu->Branch("x2", &x2,"x2/F");
	simu->Branch("x3", &x3,"x3/F");
	simu->Branch("x4", &x4,"x4/F");
	simu->Branch("x5", &x5,"x5/F");
	simu->Branch("x6", &x6,"x6/F");
	simu->Branch("x7", &x7,"x7/F");
	simu->Branch("x8", &x8,"x8/F");

    fNN = new TMultiLayerPerceptron("x1,x2,x3,x4,x5,x6,x7:30:x8",simu);

    //char* NeuralNetWeights = "NeuralNet_RingSelection_Weights.txt";
    cout << "-I- CbmRichRingSelectNeuralNet: get NeuralNet weight parameters from: " << fNeuralNetWeights << endl;
    fNN->LoadWeights(fNeuralNetWeights);
}

// -----   Exec   ----------------------------------------------------
void CbmRichRingSelectNeuralNet::DoSelect(CbmRichRing* ring)
{
    if (fVerbose > 1) ring->Print();
	
	ring->SetAngle(GetAngle(ring));
	ring->SetNofHitsOnRing(GetNofHitsOnRing(ring));

    Double_t nnPar[7];
    nnPar[0] =  ring->GetNofHits();	
    nnPar[1] =  ring->GetDistance();	
    nnPar[2] =  ring->GetAngle();
    nnPar[3] =  ring->GetNofHitsOnRing();
    nnPar[4] =  ring->GetRadialPosition();
    nnPar[5] =  ring->GetChi2();
    nnPar[6] =  ring->GetRadius();
	
    Double_t nnEval = fNN->Evaluate(0,nnPar);
    if (fVerbose > 1) cout <<"nnEval = "<<nnEval<<endl;

    if (TMath::IsNaN(nnEval) == 1) {
		cout << " -W- RingSelectNN: nnEval nan " << endl;
		nnEval = -1.;
    }

    ring->SetSelectionNN(nnEval);


}

// -------------------------------------------------------------------------
ClassImp(CbmRichRingSelectNeuralNet)