| 1 |
#include <cstdlib>
|
| 2 |
#include <vector>
|
| 3 |
#include <iostream>
|
| 4 |
#include <map>
|
| 5 |
#include <string>
|
| 6 |
|
| 7 |
#include "TFile.h"
|
| 8 |
#include "TH1.h"
|
| 9 |
#include "TTree.h"
|
| 10 |
#include "TString.h"
|
| 11 |
#include "TSystem.h"
|
| 12 |
#include "TROOT.h"
|
| 13 |
#include "TStopwatch.h"
|
| 14 |
#include "TChain.h"
|
| 15 |
#include "TMath.h"
|
| 16 |
#include "TChainElement.h"
|
| 17 |
#include "TCanvas.h"
|
| 18 |
#include "TCut.h"
|
| 19 |
#include "TStyle.h"
|
| 20 |
#include "TLegend.h"
|
| 21 |
#include "RooGaussian.h"
|
| 22 |
#include "RooNumConvPdf.h"
|
| 23 |
#include "RooGenericPdf.h"
|
| 24 |
#include "RooRealVar.h"
|
| 25 |
#include "RooPlot.h"
|
| 26 |
#include "RooDataHist.h"
|
| 27 |
#include "RooFitResult.h"
|
| 28 |
#include "RooHistPdf.h"
|
| 29 |
#include "RooAddPdf.h"
|
| 30 |
#include "RooDataSet.h"
|
| 31 |
|
| 32 |
using namespace std;
|
| 33 |
using namespace RooFit;
|
| 34 |
|
| 35 |
|
| 36 |
void fitMVA(){
|
| 37 |
|
| 38 |
//-------------------------------
|
| 39 |
// user parameters
|
| 40 |
//-------------------------------
|
| 41 |
|
| 42 |
const int nTrials = 1000; // number of trials in fit validation
|
| 43 |
const bool display = false; // display fit? Set to false for large nTrials
|
| 44 |
const bool printgif = true; // print canvases to gif files?
|
| 45 |
|
| 46 |
//---------------------------------------
|
| 47 |
// name of MVA branch in ntuple
|
| 48 |
//---------------------------------------
|
| 49 |
|
| 50 |
//const char* mvaname = "nn_hww160_ww";
|
| 51 |
const char* mvaname = "bdt_hww160_ww";
|
| 52 |
|
| 53 |
//-----------------------------------------
|
| 54 |
// selection to apply to sig, bkg samples
|
| 55 |
//-----------------------------------------
|
| 56 |
|
| 57 |
TCut jet("njets == 0");
|
| 58 |
TCut mll100("dilep->mass() < 100");
|
| 59 |
TCut OS("lq1*lq2 < 0");
|
| 60 |
TCut pt2010("lep1->pt() >= 20 && lep2->pt() >= 10");
|
| 61 |
TCut pmet1("pmet >= 20 && (type == 1 || type == 2)");
|
| 62 |
TCut pmet2("pmet >= 35 && (type == 0 || type == 3)");
|
| 63 |
TCut pmet = pmet1 || pmet2;
|
| 64 |
TCut lid("lid3 == 0");
|
| 65 |
TCut lowptb("jetLowBtag <= 2.1");
|
| 66 |
TCut softmu("nSoftMuons == 0");
|
| 67 |
TCut sel160("lep1.pt()>30 && lep2.pt()>25 && dilep.mass()<50 && dPhi<1.05");
|
| 68 |
|
| 69 |
TCut sel = jet + mll100 + OS + pt2010 + pmet;
|
| 70 |
TCut weight = "0.5 * scale1fb";
|
| 71 |
TCut selweight = sel*weight;
|
| 72 |
|
| 73 |
//--------------------------------------------------
|
| 74 |
// get sig and bkg samples
|
| 75 |
//--------------------------------------------------
|
| 76 |
|
| 77 |
char* babyPath = "/smurf/benhoob/MVA/SmurfTraining/hww160_ww/output";
|
| 78 |
|
| 79 |
TChain* sig = new TChain("tree");
|
| 80 |
sig->Add(Form("%s/hww160.root",babyPath));
|
| 81 |
|
| 82 |
TChain* bkg = new TChain("tree");
|
| 83 |
bkg->Add(Form("%s/ww.root",babyPath));
|
| 84 |
|
| 85 |
//------------------------------------
|
| 86 |
// set binning
|
| 87 |
//------------------------------------
|
| 88 |
|
| 89 |
int nbins = 0;
|
| 90 |
float xmin = 0.;
|
| 91 |
float xmax = 0.;
|
| 92 |
|
| 93 |
if( strcmp( mvaname , "nn_hww160_ww" ) == 0 ){
|
| 94 |
nbins = 20;
|
| 95 |
xmin = -0.5;
|
| 96 |
xmax = 1.5;
|
| 97 |
}
|
| 98 |
else if( strcmp( mvaname , "bdt_hww160_ww" ) == 0 ){
|
| 99 |
nbins = 30;
|
| 100 |
xmin = -1.0;
|
| 101 |
xmax = 0.5;
|
| 102 |
}else{
|
| 103 |
cout << "Unrecognized MVA " << mvaname << ", quitting" << endl;
|
| 104 |
exit(0);
|
| 105 |
}
|
| 106 |
|
| 107 |
//------------------------
|
| 108 |
// declare variables
|
| 109 |
//------------------------
|
| 110 |
|
| 111 |
RooRealVar nsig ("nsig" , "Signal Yield" , 50 , 0 , 1000 );
|
| 112 |
RooRealVar nbkg ("nbkg" , "Background Yield" , 300 , 0 , 1000 );
|
| 113 |
RooRealVar mva ("mva" , "MVA Output" , xmin , xmax );
|
| 114 |
|
| 115 |
nsig.setVal(50);
|
| 116 |
nbkg.setVal(300);
|
| 117 |
mva.setBins(nbins);
|
| 118 |
|
| 119 |
//---------------------------------------------------
|
| 120 |
// get MVA output distributions from sig, bkg samp
|
| 121 |
//---------------------------------------------------
|
| 122 |
|
| 123 |
TH1F* mva_sig = new TH1F("mva_sig","MVA Output for Signal" , nbins , xmin , xmax );
|
| 124 |
TH1F* mva_bkg = new TH1F("mva_bkg","MVA Output for Background" , nbins , xmin , xmax );
|
| 125 |
|
| 126 |
mva_sig->Sumw2();
|
| 127 |
mva_bkg->Sumw2();
|
| 128 |
|
| 129 |
TCanvas *ctemp = new TCanvas();
|
| 130 |
ctemp->cd();
|
| 131 |
sig->Draw(Form("%s >> mva_sig",mvaname),selweight);
|
| 132 |
bkg->Draw(Form("%s >> mva_bkg",mvaname),selweight);
|
| 133 |
delete ctemp;
|
| 134 |
|
| 135 |
float nsigtrue = mva_sig->Integral();
|
| 136 |
float nbkgtrue = mva_bkg->Integral();
|
| 137 |
|
| 138 |
//---------------------------------------------------
|
| 139 |
// convert MVA distributions to RooHistPdf objects
|
| 140 |
//---------------------------------------------------
|
| 141 |
|
| 142 |
RooDataHist sigpdfhist("sigpdfhist","Signal PDF Hist" , RooArgSet(mva),mva_sig);
|
| 143 |
RooDataHist bkgpdfhist("bkgpdfhist","Background PDF Hist" , RooArgSet(mva),mva_bkg);
|
| 144 |
|
| 145 |
RooHistPdf sigpdf("sigpdf", "Signal PDF" , RooArgSet(mva), sigpdfhist);
|
| 146 |
RooHistPdf bkgpdf("bkgpdf", "Background PDF" , RooArgSet(mva), bkgpdfhist);
|
| 147 |
|
| 148 |
nsig.setVal( nsigtrue );
|
| 149 |
nbkg.setVal( nbkgtrue );
|
| 150 |
|
| 151 |
RooAddPdf datapdf("datapdf", "Data PDF", RooArgList(sigpdf,bkgpdf), RooArgList(nsig,nbkg));
|
| 152 |
|
| 153 |
//-------------------------------------------------
|
| 154 |
// perform fit (nTrials iterations)
|
| 155 |
//-------------------------------------------------
|
| 156 |
|
| 157 |
float nsigfit[nTrials];
|
| 158 |
float nbkgfit[nTrials];
|
| 159 |
float nsigerrfit[nTrials];
|
| 160 |
float nbkgerrfit[nTrials];
|
| 161 |
float significance[nTrials];
|
| 162 |
TCanvas *can[nTrials];
|
| 163 |
|
| 164 |
TH1F* hsig = new TH1F("hsig","Sig Yield",100,0,100);
|
| 165 |
TH1F* hbkg = new TH1F("hbkg","Bkg Yield",100,0,200);
|
| 166 |
TH1F* hsigpull = new TH1F("hsigpull","Sig Pull",100,-5,5);
|
| 167 |
TH1F* hbkgpull = new TH1F("hbkgpull","Bkg Pull",100,-5,5);
|
| 168 |
TH1F* hsignificance = new TH1F("hsignificance","Significance",100,0,10);
|
| 169 |
|
| 170 |
for( int i = 0 ; i < nTrials ; ++i ){
|
| 171 |
|
| 172 |
nsig.setVal( nsigtrue );
|
| 173 |
nbkg.setVal( nbkgtrue );
|
| 174 |
|
| 175 |
//-------------------------------------
|
| 176 |
// generate pseudo-dataset from PDF
|
| 177 |
//-------------------------------------
|
| 178 |
|
| 179 |
RooDataSet *gendata = datapdf.generate(RooArgList(mva),nsigtrue+nbkgtrue,Extended(kTRUE));
|
| 180 |
|
| 181 |
//-------------------------------------
|
| 182 |
// perform fit with nsig fixed to 0
|
| 183 |
//-------------------------------------
|
| 184 |
|
| 185 |
nsig.setVal(0);
|
| 186 |
nbkg.setVal(gendata->sumEntries());
|
| 187 |
nsig.setConstant();
|
| 188 |
|
| 189 |
RooAbsReal* mynll_bkg = bkgpdf.createNLL( *gendata );
|
| 190 |
RooAbsReal* mynll_tot = datapdf.createNLL( *gendata );
|
| 191 |
|
| 192 |
cout << "nll_bkg " << mynll_bkg->getVal() << endl;
|
| 193 |
cout << "nll_tot " << mynll_tot->getVal() << endl;
|
| 194 |
|
| 195 |
cout << endl;
|
| 196 |
cout << "-------------------------------------------" << endl;
|
| 197 |
cout << "Performing fit with signal yield fixed to 0" << endl;
|
| 198 |
cout << "-------------------------------------------" << endl;
|
| 199 |
cout << endl;
|
| 200 |
|
| 201 |
RooFitResult *bkg_result = datapdf.fitTo( *gendata , Save() , Extended(kTRUE) );
|
| 202 |
|
| 203 |
/*
|
| 204 |
TCanvas *bkgcan = new TCanvas("bkgcan","bkgcan",600,600);
|
| 205 |
bkgcan->cd();
|
| 206 |
|
| 207 |
RooPlot* bkgframe = mva.frame();
|
| 208 |
bkgframe->SetXTitle(mvaname);
|
| 209 |
gendata->plotOn(bkgframe);
|
| 210 |
datapdf.plotOn(bkgframe,Components(sigpdf));
|
| 211 |
datapdf.plotOn(bkgframe,Components(bkgpdf),LineColor(kRed));
|
| 212 |
datapdf.plotOn(bkgframe,LineColor(kOrange));
|
| 213 |
bkgframe->Draw();
|
| 214 |
*/
|
| 215 |
|
| 216 |
//-------------------------------------
|
| 217 |
// perform fit with nsig floated in fit
|
| 218 |
//-------------------------------------
|
| 219 |
|
| 220 |
//nsig.setVal(50);
|
| 221 |
nsig.setConstant(kFALSE);
|
| 222 |
|
| 223 |
cout << endl;
|
| 224 |
cout << "----------------------------------------" << endl;
|
| 225 |
cout << "Performing fit with signal yield floated" << endl;
|
| 226 |
cout << "----------------------------------------" << endl;
|
| 227 |
cout << endl;
|
| 228 |
|
| 229 |
RooFitResult *result = datapdf.fitTo( *gendata , Save() , Extended(kTRUE) );
|
| 230 |
|
| 231 |
if( display ){
|
| 232 |
|
| 233 |
can[i] = new TCanvas(Form("can_%i",i),Form("can_%i",i),600,600);
|
| 234 |
can[i]->cd();
|
| 235 |
|
| 236 |
RooPlot* frame = mva.frame();
|
| 237 |
frame->SetXTitle(mvaname);
|
| 238 |
gendata->plotOn(frame);
|
| 239 |
datapdf.plotOn(frame,Components(sigpdf));
|
| 240 |
datapdf.plotOn(frame,Components(bkgpdf),LineColor(kRed));
|
| 241 |
datapdf.plotOn(frame,LineColor(kOrange));
|
| 242 |
frame->Draw();
|
| 243 |
|
| 244 |
if( printgif ) can[i]->Print(Form("plots/%s_%i.gif",mvaname,i));
|
| 245 |
}
|
| 246 |
|
| 247 |
//---------------------------------------------
|
| 248 |
// print output to screen
|
| 249 |
//---------------------------------------------
|
| 250 |
|
| 251 |
float nll = result->minNll();
|
| 252 |
float bkg_nll = bkg_result->minNll();
|
| 253 |
float signif = sqrt( -2 * ( nll - bkg_nll ) );
|
| 254 |
|
| 255 |
cout << endl << endl;
|
| 256 |
cout << "Fit Results-------------------------------------------" << endl;
|
| 257 |
cout << "gen events = " << gendata->sumEntries() << endl;
|
| 258 |
cout << "nll (BKG) = " << bkg_nll << endl;
|
| 259 |
cout << "nll = " << nll << endl;
|
| 260 |
cout << "sig = " << signif << endl;
|
| 261 |
cout << "nsig = " << nsig.getVal() << " +/- " << nsig.getError() << endl;
|
| 262 |
cout << "nsig(true) = " << nsigtrue << endl;
|
| 263 |
cout << "nbkg = " << nbkg.getVal() << " +/- " << nbkg.getError() << endl;
|
| 264 |
cout << "nbkg(true) = " << nbkgtrue << endl;
|
| 265 |
cout << "------------------------------------------------------" << endl;
|
| 266 |
cout << endl << endl;
|
| 267 |
|
| 268 |
|
| 269 |
nsigfit[i] = nsig.getVal();
|
| 270 |
nsigerrfit[i] = nsig.getError();
|
| 271 |
nbkgfit[i] = nbkg.getVal();
|
| 272 |
nbkgerrfit[i] = nbkg.getError();
|
| 273 |
significance[i] = signif;
|
| 274 |
|
| 275 |
hsig->Fill( nsigfit[i] );
|
| 276 |
hsigpull->Fill( ( nsigfit[i] - nsigtrue ) / nsigerrfit[i] );
|
| 277 |
hbkg->Fill( nbkgfit[i] );
|
| 278 |
hbkgpull->Fill( ( nbkgfit[i] - nbkgtrue ) / nbkgerrfit[i] );
|
| 279 |
hsignificance->Fill( significance[i] );
|
| 280 |
}
|
| 281 |
|
| 282 |
gStyle->SetOptFit(0111);
|
| 283 |
|
| 284 |
TCanvas *c1 = new TCanvas("c1","",1200,800);
|
| 285 |
c1->Divide(3,2);
|
| 286 |
|
| 287 |
c1->cd(1);
|
| 288 |
hsig->GetXaxis()->SetTitle("Sig Yield");
|
| 289 |
hsig->Draw();
|
| 290 |
hsig->Fit("gaus");
|
| 291 |
|
| 292 |
c1->cd(2);
|
| 293 |
hsigpull->GetXaxis()->SetTitle("Sig Pull");
|
| 294 |
hsigpull->Draw();
|
| 295 |
hsigpull->Fit("gaus");
|
| 296 |
|
| 297 |
c1->cd(4);
|
| 298 |
hbkg->GetXaxis()->SetTitle("Bkg Yield");
|
| 299 |
hbkg->Draw();
|
| 300 |
hbkg->Fit("gaus");
|
| 301 |
|
| 302 |
c1->cd(5);
|
| 303 |
hbkgpull->GetXaxis()->SetTitle("Bkg Pull");
|
| 304 |
hbkgpull->Draw();
|
| 305 |
hbkgpull->Fit("gaus");
|
| 306 |
|
| 307 |
c1->cd(3);
|
| 308 |
hsignificance->GetXaxis()->SetTitle("Significance");
|
| 309 |
hsignificance->Draw();
|
| 310 |
hsignificance->Fit("gaus");
|
| 311 |
|
| 312 |
if( printgif ) c1->Print(Form("plots/%s_fitval.gif",mvaname));
|
| 313 |
|
| 314 |
|
| 315 |
}
|
