void fitbkgdataCard(TString configCard="template.config",
bool dobands = true, // create baerror bands for BG models
bool dosignal = false, // plot the signal model (needs to be present)
bool blinded = true, // blind the data in the plots?
bool verbose = true ) {
gROOT->Macro("MitStyle.C");
gStyle->SetErrorX(0);
gStyle->SetOptStat(0);
gROOT->ForceStyle();
TString projectDir;
std::vector<TString> catdesc;
std::vector<TString> catnames;
std::vector<int> polorder;
double massmin = -1.;
double massmax = -1.;
double theCMenergy = -1.;
bool readStatus = readFromConfigCard( configCard,
projectDir,
catnames,
catdesc,
polorder,
massmin,
massmax,
theCMenergy
);
if( !readStatus ) {
std::cerr<<" ERROR: Could not read from card > "<<configCard.Data()<<" <."<<std::endl;
return;
}
TFile *fdata = new TFile(TString::Format("%s/CMS-HGG-data.root",projectDir.Data()),"READ");
if( !fdata ) {
std::cerr<<" ERROR: Could not open file "<<projectDir.Data()<<"/CMS-HGG-data.root."<<std::endl;
return;
}
if( !gSystem->cd(TString::Format("%s/databkg/",projectDir.Data())) ) {
std::cerr<<" ERROR: Could not change directory to "<<TString::Format("%s/databkg/",projectDir.Data()).Data()<<"."<<std::endl;
return;
}
// ----------------------------------------------------------------------
// load the input workspace....
RooWorkspace* win = (RooWorkspace*)fdata->Get("cms_hgg_workspace_data");
if( !win ) {
std::cerr<<" ERROR: Could not load workspace > cms_hgg_workspace_data < from file > "<<TString::Format("%s/CMS-HGG-data.root",projectDir.Data()).Data()<<" <."<<std::endl;
return;
}
RooRealVar *intLumi = win->var("IntLumi");
RooRealVar *hmass = win->var("CMS_hgg_mass");
if( !intLumi || !hmass ) {
std::cerr<<" ERROR: Could not load needed variables > IntLumi < or > CMS_hgg_mass < forom input workspace."<<std::endl;
return;
}
//win->Print();
hmass->setRange(massmin,massmax);
hmass->setBins(4*(int)(massmax-massmin));
hmass->SetTitle("m_{#gamma#gamma}");
hmass->setUnit("GeV");
hmass->setRange("fitrange",massmin,massmax);
hmass->setRange("blind1",100.,110.);
hmass->setRange("blind2",150.,180.);
// ----------------------------------------------------------------------
// some auxiliray vectro (don't know the meaning of all of them ... yet...
std::vector<RooAbsData*> data_vec;
std::vector<RooAbsPdf*> pdfShape_vec; // vector to store the NOT-EXTENDED PDFs (aka pdfshape)
std::vector<RooAbsPdf*> pdf_vec; // vector to store the EXTENDED PDFs
std::vector<RooAbsReal*> normu_vec; // this holds the normalization vars for each Cat (needed in bands for combined cat)
RooArgList normList; // list of range-limityed normalizations (needed for error bands on combined category)
//std::vector<RooRealVar*> coeffv;
//std::vector<RooAbsReal*> normu_vecv; // ???
// ----------------------------------------------------------------------
// define output works
RooWorkspace *wOut = new RooWorkspace("wbkg","wbkg") ;
// util;ities for the combined fit
RooCategory finalcat ("finalcat", "finalcat") ;
RooSimultaneous fullbkgpdf("fullbkgpdf","fullbkgpdf",finalcat);
RooDataSet datacomb ("datacomb", "datacomb", RooArgList(*hmass,finalcat)) ;
RooDataSet *datacombcat = new RooDataSet("data_combcat","",RooArgList(*hmass)) ;
// add the 'combcat' to the list...if more than one cat
if( catnames.size() > 1 ) {
catnames.push_back("combcat");
catdesc.push_back("Combined");
}
for (UInt_t icat=0; icat<catnames.size(); ++icat) {
TString catname = catnames.at(icat);
finalcat.defineType(catname);
// check if we're in a sub-cat or the comb-cat
RooDataSet *data = NULL;
RooDataSet *inData = NULL;
if( icat < (catnames.size() - 1) || catnames.size() == 1) { // this is NOT the last cat (which is by construction the combination)
inData = (RooDataSet*)win->data(TString("data_mass_")+catname);
if( !inData ) {
std::cerr<<" ERROR: Could not find dataset > data_mass_"<<catname.Data()<<" < in input workspace."<<std::endl;
return;
}
data = new RooDataSet(TString("data_")+catname,"",*hmass,Import(*inData)); // copy the dataset (why?)
// append the data to the combined data...
RooDataSet *datacat = new RooDataSet(TString("datacat")+catname,"",*hmass,Index(finalcat),Import(catname,*data)) ;
datacomb.append(*datacat);
datacombcat->append(*data);
// normalization for this category
RooRealVar *nbkg = new RooRealVar(TString::Format("CMS_hgg_%s_bkgshape_norm",catname.Data()),"",800.0,0.0,25e3);
// we keep track of the normalizario vars only for N-1 cats, naming convetnions hystoric...
if( catnames.size() > 2 && icat < (catnames.size() - 2) ) {
RooRealVar* cbkg = new RooRealVar(TString::Format("cbkg%s",catname.Data()),"",0.0,0.0,1e3);
cbkg->removeRange();
normu_vec.push_back(cbkg);
normList.add(*cbkg);
}
/// generate the Bernstrin polynomial (FIX-ME: add possibility ro create other models...)
fstBernModel* theBGmodel = new fstBernModel(hmass, polorder[icat], icat, catname); // using my dedicated class...
std::cout<<" model name is "<<theBGmodel->getPdf()->GetName()<<std::endl;
RooAbsPdf* bkgshape = theBGmodel->getPdf(); // the BG shape
RooAbsPdf* bkgpdf = new RooExtendPdf(TString("bkgpdf")+catname,"",*bkgshape,*nbkg); // the extended PDF
// add the extedned PDF to the RooSimultaneous holding all models...
fullbkgpdf.addPdf(*bkgpdf,catname);
// store the NON-EXTENDED PDF for usgae to compute the error bands later..
pdfShape_vec.push_back(bkgshape);
pdf_vec .push_back(bkgpdf);
data_vec .push_back(data);
} else {
data = datacombcat; // we're looking at the last cat (by construction the combination)
data_vec.push_back(data);
// sum up all the cts PDFs for combined PDF
RooArgList subpdfs;
for (int ipdf=0; ipdf<pdf_vec.size(); ++ipdf) {
subpdfs.add(*pdf_vec.at(ipdf));
}
RooAddPdf* bkgpdf = new RooAddPdf(TString("bkgpdf")+catname,"",subpdfs);
pdfShape_vec.push_back(bkgpdf);
pdf_vec .push_back(bkgpdf); // I don't think this is really needed though....
}
// generate the binned dataset (to be put into the workspace... just in case...)
RooDataHist *databinned = new RooDataHist(TString("databinned_")+catname,"",*hmass,*data);
wOut->import(*data);
wOut->import(*databinned);
}
std::cout<<" ***************** "<<std::endl;
// fit the RooSimultaneous to the combined dataset -> (we could also fit each cat separately)
fullbkgpdf.fitTo(datacomb,Strategy(1),Minos(kFALSE),Save(kTRUE));
RooFitResult *fullbkgfitres = fullbkgpdf.fitTo(datacomb,Strategy(2),Minos(kFALSE),Save(kTRUE));
// in principle we're done now, so store the results in the output workspace
wOut->import(datacomb);
wOut->import(fullbkgpdf);
wOut->import(*fullbkgfitres);
std::cout<<" ***************** "<<std::endl;
if( verbose ) wOut->Print();
std::cout<<" ***************** "<<std::endl;
wOut->writeToFile("bkgdatawithfit.root") ;
if( verbose ) {
printf("IntLumi = %5f\n",intLumi->getVal());
printf("ndata:\n");
for (UInt_t icat=0; icat<catnames.size(); ++icat) {
printf("%i ",data_vec.at(icat)->numEntries());
}
printf("\n");
}
// --------------------------------------------------------------------------------------------
// Now comesd the plotting
// chage the Statistics style...
gStyle->SetOptStat(1110);
// we want to plot in 1GeV bins (apparently...)
UInt_t nbins = (UInt_t) (massmax-massmin);
// here we'll store the curves for the bands...
std::vector<RooCurve*> fitcurves;
// loop again over the cats
TCanvas **canbkg = new TCanvas*[catnames.size()];
RooPlot** plot = new RooPlot*[catnames.size()];
TLatex** lat = new TLatex*[catnames.size()];
TLatex** lat2 = new TLatex*[catnames.size()];
std::cout<<" beofre plotting..."<<std::endl;
for (UInt_t icat=0; icat<catnames.size(); ++icat) {
TString catname = catnames.at(icat);
std::cout<<" trying to plot #"<<icat<<std::endl;
// plot the data and the fit
canbkg[icat] = new TCanvas;
plot [icat] = hmass->frame(Bins(nbins),Range("fitrange"));
std::cout<<" trying to plot #"<<icat<<std::endl;
// first plot the data invisibly... and put the fitted BG model on top...
data_vec .at(icat)->plotOn(plot[icat],RooFit::LineColor(kWhite),MarkerColor(kWhite),Invisible());
pdfShape_vec.at(icat)->plotOn(plot[icat],RooFit::LineColor(kRed),Range("fitrange"),NormRange("fitrange"));
std::cout<<" trying to plot #"<<icat<<std::endl;
// if toggled on, plot also the Data visibly
if( !blinded ) {
data_vec.at(icat)->plotOn(plot[icat]);
}
std::cout<<" trying to plot #"<<icat<<std::endl;
// some cosmetics...
plot[icat]->SetTitle("");
plot[icat]->SetMinimum(0.0);
plot[icat]->SetMaximum(1.40*plot[icat]->GetMaximum());
plot[icat]->GetXaxis()->SetTitle("m_{#gamma#gamma} (GeV/c^{2})");
plot[icat]->Draw();
std::cout<<" trying to plot #"<<icat<<std::endl;
// legend....
TLegend *legmc = new TLegend(0.68,0.70,0.97,0.90);
legmc->AddEntry(plot[icat]->getObject(2),"Data","LPE");
legmc->AddEntry(plot[icat]->getObject(1),"Bkg Model","L");
// this part computes the 1/2-sigma bands.
TGraphAsymmErrors *onesigma = NULL;
TGraphAsymmErrors *twosigma = NULL;
std::cout<<" trying *** to plot #"<<icat<<std::endl;
RooAddition* sumcatsnm1 = NULL;
if ( dobands ) { //&& icat == (catnames.size() - 1) ) {
onesigma = new TGraphAsymmErrors();
twosigma = new TGraphAsymmErrors();
// get the PDF for this cat from the vector
RooAbsPdf *thisPdf = pdfShape_vec.at(icat);
// get the nominal fir curve
RooCurve *nomcurve = dynamic_cast<RooCurve*>(plot[icat]->getObject(1));
fitcurves.push_back(nomcurve);
bool iscombcat = ( icat == (catnames.size() - 1) && catnames.size() > 1);
RooAbsData *datanorm = ( iscombcat ? &datacomb : data_vec.at(icat) );
// this si the nornmalization in the 'sliding-window' (i.e. per 'test-bin')
RooRealVar *nlim = new RooRealVar(TString::Format("nlim%s",catnames.at(icat).Data()),"",0.0,0.0,10.0);
nlim->removeRange();
if( iscombcat ) {
// ----------- HISTORIC NAMING ----------------------------------------
sumcatsnm1 = new RooAddition("sumcatsnm1","",normList); // summing all normalizations epect the last Cat
// this is the normlization of the last Cat
RooFormulaVar *nlast = new RooFormulaVar("nlast","","TMath::Max(0.1,@0-@1)",RooArgList(*nlim,*sumcatsnm1));
// ... and adding it ot the list of norms
normu_vec.push_back(nlast);
}
//if (icat == 1 && catnames.size() == 2) continue; // only 1 cat, so don't need combination
for (int i=1; i<(plot[icat]->GetXaxis()->GetNbins()+1); ++i) {
// this defines the 'binning' we use for the error bands
double lowedge = plot[icat]->GetXaxis()->GetBinLowEdge(i);
double upedge = plot[icat]->GetXaxis()->GetBinUpEdge(i);
double center = plot[icat]->GetXaxis()->GetBinCenter(i);
// get the nominal value at the center of the bin
double nombkg = nomcurve->interpolate(center);
nlim->setVal(nombkg);
hmass->setRange("errRange",lowedge,upedge);
// this is the new extended PDF whith the normalization restricted to the bin-area
RooAbsPdf *extLimPdf = NULL;
if( iscombcat ) {
extLimPdf = new RooSimultaneous("epdf","",finalcat);
// loop over the cats and generate temporary extended PDFs
for (int jcat=0; jcat<(catnames.size()-1); ++jcat) {
RooRealVar *rvar = dynamic_cast<RooRealVar*>(normu_vec.at(jcat));
if (rvar) rvar->setVal(fitcurves.at(jcat)->interpolate(center));
RooExtendPdf *ecpdf = new RooExtendPdf(TString::Format("ecpdf%s",catnames.at(jcat).Data()),"",*pdfShape_vec.at(jcat),*normu_vec.at(jcat),"errRange");
static_cast<RooSimultaneous*>(extLimPdf)->addPdf(*ecpdf,catnames.at(jcat));
}
} else
extLimPdf = new RooExtendPdf("extLimPdf","",*thisPdf,*nlim,"errRange");
RooAbsReal *nll = extLimPdf->createNLL(*datanorm,Extended(),NumCPU(1));
RooMinimizer minim(*nll);
minim.setStrategy(0);
double clone = 1.0 - 2.0*RooStats::SignificanceToPValue(1.0);
double cltwo = 1.0 - 2.0*RooStats::SignificanceToPValue(2.0);
if (iscombcat) minim.setStrategy(2);
minim.migrad();
if (!iscombcat) {
minim.minos(*nlim);
}
else {
minim.hesse();
nlim->removeAsymError();
}
if( verbose )
printf("errlo = %5f, errhi = %5f\n",nlim->getErrorLo(),nlim->getErrorHi());
onesigma->SetPoint(i-1,center,nombkg);
onesigma->SetPointError(i-1,0.,0.,-nlim->getErrorLo(),nlim->getErrorHi());
// to get the 2-sigma bands...
minim.setErrorLevel(0.5*pow(ROOT::Math::normal_quantile(1-0.5*(1-cltwo),1.0), 2)); // the 0.5 is because qmu is -2*NLL
// eventually if cl = 0.95 this is the usual 1.92!
if (!iscombcat) {
minim.migrad();
minim.minos(*nlim);
}
else {
nlim->setError(2.0*nlim->getError());
nlim->removeAsymError();
}
twosigma->SetPoint(i-1,center,nombkg);
twosigma->SetPointError(i-1,0.,0.,-nlim->getErrorLo(),nlim->getErrorHi());
// for memory clean-up
delete nll;
delete extLimPdf;
}
hmass->setRange("errRange",massmin,massmax);
if( verbose )
onesigma->Print("V");
// plot[icat] the error bands
twosigma->SetLineColor(kGreen);
twosigma->SetFillColor(kGreen);
twosigma->SetMarkerColor(kGreen);
twosigma->Draw("L3 SAME");
onesigma->SetLineColor(kYellow);
onesigma->SetFillColor(kYellow);
onesigma->SetMarkerColor(kYellow);
onesigma->Draw("L3 SAME");
plot[icat]->Draw("SAME");
// and add the error bands to the legend
legmc->AddEntry(onesigma,"#pm1 #sigma","F");
legmc->AddEntry(twosigma,"#pm2 #sigma","F");
}
std::cout<<" trying ***2 to plot #"<<icat<<std::endl;
// rest of the legend ....
legmc->SetBorderSize(0);
legmc->SetFillStyle(0);
legmc->Draw();
lat[icat] = new TLatex(103.0,0.9*plot[icat]->GetMaximum(),TString::Format("#scale[0.7]{#splitline{CMS preliminary}{#sqrt{s} = %.1f TeV L = %.2f fb^{-1}}}",theCMenergy,intLumi->getVal()));
lat2[icat] = new TLatex(103.0,0.75*plot[icat]->GetMaximum(),catdesc.at(icat));
lat[icat] ->Draw();
lat2[icat]->Draw();
// -------------------------------------------------------
// save canvas in different formats
canbkg[icat]->SaveAs(TString("databkg") + catname + TString(".pdf"));
canbkg[icat]->SaveAs(TString("databkg") + catname + TString(".eps"));
canbkg[icat]->SaveAs(TString("databkg") + catname + TString(".root"));
}
return;
}
