#include "trainNetwork.C"

void hbbNN(Int_t ntrain=15) {

// Example of a Multi Layer Perceptron

// For a Tevatron search for SUSY Higgs in bh->bbb, a neural network

// was used to separate the signal from the background passing

// some selection cuts. Here is a simplified version of this network,

// taking into account only bbb events.

//Author: P. Jonsson

//Part 1, preparing the inputs

//Booking some histograms for signal and background,

//Add histograms yourself for the dijet mass (MH[0]) of the leading pT

//(=transverse momenta) jets for background and signal

TH1F *detas = new TH1F("detas", "detas", 50, .0, 4.);

TH1F *detab = new TH1F("detab", "detab", 50, .0, 4.);

TH1F *dphis = new TH1F("dphis", "dphis", 50, .0, 3.15);

TH1F *dphib = new TH1F("dphib", "dphib", 50, .0, 3.15);

TH1F *angles = new TH1F("angles", "angles", 50, .0, 2.);

TH1F *angleb = new TH1F("angleb", "angleb", 50, .0, 2.);

TH1F *pbalances = new TH1F("pbalances", "pbalances", 50, .0, 1.);

TH1F *pbalanceb = new TH1F("pbalanceb", "pbalanceb", 50, .0, 1.);

TH1F *etahs = new TH1F("etahs", "etahs", 50, -2.5, 2.5);

TH1F *etahb = new TH1F("etahb", "etahb", 50, -2.5, 2.5);

TH1F *sphericitys = new TH1F("sphericitys", "sphericitys", 50, .0, 1.);

TH1F *sphericityb = new TH1F("sphericityb", "sphericityb", 50, .0, 1.);

if (!gROOT->GetClass("TMultiLayerPerceptron")) {

gSystem->Load("libMLP");

}

// Prepare inputs

// The 2 trees are merged into one, and a "type" branch,

// equal to 1 for the signal and 0 for the background is added.

TChain * signal = new TChain("higgsSearch","");

//Use a Higgs signal at a mass of 110 GeV

signal->Add("LHinput-cut_2b25_1j15-Tight-MC_higgs110GeV.root/higgsSearch");

//Use a background of multijet bbb events

TChain *background = new TChain("higgsSearch","");

background->Add("LHinput-cut_2b25_1j15-Tight-MC_bbb.root/higgsSearch");

//book combined tree of selected events top use for training

TFile *f = new TFile("training.root","recreate");

TTree *simu = new TTree("MonteCarlo", "Filtered Monte Carlo Events");

Double_t nnoutxlm[4], MH[6], dEta[6], dPhi[6], Angle[6], EtaH[6], pBalance[6], Sphericity;

Double_t dphi, deta, angle, etah, pbalance, sphericity;

Int_t Njets, typeH[6], typeO, nsignal, nbkg;

//counters for signal and background events

nsignal=0;

nbkg=0;

//

//Set branch addresses to use from signal and background trees

//Description of variables inline below.

//Draw histograms of these variables for signal and background on top

//of eachother in different colors.

//Save these plots (you can look up how to save plots to files at root.cern.ch)

//Invariant di-jet mass

signal->SetBranchAddress("MH", &MH);

//Delta eta, difference in pseudorapidity (spatial coodinate used to

//describe angle with the beam axis = -ln(tan(theta/2))

signal->SetBranchAddress("dEta", &dEta);

//Delta phi, azimuthal angle (going from 0 to 2pi around the beampipe)

signal->SetBranchAddress("dPhi", &dPhi);

//Angle between the leading jet and the combined jet-pair

signal->SetBranchAddress("Eta1mEtaH", &Angle);

//Pseudorapidity of the combined jet-pair

signal->SetBranchAddress("EtaH", &EtaH);

//Which jet-pair comes from the true Higgs decay

signal->SetBranchAddress("type", &typeH);

//Momentum balance of the jet-pair: p1-p2/p1+p2

signal->SetBranchAddress("pBalance", &pBalance);

//Number of jets in the event

signal->SetBranchAddress("Njets", &Njets);

//Event sphericity (measure of how spherical, round, the event is)

signal->SetBranchAddress("FW_H2", &Sphericity);

//same for background

background->SetBranchAddress("MH", &MH);

background->SetBranchAddress("dEta", &dEta);

background->SetBranchAddress("dPhi", &dPhi);

background->SetBranchAddress("Eta1mEtaH", &Angle);

background->SetBranchAddress("EtaH", &EtaH);

background->SetBranchAddress("pBalance", &pBalance);

background->SetBranchAddress("Njets", &Njets);

background->SetBranchAddress("FW_H2", &Sphericity);

//Make input branches of variables to use for the training

simu->Branch("dEta", &deta, "dEta/D");

simu->Branch("dPhi", &dphi, "dPhi/D");

simu->Branch("Angle", &angle, "Angle/D");

simu->Branch("pBalance", &pbalance, "pBalance/D");

simu->Branch("EtaH", &etah, "EtaH/D");

simu->Branch("Sphericity", &sphericity, "Sphericity/D");

simu->Branch("type", &typeO, "type/I");

//loop over signal and select events to use for training

Int_t i;

for (i = 0; i < signal->GetEntries(); i++) {

signal->GetEntry(i);

//only consider events with 3 jets

if (Njets==3){

typeO=0;

//select only events where the two leading pT jets come from the

//Higgs decay (typeH[0]==1), and these two jets are separated by

//sqrt(dEta[0]*dEta[0]+dPhi[0]*dPhi[0]) > 1, and the invariant

//di-jet mass is greater than 50 GeV

if (typeH[0]==1 && sqrt(dEta[0]*dEta[0]+dPhi[0]*dPhi[0]) > 1. && MH[0]>50.){

deta = dEta[0];

dphi =2*acos(0)- dPhi[0];

angle = Angle[0];

pbalance = pBalance[0];

etah = EtaH[0];

sphericity = Sphericity;

typeO=1;

}

//fill selected signal events (type0==1)

if (typeO==1){

//fill the histograms like this:

detas->Fill(deta);

dphis->Fill(dphi);

angles->Fill(angle);

pbalances->Fill(pbalance);

etahs->Fill(etah);

sphericitys->Fill(sphericity);

//fill the training tree

simu->Fill();

nsignal++;

}

}

}

cout << nsignal <<endl;

//now loop over background events

for (i = 0; i < background->GetEntries(); i++) {

background->GetEntry(i);

if (Njets==3){

typeO=1;

//similar for background events

if (sqrt(dEta[0]*dEta[0]+dPhi[0]*dPhi[0]) > 1. && MH[0]>50.){

deta = dEta[0];

dphi =2*acos(0)- dPhi[0];

angle = Angle[0];

pbalance = pBalance[0];

etah = EtaH[0];

sphericity = Sphericity;

typeO=0;

}

//fill the selected background events until you have the same

//number as signal

if (typeO==0 && nbkg<nsignal){

//fill the background histograms here:

detab->Fill(deta);

dphib->Fill(dphi);

angleb->Fill(angle);

pbalanceb->Fill(pbalance);

etahb->Fill(etah);

sphericityb->Fill(sphericity);

nbkg++;

simu->Fill();

}

}

}

cout << nbkg << endl;

/*Plot the simulated and background variable distributions */

TCanvas *c1 = new TCanvas("c2", "Signal and Background Distributions");

c1->Divide(2,3);

c1->cd(1); dphis->Draw(); dphib-> SetLineColor(kRed); dphib->Draw("same");

c1->cd(2); detas->Draw(); detab-> SetLineColor(kRed); detab->Draw("same");

c1->cd(3); angles->Draw(); angleb-> SetLineColor(kRed); angleb->Draw("same");

c1->cd(4); pbalances ->Draw(); pbalanceb-> SetLineColor(kRed); pbalanceb ->Draw("same");

c1->cd(5); etahs->Draw(); etahb-> SetLineColor(kRed); etahb->Draw("same");

c1->cd(6); sphericitys->Draw(); sphericityb-> SetLineColor(kRed); sphericityb->Draw("same");

c1->SaveAs("sig_background_distr.pdf");

for(Int_t i = 1; i++; i < 5)

{

stringstream ss;

ss << i*5;

TString structure = "@dEta, @dPhi, @EtaH, @pBalance, @Sphericity, @Angle:" + ss.str() + ":1:type";;

trainNetwork(simu, structure, 100);

}

}