def RooFitHist(inputhist, title='title', path='.'):
# RooFit.gErrorIgnoreLevel = RooFit.kInfo # <6.02
# RooFit.SetSilentMode()# >6.02
# RooMsgService().instance().SetSilentMode(kTRUE)
fitbinning = array('d')
binwidth = 200
#NBins=(14000/binwidth) - ( (1040/binwidth) + 1 )
NBins = (14000 / binwidth) - ((1040 / binwidth) + 1)
for i in range(NBins + 1):
fitbinning.append(1050 + i * binwidth)
# print(fitbinning)
hist = inputhist.Rebin(NBins, "fit parameter", fitbinning)
meanstart = hist.GetBinCenter(hist.GetMaximumBin())
sigmastart = hist.GetRMS()
print('meanstart:', meanstart, 'sigmastart:', sigmastart)
# inputhist.Draw()
# hist.Draw()
# hold=raw_input('press enter to exit.')
gStyle.SetOptFit(1100)
gStyle.SetOptTitle(0)
RooFit.SumW2Error(kTRUE)
mjj = RooRealVar('mjj', 'M_{jj-AK8}', fitbinning[0],
fitbinning[len(fitbinning) - 1], 'GeV')
mjjral = RooArgList(mjj)
dh = RooDataHist('dh', 'dh', mjjral, RooFit.Import(hist))
shapes = {}
#Gaussian
gaussmean = RooRealVar('#mu_{gauss}', 'mass mean value', meanstart, 0,
2 * meanstart)
gausssigma = RooRealVar('#sigma_{gauss}', 'mass resolution', sigmastart, 0,
2 * sigmastart)
gauss = RooGaussian('gauss', 'gauss', mjj, gaussmean, gausssigma)
shapes.update({'Gauss': gauss})
#CrystalBall
mean = RooRealVar('#mu', 'mean', meanstart, 0, 2 * meanstart)
sigma = RooRealVar('#sigma', 'sigma', sigmastart, 0, 2 * sigmastart)
alpha = RooRealVar('#alpha', 'Gaussian tail', -1000, 0)
n = RooRealVar('n', 'Normalization', -1000, 1000)
cbshape = RooCBShape('cbshape', 'crystalball PDF', mjj, mean, sigma, alpha,
n)
shapes.update({'CrystalBall': cbshape})
#Voigt
voigtmean = RooRealVar('#mu', 'mass mean value', meanstart, 0,
2 * meanstart)
voigtwidth = RooRealVar('#gamma', 'width of voigt', 0, 5000)
voigtsigma = RooRealVar('#sigma', 'mass resolution', sigmastart, 0,
2 * sigmastart)
voigt = RooVoigtian('voigt', 'voigt', mjj, voigtmean, voigtwidth,
voigtsigma)
shapes.update({'Voigt': voigt})
#BreitWigner
bwmean = RooRealVar('#mu', 'mass mean value', meanstart, 0, 2 * meanstart)
bwwidth = RooRealVar('#sigma', 'width of bw', sigmastart, 0,
2 * sigmastart)
bw = RooBreitWigner('bw', 'bw', mjj, bwmean, bwwidth)
shapes.update({'BreitWigner': bw})
#Landau
landaumean = RooRealVar('#mu_{landau}', 'mean landau', meanstart, 0,
2 * meanstart)
landausigma = RooRealVar('#sigma_{landau}', 'mass resolution', sigmastart,
0, 2 * sigmastart)
landau = RooLandau('landau', 'landau', mjj, landaumean, landausigma)
shapes.update({'Landau': landau})
#LandauGauss Convolution
landaugauss = RooFFTConvPdf('landaugauss', 'landau x gauss', mjj, landau,
gauss)
shapes.update({'LandauGauss': landaugauss})
#Logistics
# logisticsmean=RooRealVar('#mu_{logistics}','mean logistics',meanstart,0,2*meanstart)
# logisticssigma= RooRealVar('#sigma_{logistics}','mass resolution',sigmastart,0,2*sigmastart)
# logistics=RooLogistics('logistics','logistics',mjj,logisticsmean,logisticssigma)
# shapes.update({'Logistics':logistics})
#ExpAndGauss
# expgaussmean=RooRealVar('#mu_{expgauss}','mean expgauss',meanstart,0,2*meanstart)
# expgausssigma= RooRealVar('#sigma_{expgauss}','mass resolution',sigmastart,0,2*sigmastart)
# expgausstrans= RooRealVar('trans','trans',0,100)
# expgauss=RooExpAndGauss('expgauss','expgauss',mjj,expgaussmean,expgausssigma,expgausstrans)
# shapes.update({'ExpAndGauss':expgauss})
#BifurGauss
BifurGaussmean = RooRealVar('#mu_{BifurGauss}', 'mean BifurGauss',
meanstart, 0, 2 * meanstart)
BifurGausslsigma = RooRealVar('#sigma_{left}', 'mass resolution',
sigmastart, 0, 2 * sigmastart)
BifurGaussrsigma = RooRealVar('#sigma_{right}', 'mass resolution',
sigmastart, 0, 2 * sigmastart)
BifurGauss = RooBifurGauss('BifurGauss', 'BifurGauss', mjj, BifurGaussmean,
BifurGausslsigma, BifurGaussrsigma)
shapes.update({'BifurGauss': BifurGauss})
#Chebychev
Chebychev1 = RooRealVar('c0', 'Chebychev0', -1000, 1000)
Chebychev2 = RooRealVar('c1', 'Chebychev1', -1000, 1000)
Chebychev3 = RooRealVar('c2', 'Chebychev2', 2, -1000, 1000)
Chebychev = RooChebychev('Chebychev', 'Chebychev', mjj,
RooArgList(Chebychev1, Chebychev2, Chebychev3))
shapes.update({'Chebychev': Chebychev})
#Polynomial
Polynomial1 = RooRealVar('Polynomial1', 'Polynomial1', 100, 0, 1000)
Polynomial2 = RooRealVar('Polynomial2', 'Polynomial2', 100, 0, 1000)
Polynomial = RooPolynomial('Polynomial', 'Polynomial', mjj,
RooArgList(Polynomial1, Polynomial2))
shapes.update({'Polynomial': Polynomial})
# mjj.setRange("FitRange",1050.,14000.)
# for fname in ['Gauss','Logistics','BifurGauss']:
for fname in ['BifurGauss']:
plottitle = '%s Fit of %s' % (fname, title)
shape = shapes[fname]
# shape.fitTo(dh,RooFit.Range("FitRange"),RooFit.SumW2Error(True))
shape.fitTo(dh, RooFit.SumW2Error(True))
frame = mjj.frame(RooFit.Title(plottitle))
frame.GetYaxis().SetTitleOffset(2)
dh.plotOn(frame, RooFit.MarkerStyle(4))
shape.plotOn(frame, RooFit.LineColor(2))
ndof = dh.numEntries() - 3
#chiSquare legend
chi2 = frame.chiSquare()
probChi2 = TMath.Prob(chi2 * ndof, ndof)
chi2 = round(chi2, 2)
probChi2 = round(probChi2, 2)
leg = TLegend(0.5, 0.5, 0.9, 0.65)
leg.SetBorderSize(0)
leg.SetFillStyle(0)
shape.paramOn(frame, RooFit.Layout(0.5, 0.9, 0.9))
leg.AddEntry(0, '#chi^{2} =' + str(chi2), '')
leg.AddEntry(0, 'Prob #chi^{2} = ' + str(probChi2), '')
leg.SetTextSize(0.04)
frame.addObject(leg)
canv = TCanvas(plottitle, plottitle, 700, 700)
canv.SetLogy()
canv.SetLeftMargin(0.20)
canv.cd()
frame.SetMinimum(10**(-3))
frame.Draw()
canv.Print(path + '/%s__%s.eps' % (title, fname))
return chi2
def RooFitHist(inputhist,title='title',path='.'):
# RooFit.gErrorIgnoreLevel = RooFit.kInfo # <6.02
# RooFit.SetSilentMode()# >6.02
# RooMsgService().instance().SetSilentMode(kTRUE)
fitbinning=array('d')
binwidth=200
#NBins=(14000/binwidth) - ( (1040/binwidth) + 1 ) Mjj
NBins=(8000/binwidth) - ( (200/binwidth)+1 )#pT
for i in range(NBins+1):
fitbinning.append(210+i*binwidth)# Mjj 1050
# print(fitbinning)
#import numpy as np
#fitbinning=np.linspace(0,8000,81)
hist=inputhist.Rebin(NBins,"fit parameter",fitbinning)
#hist=inputhist.Rebin(80,"fit parameter", fitbinning)
#meanstart=hist.GetBinCenter(2000)
meanstart=hist.GetBinCenter(hist.GetMaximumBin())#maximum
sigmastart=hist.GetRMS()
#sigmastart=3000
print('meanstart:',meanstart,'sigmastart:',sigmastart)
# inputhist.Draw()
# hist.Draw()
# hold=raw_input('press enter to exit.')
gStyle.SetOptFit(1111)
gStyle.SetOptTitle(0)
RooFit.SumW2Error(kTRUE)
RooFit.Extended(kTRUE)
# RooDataHist::adjustBinning(dh): fit range of variable mjj expanded to nearest bin boundaries: [1050,13850] --> [1050,13850]
mjj=RooRealVar('mjj','P_{T-AK8}',fitbinning[0],fitbinning[len(fitbinning)-1],'GeV')
mjjral=RooArgList(mjj)
dh=RooDataHist('dh','dh',mjjral,RooFit.Import(hist))
#shape.fitTo(dh,RooFit.Range("FitRange"),RooFit.Extended(True),RooFit.SumW2Error(False))
shapes={}
#Gaussian not really
# 3rd, 4th and 5th arguments are: (starting value, minimum possible value, maximum possible value) -- not goot
gaussmean = RooRealVar('#mu_{gauss}','mass mean value',meanstart,0,2*meanstart)
gausssigma= RooRealVar('#sigma_{gauss}','mass resolution',sigmastart,0,2*meanstart)
gauss=RooGaussian('gauss','gauss',mjj,gaussmean,gausssigma)
shapes.update({'Gauss':gauss})
#poisson
poissonmean = RooRealVar('#mu_{poisson}', 'pT mean value', meanstart,0,2*meanstart)
poissonsigma = RooRealVar('#sigma_{poisson]', 'pT resolution', sigmastart, 0, 2*sigmastart)
poisson = RooPoisson('poisson', 'poisson', mjj, poissonmean, False)
shapes.update({'Poisson': poisson})
#Landau -- not good
landaumean=RooRealVar('#mu_{landau}','mean landau',meanstart,0,2*meanstart)
landausigma= RooRealVar('#sigma_{landau}','mass resolution',sigmastart,0,2*sigmastart)#bzw8
landau=RooLandau('landau','landau',mjj,landaumean,landausigma)
shapes.update({'Landau':landau})
#CrystalBall -> this is close to be good but matrix error :(
mean = RooRealVar('#mu','mean',meanstart,0,2*meanstart)
sigma= RooRealVar('#sigma','sigma',sigmastart,0,2*sigmastart)
alpha=RooRealVar('#alpha','Gaussian tail',-1000,0)
n=RooRealVar('n','Normalization',-1000,1000)
cbshape=RooCBShape('cbshape','crystalball PDF',mjj,mean,sigma,alpha,n)
shapes.update({'CrystalBall':cbshape})
#Voigt ---pff
voigtmean = RooRealVar('#mu','mass mean value',meanstart,0,2*meanstart)
voigtwidth = RooRealVar('#gamma','width of voigt',0,100)
voigtsigma= RooRealVar('#sigma','mass resolution',sigmastart,0,150)
voigt=RooVoigtian('voigt','voigt',mjj,voigtmean,voigtwidth,voigtsigma)
shapes.update({'Voigt':voigt})
#BreitWigner--worst
bwmean = RooRealVar('#mu','mass mean value',meanstart,0,2*meanstart)
bwwidth = RooRealVar('#sigma','width of bw',sigmastart,100, 150)
bw=RooBreitWigner('bw','bw',mjj,bwmean,bwwidth)
shapes.update({'BreitWigner':bw})
#Logistics
#logisticsmean=RooRealVar('#mu_{logistics}','mean logistics',meanstart,0,2*meanstart)
#logisticssigma= RooRealVar('#sigma_{logistics}','mass resolution',sigmastart,0,2*sigmastart)
#logistics=RooLogistics('logistics','logistics',mjj,logisticsmean,logisticssigma)
#shapes.update({'Logistics':logistics})
#ExpAndGauss
expgaussmean=RooRealVar('#mu_{expgauss}','mean expgauss',meanstart,490,900)
expgausssigma= RooRealVar('#sigma_{expgauss}','mass resolution',sigmastart,20,3*sigmastart)
expgausstrans= RooRealVar('trans','trans',0,1000)
ExpAndGauss=RooExpAndGauss('expgauss','expgauss',mjj,expgaussmean,expgausssigma,expgausstrans)
shapes.update({'ExpAndGauss':ExpAndGauss})
#Exp
#expmean=RooRealVar('')
#BifurGauss -bad
BifurGaussmean=RooRealVar('#mu_{BifurGauss}','mean BifurGauss',meanstart,0,2*meanstart)
BifurGausslsigma= RooRealVar('#sigma_{left}','mass resolution',sigmastart,200,2*sigmastart)#2*sigmastart
BifurGaussrsigma= RooRealVar('#sigma_{right}','mass resolution',sigmastart,200,2*sigmastart)
BifurGauss=RooBifurGauss('BifurGauss','BifurGauss',mjj,BifurGaussmean,BifurGausslsigma,BifurGaussrsigma)
shapes.update({'BifurGauss':BifurGauss})
#Chebychev -nope
Chebychev1=RooRealVar('c0','Chebychev0',-1000,1000)
Chebychev2= RooRealVar('c1','Chebychev1',-1000,1000)
Chebychev3= RooRealVar('c2','Chebychev2',2,-1000,1000)
Chebychev=RooChebychev('Chebychev','Chebychev',mjj,RooArgList(Chebychev1,Chebychev2,Chebychev3))
shapes.update({'Chebychev':Chebychev})
#Polynomial -nope
Polynomial1=RooRealVar('Polynomial1','Polynomial1',100,0,1000)
Polynomial2= RooRealVar('Polynomial2','Polynomial2',100,0,1000)
Polynomial=RooPolynomial('Polynomial','Polynomial',mjj,RooArgList(Polynomial1,Polynomial2))
shapes.update({'Polynomial':Polynomial})
#pareto
#___________________________________________________________We will see
#Convolutions
#-> use bin > 1000 for better accuracy -----cyclic trouble
#-> Convolution depends on order!!! RooFFTConvPdf the first p.d.f. is the theory model and that the second p.d.f. is the resolution model
#
#LandauGauss Convolution - really bad
landaugauss=RooFFTConvPdf('landaugauss','landau x gauss',mjj,landau, gauss)
#landaugauss.setBufferFraction(126)
shapes.update({'LandauGauss':landaugauss})
#GaussLandau Convolution -> Better but NOT posdef. ...but for 100 it is
gausslandau=RooFFTConvPdf('gausslandau','gauss x landau ',mjj,gauss,landau)
#gausslandau.setBufferFraction(126)
shapes.update({'GaussLandau':gausslandau})
#CrystalBallLandau Convolution cbshape x landau looks better -> status failed
crystlandau=RooFFTConvPdf('crystallandau','cbshape x landau', mjj, landau, cbshape)
crystlandau.setBufferFraction(39)
shapes.update({'CrystLandau': crystlandau})
#BifurGaussLandau Convolution -> Better in look, NO matrix error for Binwidth 200
BifurGaussLandau=RooFFTConvPdf('bifurgausslandau','landau x bifurgauss',mjj,landau, BifurGauss)
BifurGaussLandau.setBufferFraction(39) #against over cycling
shapes.update({'BifurGaussLandau':BifurGaussLandau})
#CrystalGauss Convolution looks better -> status failed
crystgauss=RooFFTConvPdf('crystalgauss','cbshape x gauss', mjj, cbshape, gauss)
#crystgauss.setBufferFraction(39)
shapes.update({'CrystGauss': crystgauss})
#BreitWignerLandau Convolution (Breitwigner = Lorentz)-> status OK....
BreitWignerLandau=RooFFTConvPdf('breitwignerlandau','breitwigner x landau',mjj,landau,bw,3)
#BreitWignerLandau.setBufferFraction(48) #setBufferFraction(fraction of the sampling array size) ->cyclic behaviour fix
#crystgauss.setShift(0,0)
#s1 and s2 are the amounts by which the sampling ranges for pdf's are shifted respectively
#(0,-(xmin+xmax)/2) replicates the default behavior
#(0,0) disables the shifting feature altogether
shapes.update({'BreitWignerLandau':BreitWignerLandau})
for fname in ['ExpAndGauss']:
plottitle='%s Fit of %s'%(fname,title)
shape=shapes[fname]
# shape.fitTo(dh,RooFit.Range("FitRange"),RooFit.SumW2Error(True))
#shape.fitTo(dh,RooFit.Extended(True),RooFit.SumW2Error(True))
mjj.setRange("FitRange",500,4000)#tried
shape.fitTo(dh,RooFit.Range("FitRange"),RooFit.Extended(False),RooFit.SumW2Error(True))
frame=mjj.frame(RooFit.Title(plottitle))
#frame.GetYaxis().SetTitleOffset(2)
dh.plotOn(frame,RooFit.MarkerStyle(4))
shape.plotOn(frame,RooFit.LineColor(2))
ndof=dh.numEntries()-3
print ('ndof', ndof)
#chiSquare legend
chi2 = frame.chiSquare()#there are 2 chiSquare. the 2cond returns chi2/ndf /// \return \f$ \chi^2 / \mathrm{ndf} \f$
print ('chi2', chi2)
probChi2 = TMath.Prob(chi2*ndof, ndof)# why chi2*ndof ?! makes no sense to me
#Double_t Prob(Double_t chi2, Int_t ndf)
#Computation of the probability for a certain Chi-squared (chi2)
#and number of degrees of freedom (ndf).
#P(a,x) represents the probability that the observed Chi-squared
#for a correct model should be less than the value chi2.
#The returned probability corresponds to 1-P(a,x), !!!!
#which denotes the probability that an observed Chi-squared exceeds
#the value chi2 by chance, even for a correct model.
#--- NvE 14-nov-1998 UU-SAP Utrecht
#probChi2=TMath.Prob(chi2, ndof)
chi2 = round(chi2,2)
#probChi2 = round(probChi2,2)
leg = TLegend(0.5,0.5,0.5,0.65)#0.9
leg.SetBorderSize(0)
leg.SetFillStyle(0)
shape.paramOn(frame, RooFit.Layout(0.5,0.9,0.9))
leg.AddEntry(0,'#chi^{2} ='+str(chi2),'')
leg.AddEntry(0,'Prob #chi^{2} = '+str(probChi2),'')
leg.SetTextSize(0.04)
frame.addObject(leg)
canv=TCanvas(plottitle,plottitle,700,700)
canv.SetLogy()
canv.SetLeftMargin(0.20)
canv.cd()
frame.SetMaximum(10**(1))
frame.SetMinimum(10**(-11))#from -3 -> -6
frame.Draw()
#canv.Print(path+'/%s__%sS2MoreLessY.pdf'%(title,fname))
raw_input('press enter to continue')
return chi2
l0=RooRealVar("l0","l0",100.,0.,1000.)
l1=RooRealVar("l1","l1",1.,0.,1000.)
l2=RooRealVar("l2","l2",1.,0.,1000.)
#sigbkg=RooPolynomial("sigbkg"+str(cut),"bkg",mass,RooArgList(l0,l1,l2))
#sigbkg=RooChebychev("sigbkg"+str(cut),"bkg",mass,RooArgList(l0,l1))
sigbkg=RooLogistics("sigbkg"+str(cut),"bkg",mass,l0,l1)
#sigbkg=RooExponential("sigbkg"+str(cut),"sigbkg",mass,l0)
nsigref=RooRealVar("nsigref","number of signal events",100,0,1e10)
sigmodel=RooAddPdf("sigmodel"+str(cut),"sig+bkg",RooArgList(sigbkg,sig),RooArgList(nbkg,nsigref))
s0.setConstant(False)
s1.setConstant(False)
sigmodel.fitTo(signal,RooFit.SumW2Error(True)) #,RooFit.Range("signal")
sigmodel.fitTo(signal,RooFit.SumW2Error(True)) #,RooFit.Range("signal")
sigmodel.fitTo(signal,RooFit.SumW2Error(True)) #,RooFit.Range("signal")
chi2=RooChi2Var("chi2","chi2",sigmodel,signal,RooFit.DataError(RooAbsData.SumW2))
nbins=data.numEntries()
nfree=sigmodel.getParameters(data).selectByAttrib("Constant",False).getSize()
s0.setConstant(True)
s1.setConstant(True)
if cut>=1:
fullintegral=sumsighist.Integral()
else:
fullintegral=sighist.Integral()
print "SIGNAL FRACTION",nsigref.getValV()/(nsigref.getValV()+nbkg.getValV())
if nsigref.getValV()==0: continue
if fit=="data":
xframe=mass.frame(RooFit.Title("signal fraction in peak ="+str(int(nsigref.getValV()/(nsigref.getValV()+nbkg.getValV())*1000.)/1000.)+"+-"+str(int(nsigref.getError()/(nsigref.getValV()+nbkg.getValV())*1000.)/1000.)+", #chi^{2}/DOF = "+str(int((chi2.getVal()/(nbins-nfree))*10.)/10.)))
signal.plotOn(xframe,RooFit.DataError(RooAbsData.SumW2))
sigmodel.plotOn(xframe,RooFit.Normalization(1.0,RooAbsReal.RelativeExpected))
nsigref = RooRealVar("nsigref", "number of signal events", 100,
0, 1e10)
sigmodel = RooAddPdf("sigmodel" + str(cut), "sig+bkg",
RooArgList(sigbkg, sig),
RooArgList(nbkg, nsigref))
s0.setConstant(False)
s1.setConstant(False)
sigmodel.fitTo(
signal, RooFit.SumW2Error(True)) #,RooFit.Range("signal")
sigmodel.fitTo(
signal, RooFit.SumW2Error(True)) #,RooFit.Range("signal")
sigmodel.fitTo(
signal, RooFit.SumW2Error(True)) #,RooFit.Range("signal")
chi2 = RooChi2Var("chi2", "chi2", sigmodel, signal,
RooFit.DataError(RooAbsData.SumW2))
nbins = data.numEntries()
nfree = sigmodel.getParameters(data).selectByAttrib(
"Constant", False).getSize()
s0.setConstant(True)
s1.setConstant(True)
if cut >= 1:
fullintegral = sumsighist.Integral()
else:
fullintegral = sighist.Integral()
print "SIGNAL FRACTION", nsigref.getValV() / (
nsigref.getValV() + nbkg.getValV())
if nsigref.getValV() == 0: continue
if fit == "data":
xframe = mass.frame(