plt.savefig('%s/altDiphoROC.pdf' % plotDir)
plt.figure(3)
xg.plot_importance(diphoModel)
#plt.show()
plt.savefig('%s/diphoImportances.pdf' % plotDir)
plt.figure(4)
xg.plot_importance(altDiphoModel)
#plt.show()
plt.savefig('%s/altDiphoImportances.pdf' % plotDir)
#draw sig vs background distribution
nOutputBins = 50
theCanv = useSty.setCanvas()
sigScoreW = diphoTestFW * (diphoTestY == 1)
sigScoreHist = r.TH1F('sigScoreHist', 'sigScoreHist', nOutputBins, 0., 1.)
useSty.formatHisto(sigScoreHist)
sigScoreHist.SetTitle('')
sigScoreHist.GetXaxis().SetTitle('Diphoton BDT score')
fill_hist(sigScoreHist, diphoPredY, weights=sigScoreW)
bkgScoreW = diphoTestFW * (diphoTestY == 0)
bkgScoreHist = r.TH1F('bkgScoreHist', 'bkgScoreHist', nOutputBins, 0., 1.)
useSty.formatHisto(bkgScoreHist)
bkgScoreHist.SetTitle('')
bkgScoreHist.GetXaxis().SetTitle('Diphoton BDT score')
fill_hist(bkgScoreHist, diphoPredY, weights=bkgScoreW)
#apply transformation to flatten ggH
for iBin in range(1, nOutputBins + 1):
sigVal = sigScoreHist.GetBinContent(iBin)
bkgVal = bkgScoreHist.GetBinContent(iBin)
sigScoreHist.SetBinContent(iBin, 1.)
for iBin, iPar in enumerate(pars.values()):
for jBin, jPar in enumerate(pars.values()):
proc = theMatrix.GetXaxis().GetBinLabel(iPar + 1)
#print 'Got proc %s, expecting proc %s'%(proc, revPars[iPar])
theVal = theMatrix.GetBinContent(iPar + 1, jPar + 1)
#print 'Value for correlation between %s and %s is %.3f'%(revPars[iPar],revPars[jPar],theVal)
theHist.GetXaxis().SetBinLabel(iBin + 1, prettyProc(revPars[iPar]))
theHist.GetYaxis().SetBinLabel(jBin + 1, prettyProc(revPars[jPar]))
theHist.Fill(iBin, jBin, theVal)
set_color_palette('frenchFlag')
r.gStyle.SetNumberContours(500)
r.gStyle.SetPaintTextFormat('1.2f')
#canv = r.TCanvas('canv','canv')
canv = setCanvas()
formatHisto(theHist)
theHist.GetXaxis().SetTickLength(0.)
theHist.GetXaxis().SetLabelSize(0.05)
theHist.GetYaxis().SetTickLength(0.)
theHist.GetYaxis().SetLabelSize(0.05)
theHist.GetZaxis().SetRangeUser(-1., 1.)
theHist.GetZaxis().SetTickLength(0.)
theHist.Draw('colz,text')
drawCMS(True)
drawEnPu(lumi='%2.1f fb^{-1}' % lumi)
canv.Print('Plots/corrMatrixFormal_%s.png' % ext)
canv.Print('Plots/corrMatrixFormal_%s.pdf' % ext)
bestSlo = sigCountLo
bestBlo = bkgCountLo
bestShi = sigCountHi
bestBhi = bkgCountHi
printStr += 'for proc %g the best outcome was tot signif %1.2f:\n' % (
iProc, bestSignif)
printStr += 'cutHi %1.3f, Shi %1.2f, Bhi %1.2f, signifHi %1.2f \n' % (
bestCutHi, bestShi, bestBhi, bestSignifHi)
printStr += 'cutLo %1.3f, Slo %1.2f, Blo %1.2f, signifLo %1.2f \n' % (
bestCutLo, bestSlo, bestBlo, bestSignifLo)
printStr += '\n'
#now do some checking that random search found a good minimum
cutLoCheck = r.TGraph()
cutLoCheck.SetName('cutLoCheck_%g' % iProc)
useSty.formatHisto(cutLoCheck.GetHistogram())
cutLoCheck.GetXaxis().SetTitle('Change in cut')
cutLoCheck.GetYaxis().SetTitle('Significance')
cutLoCheck.SetTitle('')
for iCutLo, cutLo in enumerate(
np.arange(bestCutLo - 0.1, bestCutLo + 0.1, 0.01)):
cutHi = bestCutHi
sigHistHi = r.TH1F('sigHistHiTemp', 'sigHistHiTemp', 160, 100, 180)
sigWeightsHi = diphoFW * (diphoY == 1) * (diphoS - 3 == iProc) * (
diphoPredY > cutHi) * (diphoR == iProc)
fill_hist(sigHistHi, diphoM, weights=sigWeightsHi)
sigCountHi = 0.68 * lumi * sigHistHi.Integral()
sigWidthHi = getRealSigma(sigHistHi)
bkgHistHi = r.TH1F('bkgHistHiTemp', 'bkgHistHiTemp', 160, 100, 180)
bkgWeightsHi = dataFW * (dataPredY > cutHi) * (dataR == iProc)
fill_hist(bkgHistHi, dataM, weights=bkgWeightsHi)
print 'reconstructed number of jets = %s' % classTestN
print 'predicted number of jets = %s' % classPredJ
print 'truth number of jets = %s' % classTestJ
print
print 'predicted number of jets = %s' % classPredJ
print 'reconstructed diphoton pt = %s' % classTestP
print 'predicted class = %s' % classPredY
print 'truth class = %s' % classTestY
print
#define reco hists
canv = useSty.setCanvas()
truthHist = r.TH1F('truthHist', 'truthHist', nClasses, -0.5, nClasses - 0.5)
truthHist.SetTitle('')
useSty.formatHisto(truthHist)
predHist = r.TH1F('predHist', 'predHist', nClasses, -0.5, nClasses - 0.5)
useSty.formatHisto(predHist)
predHist.SetTitle('')
rightHist = r.TH1F('rightHist', 'rightHist', nClasses, -0.5, nClasses - 0.5)
useSty.formatHisto(rightHist)
rightHist.SetTitle('')
wrongHist = r.TH1F('wrongHist', 'wrongHist', nClasses, -0.5, nClasses - 0.5)
useSty.formatHisto(wrongHist)
wrongHist.SetTitle('')
for iBin in range(1, nClasses + 1):
truthHist.GetXaxis().SetBinLabel(iBin, binNames[iBin - 1])
predHist.GetXaxis().SetBinLabel(iBin, binNames[iBin - 1])
rightHist.GetXaxis().SetBinLabel(iBin, binNames[iBin - 1])
wrongHist.GetXaxis().SetBinLabel(iBin, binNames[iBin - 1])
def main():
for fileName in fullFileNames:
if 'M125' not in fileName: continue
theProc = fileName.split('pythia8_')[1].split('.root')[0]
theProc0 = theProc.split('_')[0]
#if theProc not in procsOfInterest: continue
print 'processing %s'%theProc
theFile = r.TFile(fileName, 'READ')
theWS = theFile.Get('tagsDumper/cms_hgg_13TeV')
for cat in cats:
dataName = '%s_125_13TeV_%s'%(nameMap[theProc0], cat)
sumEntries = theWS.data(dataName).sumEntries()
if sumEntries < 0.: sumEntries = 0.
if theProc not in procsOfInterest:
theProc = 'OTHER'
sumwProcCatMap[ (theProc,cat) ] += sumEntries
sumwProcMap[ theProc ] += sumEntries
sumwCatMap[ cat ] += sumEntries
print 'got all values'
nBinsX = len(procsOfInterest)
nBinsY = len(cats)
procHist = r.TH2F('procHist','procHist', nBinsX, -0.5, nBinsX-0.5, nBinsY, -0.5, nBinsY-0.5)
procHist.SetTitle('')
catHist = r.TH2F('catHist','catHist', nBinsX, -0.5, nBinsX-0.5, nBinsY, -0.5, nBinsY-0.5)
catHist.SetTitle('')
for iProc,proc in enumerate(procsOfInterest):
#print 'iProc,proc',iProc,proc
for iCat,cat in enumerate(cats):
procWeight = 100. * sumwProcCatMap[(proc,cat)] / sumwProcMap[proc]
if procWeight < 0.5: procWeight=-1
catWeight = 100. * sumwProcCatMap[(proc,cat)] / sumwCatMap[cat]
if catWeight < 0.5: catWeight=-1
catLabel = cat.split('ECO_')[1]
catLabel = catLabel.replace('PTH_0_60','low')
catLabel = catLabel.replace('PTH_60_120','med')
catLabel = catLabel.replace('PTH_120_200','high')
catLabel = catLabel.replace('PTH_GT200','BSM')
catLabel = catLabel.replace('GE2J','2J')
catLabel = catLabel.replace('PTJET1_GT200','BSM')
catLabel = catLabel.replace('VBFTOPO','VBF')
catLabel = catLabel.replace('JET3VETO','2J-like')
catLabel = catLabel.replace('JET3','3J-like')
catLabel = catLabel.replace('VH2JET','VH-like')
catLabel = catLabel.replace('REST','rest')
catLabel = catLabel.replace('_',' ')
procHist.Fill( iProc, iCat, procWeight )
procHist.GetXaxis().SetBinLabel( iProc+1, prettyProc(proc) )
procHist.GetYaxis().SetBinLabel( iCat+1, catLabel )
catHist.Fill( iProc, iCat, catWeight )
catHist.GetXaxis().SetBinLabel( iProc+1, prettyProc(proc) )
catHist.GetYaxis().SetBinLabel( iCat+1, catLabel )
#canv = r.TCanvas('canv','canv')
set_color_palette('ed_noice')
canv = setCanvas()
formatHisto(procHist)
procHist.SetStats(0)
procHist.GetXaxis().SetTitle('STXS process')
procHist.GetXaxis().SetTitleOffset(3)
procHist.GetXaxis().SetTickLength(0.)
procHist.GetXaxis().LabelsOption('v')
procHist.GetYaxis().SetTitle('Event category')
procHist.GetYaxis().SetTitleOffset(2.7)
procHist.GetYaxis().SetTickLength(0.)
procHist.GetZaxis().SetTitle('Signal category destination (%)')
procHist.SetMinimum(-0.00001)
procHist.SetMaximum(100.)
procHist.Draw('colz,text')
lines = []
for iProc,proc in enumerate(procsOfInterest):
lines.append(r.TLine(iProc+0.5, -0.5, iProc+0.5, len(cats)-0.5))
lines[-1].SetLineColorAlpha(r.kGray, 0.5)
lines[-1].SetLineWidth(1)
lines.append(r.TLine(-0.5, 17.5, len(procsOfInterest)-0.5, 17.5)) #horiontal ggH VBF divider
lines[-1].SetLineColorAlpha(r.kBlack, 0.5)
lines[-1].SetLineWidth(1)
lines.append(r.TLine(10.5, -0.5, 10.5, len(cats)-0.5)) #vertical ggH VBF divider
lines[-1].SetLineColorAlpha(r.kBlack, 0.5)
lines[-1].SetLineWidth(1)
for line in lines: line.Draw()
drawCMS(True)
#drawEnPu(lumi='%.1f fb^{-1}'%lumi)
drawEnYear(year=opts.year)
canv.Print('procHist_%s.pdf'%ext)
canv.Print('procHist_%s.png'%ext)
formatHisto(catHist)
catHist.SetStats(0)
catHist.GetXaxis().SetTitle('STXS process')
catHist.GetXaxis().SetTitleOffset(3)
catHist.GetXaxis().SetTickLength(0.)
catHist.GetXaxis().LabelsOption('v')
catHist.GetYaxis().SetTitle('Event category')
catHist.GetYaxis().SetTitleOffset(2.7)
catHist.GetYaxis().SetTickLength(0.)
catHist.GetZaxis().SetTitle('Category signal composition (%)')
catHist.SetMinimum(-0.00001)
catHist.SetMaximum(100.)
catHist.Draw('colz,text')
lines = []
for iCat,cat in enumerate(cats):
if cat.count('Tag1') or cat.count('RECO_1J_PTH_GT200') or cat.count('RECO_VBFTOPO_REST'):
lines.append(r.TLine(-0.5, iCat+0.5, len(procsOfInterest)-0.5, iCat+0.5))
lines[-1].SetLineColorAlpha(r.kGray, 0.5)
lines[-1].SetLineWidth(1)
else:
lines.append(r.TLine(-0.5, iCat+0.5, len(procsOfInterest)-0.5, iCat+0.5))
lines[-1].SetLineColorAlpha(r.kGray, 0.25)
lines[-1].SetLineWidth(1)
lines.append(r.TLine(-0.5, 17.5, len(procsOfInterest)-0.5, 17.5)) #horiontal ggH VBF divider
lines[-1].SetLineColorAlpha(r.kBlack, 0.5)
lines[-1].SetLineWidth(1)
lines.append(r.TLine(10.5, -0.5, 10.5, len(cats)-0.5)) #vertical ggH VBF divider
lines[-1].SetLineColorAlpha(r.kBlack, 0.5)
lines[-1].SetLineWidth(1)
for line in lines: line.Draw()
drawCMS(True)
#drawEnPu(lumi='%.1f fb^{-1}'%lumi)
drawEnYear(year=opts.year)
canv.Print('catHist_%s.pdf'%ext)
canv.Print('catHist_%s.png'%ext)
#save hists
outFile = r.TFile('stxsStage1Hists_%s.root'%ext,'RECREATE')
procHist.Write()
catHist.Write()
outFile.Close()