# globals, and aliases
from ROOT import gPad, gStyle, gSystem, kRed, kBlue, kAzure, kGreen, kBlack
const = RooRealConstant.value
# setup
gStyle.SetOptStat('nemrou')
canvas = TCanvas('canvas', 'canvas', 800, 600)
canvas.Print('{}['.format(plotfile))
# read workspace
ffile = TFile.Open(rfile, 'read')
workspace = ffile.Get('workspace')
pdfs = RooArgList(workspace.allPdfs())
for i in range(pdfs.getSize()):
name = pdfs[i].GetName()
if name.find(mode) >= 0:
mykpdf = pdfs[i]
assert(mykpdf)
## variables
time = RooRealVar('time', 'Time [ps]', 0.2, 15.0)
time.setBins(bins)
time.setBins(bins*3, 'cache')
kfactor = workspace.var('kfactorVar')
kfactor.setRange(0.85, 1.05)
gamma = RooRealVar('gamma', 'gamma', 0.661, 0., 3.)
kgamma = RooProduct('kgamma', 'kgamma', RooArgList(gamma, kfactor))
dGamma = RooRealVar('dGamma', 'dGamma', -0.106, -3., 3.)
usig2 = 0.
totalYield = 0.
sigYield = 0.
sigErrs = {}
sigYieldFilename = 'last_H%i_%s_%iJets_signalYield.txt' % (opts.mH,
modeString, opts.Nj)
sigYieldsFile = open(sigYieldFilename, 'w')
WpJNonPoissonError = 0
print
print '-------------------------------'
print 'Yields in signal box'
print '-------------------------------'
for i in range(0, yields.getSize()):
theName = yields.at(i).GetName()
if theName[0] == 'n':
totalYield += yields.at(i).getVal()
theIntegral = 1.
if (theName == 'nDiboson'):
theIntegral = dibosonInt.getVal()/dibosonFullInt.getVal()
elif (theName == 'nWjets'):
theIntegral = WpJInt.getVal()/WpJFullInt.getVal()
if (yields.at(i).getError()**2 > yields.at(i).getVal()):
WpJNonPoissonError = sqrt(yields.at(i).getError()**2 - \
yields.at(i).getVal())
else:
WpJNonPoissonError = 0.
elif (theName == 'nTTbar'):
theIntegral = ttbarInt.getVal()/ttbarFullInt.getVal()
def findOnePe(hist, ws, name='x', Npe = 1):
fitPed(hist, ws, name)
x = ws.var(name)
ped = ws.pdf('ped')
pedWidth = ws.var('pedWidth')
pdfs = RooArgList(ped)
pdfList = []
fped = RooRealVar('fped', 'f_{ped}', 0.8, 0., 1.)
fractions = RooArgList(fped)
fList = []
peList = []
peMean = RooRealVar('peMean', 'mean_{pe}', 6., 0., 20.)
peWidth = RooRealVar('peWidth', 'width_{pe}', pedWidth.getVal(), 0., 10.)
for i in range(0, Npe):
pem = RooFormulaVar('pem{0}'.format(i+1), '@0+{0}*@1'.format(i+1),
RooArgList(ws.var('pedMean'), peMean))
peList.append(pem)
npepdf = RooGaussian('pe{0}pdf'.format(i+1), 'pe{0}pdf'.format(i+1),
x, pem, pedWidth)
pdfs.add(npepdf)
pdfList.append(npepdf)
fnpe = RooRealVar('fpe{0}'.format(i+1), 'fpe{0}'.format(i+1),
0.5, -0.1, 1.0)
fractions.add(fnpe)
fList.append(fnpe)
#bgMean = RooRealVar("bgMean", "bgMean", 6.0, x.getMin(), x.getMax())
bgScale = RooRealVar("bgScale", "bgScale", 0.5, -1.0, Npe + 1.0)
bgMean = RooFormulaVar("bgMean", "@1+@0*@2",
RooArgList(peMean, ws.var('pedMean'), bgScale))
bgWidthL = RooRealVar("bgWidthL", "bgWidthL", pedWidth.getVal()*2,
0., 25.)
bgWidthR = RooRealVar("bgWidthR", "bgWidthR", pedWidth.getVal()*7,
0., 25.)
bgGauss = RooBifurGauss("bgGauss", "bgGauss", x, bgMean,
bgWidthR, bgWidthR)
if (Npe > 1):
pdfs.add(bgGauss)
else:
fractions.remove(fractions.at(fractions.getSize()-1))
## pem = RooFormulaVar('pem', '@0+@1', RooArgList(peMean, ws.var('pedMean')))
## firstPe = RooGaussian('firstPe', 'firstPe', x, pem, peWidth)
## pdfs.Print("v")
## fractions.Print("v")
pedPlusOne = RooAddPdf('pedPlusOne', 'pedPlusOne', pdfs, fractions, True)
## pedWidth = ped.GetParameter(2)
## pedMean = ped.GetParameter(1)
## pedA = ped.GetParameter(0)
secondMax = hist.GetMaximumBin() + 1
goingDown = True
maxVal = hist.GetBinContent(secondMax)
foundMax = False
while (not foundMax) and (secondMax < hist.GetNbinsX()):
tmpVal = hist.GetBinContent(secondMax+1)
if (tmpVal < maxVal):
if not goingDown:
foundMax = True
else:
goingDown = True
maxVal = tmpVal
secondMax += 1
elif (tmpVal > maxVal):
goingDown = False
maxVal = tmpVal
secondMax += 1
else:
maxVal = tmpVal
secondMax += 1
secondMaxx = hist.GetBinCenter(secondMax)
print 'found 2nd maximum in bin',secondMax,'value',secondMaxx
## peMean.setVal(secondMaxx)
## bgMean.setVal(secondMaxx*0.6)
x.setRange('pedPlus_fit', x.getMin(), ws.var('pedMean').getVal()+pedWidth.getVal()*6.*(Npe+0))
pedPlusOne.fitTo(ws.data('ds'), RooFit.Minos(False),
RooFit.Range('pedPlus_fit'),
RooFit.PrintLevel(1))
getattr(ws, 'import')(pedPlusOne)
parlist = fitresult.floatParsFinal()
cmatrix = fitresult.covarianceMatrix()
veclist = RooArgList()
for i in range(parlist.getSize()):
name = "%s_%d" % (parlist[i].GetName(), i)
veclist.add(parlist[i].clone(name))
multigauss = RooMultiVarGaussian("multigauss", "multigauss", veclist, parlist, cmatrix)
dset = multigauss.generate(RooArgSet(veclist), 1000)
fns = []
for entry in range(dset.numEntries()):
vecset = dset.get(entry)
veclist = RooArgList(vecset)
for pars in range(veclist.getSize()):
acceptance.SetParameter(pars, veclist[pars].getVal())
fns += [acceptance.Clone("%s_%d" % (acceptance.GetName(), entry))]
avgfn = BinnedAvgFunction(fns, xbincs)
avgfn.calculate()
accfns += [avgfn.get_avg_fn()]
accfnerrs += [avgfn.get_avg_fn_var()]
means = numpy.zeros(nbins, dtype=float)
varis = numpy.zeros(nbins, dtype=float)
for ibin in range(nbins):
means[ibin] = accfns[0][ibin] / accfns[1][ibin]
varis[ibin] = accfnerrs[0][ibin] + accfnerrs[1][ibin]
usig2 = 0.
totalYield = 0.
sigYield = 0.
sigErrs = {}
sigYieldFilename = 'last_H%i_%s_%iJets_signalYield.txt' % (opts.mH,
modeString, opts.Nj)
sigYieldsFile = open(sigYieldFilename, 'w')
WpJNonPoissonError = 0
print
print '-------------------------------'
print 'Yields in signal box'
print '-------------------------------'
for i in range(0, yields.getSize()):
theName = yields.at(i).GetName()
if theName[0] == 'n':
totalYield += yields.at(i).getVal()
theIntegral = 1.
if (theName == 'nDiboson'):
theIntegral = dibosonInt.getVal()/dibosonFullInt.getVal()
elif (theName == 'nWjets'):
theIntegral = WpJInt.getVal()/WpJFullInt.getVal()
if (yields.at(i).getError()**2 > yields.at(i).getVal()):
WpJNonPoissonError = sqrt(yields.at(i).getError()**2 - \
yields.at(i).getVal())
else:
WpJNonPoissonError = 0.
elif (theName == 'nTTbar'):
theIntegral = ttbarInt.getVal()/ttbarFullInt.getVal()
def accbuilder(time, knots, coeffs):
# build acceptance function
from copy import deepcopy
myknots = deepcopy(knots)
mycoeffs = deepcopy(coeffs)
from ROOT import (RooBinning, RooArgList, RooPolyVar,
RooCubicSplineFun)
if (len(myknots) != len(mycoeffs) or 0 >= min(len(myknots), len(mycoeffs))):
raise ValueError('ERROR: Spline knot position list and/or coefficient'
'list mismatch')
# create the knot binning
knotbinning = WS(ws, RooBinning(time.getMin(), time.getMax(), 'knotbinning'))
for v in myknots:
knotbinning.addBoundary(v)
knotbinning.removeBoundary(time.getMin())
knotbinning.removeBoundary(time.getMax())
knotbinning.removeBoundary(time.getMin())
knotbinning.removeBoundary(time.getMax())
oldbinning, lo, hi = time.getBinning(), time.getMin(), time.getMax()
time.setBinning(knotbinning, 'knotbinning')
time.setBinning(oldbinning)
time.setRange(lo, hi)
del knotbinning
del oldbinning
del lo
del hi
# create the knot coefficients
coefflist = RooArgList()
i = 0
for v in mycoeffs:
coefflist.add(WS(ws, RooRealVar('SplineAccCoeff%u' % i,
'SplineAccCoeff%u' % i, v)))
i = i + 1
del mycoeffs
coefflist.add(one)
i = i + 1
myknots.append(time.getMax())
myknots.reverse()
fudge = (myknots[0] - myknots[1]) / (myknots[2] - myknots[1])
lastmycoeffs = RooArgList(
WS(ws, RooConstVar('SplineAccCoeff%u_coeff0' % i,
'SplineAccCoeff%u_coeff0' % i, 1. - fudge)),
WS(ws, RooConstVar('SplineAccCoeff%u_coeff1' % i,
'SplineAccCoeff%u_coeff1' % i, fudge)))
del myknots
coefflist.add(WS(ws, RooPolyVar(
'SplineAccCoeff%u' % i, 'SplineAccCoeff%u' % i,
coefflist.at(coefflist.getSize() - 2), lastmycoeffs)))
del i
# create the spline itself
tacc = WS(ws, RooCubicSplineFun('SplineAcceptance', 'SplineAcceptance', time,
'knotbinning', coefflist))
del lastmycoeffs
# make sure the acceptance is <= 1 for generation
m = max([coefflist.at(j).getVal() for j in
xrange(0, coefflist.getSize())])
from ROOT import RooProduct
c = WS(ws, RooConstVar('SplineAccNormCoeff', 'SplineAccNormCoeff', 0.99 / m))
tacc_norm = WS(ws, RooProduct('SplineAcceptanceNormalised',
'SplineAcceptanceNormalised', RooArgList(tacc, c)))
del c
del m
del coefflist
return tacc, tacc_norm
def findOnePe(hist, ws, name='x', Npe=1):
fitPed(hist, ws, name)
x = ws.var(name)
ped = ws.pdf('ped')
pedWidth = ws.var('pedWidth')
pdfs = RooArgList(ped)
pdfList = []
fped = RooRealVar('fped', 'f_{ped}', 0.8, 0., 1.)
fractions = RooArgList(fped)
fList = []
peList = []
peMean = RooRealVar('peMean', 'mean_{pe}', 6., 0., 20.)
peWidth = RooRealVar('peWidth', 'width_{pe}', pedWidth.getVal(), 0., 10.)
for i in range(0, Npe):
pem = RooFormulaVar('pem{0}'.format(i + 1), '@0+{0}*@1'.format(i + 1),
RooArgList(ws.var('pedMean'), peMean))
peList.append(pem)
npepdf = RooGaussian('pe{0}pdf'.format(i + 1),
'pe{0}pdf'.format(i + 1), x, pem, pedWidth)
pdfs.add(npepdf)
pdfList.append(npepdf)
fnpe = RooRealVar('fpe{0}'.format(i + 1), 'fpe{0}'.format(i + 1), 0.5,
-0.1, 1.0)
fractions.add(fnpe)
fList.append(fnpe)
#bgMean = RooRealVar("bgMean", "bgMean", 6.0, x.getMin(), x.getMax())
bgScale = RooRealVar("bgScale", "bgScale", 0.5, -1.0, Npe + 1.0)
bgMean = RooFormulaVar("bgMean", "@1+@0*@2",
RooArgList(peMean, ws.var('pedMean'), bgScale))
bgWidthL = RooRealVar("bgWidthL", "bgWidthL",
pedWidth.getVal() * 2, 0., 25.)
bgWidthR = RooRealVar("bgWidthR", "bgWidthR",
pedWidth.getVal() * 7, 0., 25.)
bgGauss = RooBifurGauss("bgGauss", "bgGauss", x, bgMean, bgWidthR,
bgWidthR)
if (Npe > 1):
pdfs.add(bgGauss)
else:
fractions.remove(fractions.at(fractions.getSize() - 1))
## pem = RooFormulaVar('pem', '@0+@1', RooArgList(peMean, ws.var('pedMean')))
## firstPe = RooGaussian('firstPe', 'firstPe', x, pem, peWidth)
## pdfs.Print("v")
## fractions.Print("v")
pedPlusOne = RooAddPdf('pedPlusOne', 'pedPlusOne', pdfs, fractions, True)
## pedWidth = ped.GetParameter(2)
## pedMean = ped.GetParameter(1)
## pedA = ped.GetParameter(0)
secondMax = hist.GetMaximumBin() + 1
goingDown = True
maxVal = hist.GetBinContent(secondMax)
foundMax = False
while (not foundMax) and (secondMax < hist.GetNbinsX()):
tmpVal = hist.GetBinContent(secondMax + 1)
if (tmpVal < maxVal):
if not goingDown:
foundMax = True
else:
goingDown = True
maxVal = tmpVal
secondMax += 1
elif (tmpVal > maxVal):
goingDown = False
maxVal = tmpVal
secondMax += 1
else:
maxVal = tmpVal
secondMax += 1
secondMaxx = hist.GetBinCenter(secondMax)
print 'found 2nd maximum in bin', secondMax, 'value', secondMaxx
## peMean.setVal(secondMaxx)
## bgMean.setVal(secondMaxx*0.6)
x.setRange('pedPlus_fit', x.getMin(),
ws.var('pedMean').getVal() + pedWidth.getVal() * 6. * (Npe + 0))
pedPlusOne.fitTo(ws.data('ds'), RooFit.Minos(False),
RooFit.Range('pedPlus_fit'), RooFit.PrintLevel(1))
getattr(ws, 'import')(pedPlusOne)
from ROOT import gPad, gStyle, kRed, kBlue, kAzure, kGreen, kBlack
gStyle.SetOptStat('nemrou')
canvas = TCanvas('canvas', 'canvas', 800, 600)
canvas.Print('{}['.format(plotfile))
ffile = TFile.Open(rfile, 'read')
if ifpdf:
workspace = ffile.Get('workspace')
kfactor = workspace.var('kfactorVar')
kfactor.setRange(0.9, 1.1)
pdfs = RooArgList(workspace.allPdfs())
for i in range(pdfs.getSize()):
fr = kfactor.frame()
pdfs[i].plotOn(fr, RooFit.LineColor(kBlack), RooFit.FillColor(kAzure+1),
RooFit.DrawOption('lf')) # FIXME: doesn't draw the line!'
fr.Draw()
canvas.Print(plotfile)
else:
modes = {}
klist = ffile.GetListOfKeys()
for item in klist:
name = item.GetName()
if not name.startswith('mBresn'):
print 'MSG: Skipping, unknown object: %s' % name
continue # ntuples are named mBresn_*
sample = name.split('_')
modes[sample[1]] = modes.get(sample[1],[]) + [item] # up and down for each mode
# knot coefficients
coefflist = RooArgList()
for i, v in enumerate(spline_coeffs):
coefflist.add(const(v))
i = len(spline_coeffs)
coefflist.add(one)
spline_knots.append(time.getMax())
spline_knots.reverse()
fudge = (spline_knots[0] - spline_knots[1]) / (spline_knots[2] -
spline_knots[1])
lastmycoeffs = RooArgList()
lastmycoeffs.add(const(1. - fudge))
lastmycoeffs.add(const(fudge))
polyvar = RooPolyVar('{}_SplineAccCoeff{}'.format(mode, i), '',
coefflist.at(coefflist.getSize() - 2), lastmycoeffs)
coefflist.add(polyvar)
del i
# create the spline itself
tacc = RooCubicSplineFun('{}_SplineAcceptance'.format(mode), '', time,
'{}_knotbinning'.format(mode), coefflist)
#del lastmycoeffs, coefflist
from ROOT import TCanvas, TLine
#tacc.DrawClass();
aCanvas = canvas
import sys
parlist = fitresult.floatParsFinal()
cmatrix = fitresult.covarianceMatrix()
veclist = RooArgList()
for i in range(parlist.getSize()):
name = '%s_%d' % (parlist[i].GetName(), i)
veclist.add(parlist[i].clone(name))
multigauss = RooMultiVarGaussian('multigauss', 'multigauss', veclist, parlist, cmatrix)
dset = multigauss.generate(RooArgSet(veclist), 1000)
fns = []
for entry in range(dset.numEntries()):
vecset = dset.get(entry)
veclist = RooArgList(vecset)
for pars in range(veclist.getSize()):
acceptance.SetParameter(pars, veclist[pars].getVal())
fns += [acceptance.Clone('%s_%d' % (acceptance.GetName(), entry))]
avgfn = BinnedAvgFunction(fns, xbincs)
avgfn.calculate()
accfns += [avgfn.get_avg_fn()]
accfnerrs += [avgfn.get_avg_fn_var()]
means = numpy.zeros(nbins, dtype=float)
varis = numpy.zeros(nbins, dtype=float)
for ibin in range(nbins):
means[ibin] = accfns[0][ibin] / accfns[1][ibin]
varis[ibin] = accfnerrs[0][ibin] + accfnerrs[1][ibin]
inFile.ls()
#s = raw_input("Press Enter to continue");
theTH1DHist = TH1F()
inFile.GetObject("etaHist", theTH1DHist)
#rooRealVarList += [RooRealVar('x','x',0.0,1.0)];
x = RooRealVar('x', 'x', 0.0, 1.0)
xRegions, x = divideEtaTH1Ds(theTH1DHist, x, 5, fNum)
fNum += 1
#xList += [RooAbsReal(x)]
xRegionsList.add(xRegions.clone(xRegions.GetName() + "_clone"))
#etasetDataset.add();
import sys
#sys.exit(0);
for i in range(xRegionsList.getSize()):
xRegionsList.at(0).Print()
#.Print();#writeToStream(ROOT.cout,True);
print "\n\nNOCRASH2\n\n"
from ROOT import TFile
import time
os.chdir(os.environ['B2DXFITTERSROOT'] + '/tutorial')
f = TFile('xRegionsList_%f.root' % time.time(), 'recreate')
f.WriteTObject(xRegionsList, 'xRegionsList_%f' % time.time())
f.Close()
del f
def buildBDecayTimePdf(
config, # configuration dictionary
name, # 'Signal', 'DsPi', ...
ws, # RooWorkspace into which to put the PDF
time, timeerr, qt, qf, mistag, tageff, # potential observables
Gamma, DeltaGamma, DeltaM, # decay parameters
C, D, Dbar, S, Sbar, # CP parameters
timeresmodel = None, # decay time resolution model
acceptance = None, # acceptance function
timeerrpdf = None, # pdf for per event time error
mistagpdf = None, # pdf for per event mistag
mistagobs = None, # real mistag observable
kfactorpdf = None, # distribution k factor smearing
kvar = None, # variable k which to integrate out
aprod = None, # production asymmetry
adet = None, # detection asymmetry
atageff = None # asymmetry in tagging efficiency
):
"""
build a B decay time pdf
parameters:
-----------
config -- config dictionary
name -- name prefix for newly created objects
ws -- workspace into which to import created objects
time -- time observable
timeerr -- time error (constant for average, or per-event obs.)
qt -- tagging decision (RooCategory)
qf -- final state charge (RooCategory)
mistag -- mistag (either average, or calibrated omega(eta))
tageff -- tagging efficiency
Gamma -- width of decay (1/lifetime)
DeltaGamma -- width difference
DeltaM -- mass difference
C -- term in front of cos in RooBDecay
D -- term in front of sinh in RooBDecay (f final state)
Dbar -- term in front of sinh in RooBDecay (fbar final state)
S -- term in front of sin in RooBDecay (f final state)
Sbar -- term in front of sin in RooBDecay (fbar final state)
timeresmodel -- decay time resolution model (or None for delta fn)
acceptance -- acceptance (or None for flat efficiency(decay time))
timeerrpdf -- decay time error distribution (or None for average)
mistagpdf -- per event mistag distribution(s) (or None for average)
mistagobs -- (uncalibrated) per event mistag observable (or None)
kfactorpdf -- k factor distribution (None for no k-factor correction)
kvar -- k factor variable (None for no k-factor correction)
aprod -- production asymmetry (None for zero asymmetry)
adet -- detection asymmetry (None for zero asymmetry)
atageff -- tagging eff. asymmetry (None, or list, 1 per tagger/cat.)
returns:
--------
a time pdf built according to specification
This routine is a rather flexible beast and can accomodate a multitude of
use cases:
- ideal, average or per-event decay time resolution (pdf needed for
per-event case)
- ideal, average, per category, or per-event mistag:
* ideal: mistag = zero, mistagpdf = mistagobs = None
* average: mistag = RooRealVar/RooConstVar, mistagpdf = mistagobs = None
* per category: qt = -Ncat, ... , 0, 1, ..., Ncat, mistag = RooArgList
of per-category average mistags (1, ..., Ncat),i
mistagpdf = mistagobs = None (can be used for calibrations!)
* per event: qt = -1, 0, +1, mistag = calibrated mistag omega(eta),
mistagpdf = P(eta), mistagobs = eta
* per event, mutually exclusive taggers: like per category, but
mistagpdf becomes a list of pdfs as well
* one can have as many tagging asymmetries as there are
categories/mutually exclusive taggers (for a single one, just pass the
RooRealVar, else a RooArgList of RooRealVars)
* if the calibration needs to differ between B and Bbar, mistag can also
be a list of two calibrations "[ calB, calBbar ]", or a list of
RooArgLists (one per category/mutually exclusive tagger)
- optionally, a k-factor correction can be applied for partially
reconstructed or misidentified modes
See the classes RooBDecay (in RooFit), DecRateCoeff (B2DXFitters) and
RooKResModel (also B2DXFitters) for details on the precise meaning and
definition of these variables.
relevant config dictionary keys:
--------------------------------
'Debug':
print all arguments to buildBDecayTimePdf before doing anything -
useful to see what "really goes in" when it doesn't do what it
should...
'UseKFactor':
if True, apply the k-factor correction, if False, k-factor correction
can be disabled globally (useful for fast fitback look toys,
systematic studies, etc.)
'ParameteriseIntegral':
if True, save a huge amount of CPU by parametrising the resolutition
model integral and its convolutions with the decay time pdf as a
function of (per-event) decay time; see parameteriseResModelIntegrals
for details
For more information on how various bits affect the pdf built, see also
the documentation for the following helper routines:
applyBinnedAcceptance, applyKFactorSmearing, applyDecayTimeErrPdf,
applyUnbinnedAcceptance, parameteriseResModelIntegrals
"""
# Look in LHCb-INT-2011-051 for the conventions used
from ROOT import ( RooConstVar, RooProduct, RooTruthModel, RooGaussModel,
Inverse, RooBDecay, RooProdPdf, RooArgSet, DecRateCoeff,
RooArgList )
if config['Debug']:
print 72 * '#'
kwargs = {
'config': config,
'name': name,
'ws': ws,
'time': time,
'timeerr': timeerr,
'qt': qt,
'qf': qf,
'mistag': mistag,
'tageff': tageff,
'Gamma': Gamma,
'DeltaGamma': DeltaGamma,
'DeltaM': DeltaM,
'C': C, 'D': D, 'Dbar':Dbar, 'S': S, 'Sbar': Sbar,
'timeresmodel': timeresmodel,
'acceptance': acceptance,
'timeerrpdf': timeerrpdf,
'mistagpdf': mistagpdf,
'mistagobs': mistagobs,
'kfactorpdf': kfactorpdf,
'kvar': kvar,
'aprod': aprod,
'adet': adet,
'atageff': atageff
}
print 'buildBDecayTimePdf('
for kw in kwargs:
print '\t%s = %s' % (kw, kwargs[kw])
print '\t)'
print 72 * '#'
# constants used
zero = WS(ws, RooConstVar('zero', 'zero', 0.))
one = WS(ws, RooConstVar('one', 'one', 1.))
if None == aprod: aprod = zero
if None == adet: adet = zero
if None == atageff: atageff = [ zero ]
if None == mistagpdf:
mistagobs = None
else: # None != mistagpdf
if None == mistagobs:
raise NameError('mistag pdf set, but no mistag observable given')
# if no time resolution model is set, fake one
if timeresmodel == None:
timeresmodel = WS(ws, RooTruthModel('%s_TimeResModel' % name,
'%s time resolution model' % name, time))
elif timeresmodel == 'Gaussian':
timeresmodel = WS(ws, RooGaussModel('%s_TimeResModel' % name,
'%s time resolution model' % name, time, zero, timeerr))
# apply acceptance (if needed)
timeresmodel = applyBinnedAcceptance(
config, ws, time, timeresmodel, acceptance)
if config['UseKFactor']:
timeresmodel = applyKFactorSmearing(config, ws, time, timeresmodel,
kvar, kfactorpdf, [ Gamma, DeltaGamma, DeltaM ])
if config['ParameteriseIntegral']:
parameteriseResModelIntegrals(config, ws, timeerrpdf, timeerr, timeresmodel)
# if there is a per-event mistag distributions and we need to do things
# correctly
if None != mistagpdf:
otherargs = [ mistagobs, RooArgList(*mistagpdf), RooArgList(*tageff) ]
else:
otherargs = [ RooArgList(*tageff) ]
bcalib = RooArgList()
bbarcalib = RooArgList()
for t in mistag:
bcalib.add(t[0])
if len(t) > 1:
bbarcalib.add(t[1])
otherargs.append(bcalib)
if (bbarcalib.getSize()):
otherargs.append(bbarcalib)
otherargs += [ aprod, adet, RooArgList(*atageff) ]
flag = 0
if 'Bs2DsK' == name and 'CADDADS' == config['Bs2DsKCPObs']:
flag = DecRateCoeff.AvgDelta
# build coefficients to go into RooBDecay
cosh = WS(ws, DecRateCoeff('%s_cosh' % name, '%s_cosh' % name,
DecRateCoeff.CPEven, qf, qt, one, one, *otherargs))
sinh = WS(ws, DecRateCoeff('%s_sinh' % name, '%s_sinh' % name,
flag | DecRateCoeff.CPEven, qf, qt, D, Dbar, *otherargs))
cos = WS(ws, DecRateCoeff('%s_cos' % name, '%s_cos' % name,
DecRateCoeff.CPOdd, qf, qt, C, C, *otherargs))
sin = WS(ws, DecRateCoeff('%s_sin' % name, '%s_sin' % name,
flag | DecRateCoeff.CPOdd | DecRateCoeff.Minus,
qf, qt, S, Sbar, *otherargs))
del flag
del otherargs
# build (raw) time pdf
tau = WS(ws, Inverse('%sTau' % Gamma.GetName(),
'%s #tau' % Gamma.GetName(), Gamma))
retVal = WS(ws, RooBDecay(
'%s_RawTimePdf' % name, '%s raw time pdf' % name,
time, tau, DeltaGamma, cosh, sinh, cos, sin,
DeltaM, timeresmodel, RooBDecay.SingleSided))
retVal = applyDecayTimeErrPdf(config, name, ws, time, timeerr, qt, qf,
mistagobs, retVal, timeerrpdf, mistagpdf)
# if we do not bin the acceptance, we apply it here
retVal = applyUnbinnedAcceptance(config, name, ws, retVal, acceptance)
retVal.SetNameTitle('%s_TimePdf' % name, '%s full time pdf' % name)
# return the copy of retVal which is inside the workspace
return WS(ws, retVal)
def readDataSet(
config, # configuration dictionary
ws, # workspace to which to add data set
observables, # observables
rangeName = None # name of range to clip dataset to
):
"""
read data set from given Ntuple (or a RooDataSet inside a workspace) into
a RooDataSet
arguments:
config -- configuration dictionary (see below for relevant keys)
ws -- workspace into which to import data from tuple
observables -- RooArgSet containing the observables to be read
rangeName -- optional, can be the name of a range of one observable, if
the data read from the tuple needs to be explicitly clipped
to that range for some reason
the routine returns the data set that has been read in and stored inside
ws.
relevant configuration dictionary keys:
'DataFileName' -- file name of data file from which to read ntuple or data
set
'DataSetNames' -- name (TTree or RooDataSet) of the data set to be read;
more than one can be given in a dictionary, providing a
mapping between the sample name and the data set to be
read (see below for an explanation)
'DataWorkSpaceName'
-- name of workspace to read data from (if any - leave None
for reading tuples)
'DataSetCuts' -- cuts to apply to data sets on import - anything that
RooDataSet.reduce will understand is permissible here
(set to None to not apply any cuts on import)
'DataSetVarNameMapping'
-- mapping from variable names in set of observables to
what these variable names are called in the
tuple/workspace to be imported
"special" observable names:
The routine treats some variable names special on import based on their
likely meaning:
'weight' -- this variable name must be used to read in (s)weighted events
'qf' -- final state charge (e.g. +1 for K+ vs -1 for K-); only the
sign is important here, and the import code enforces that
'qt' -- tagging decision; this can be any integer number (positive or
negative); if the tuple should contain a float/double with
that information, it is rounded appropriately
'mistag' -- predicted mistag; the import code makes sure that events with
qt == 0 have mistag = 0.5
'sample' -- if more than one final state is studied (e.g. Ds final states
phipi, kstk, nonres, kpipi, pipipi), the events for these
subsamples often reside in different samples; therefore the
config dictionary entry 'DataSetNames' can contain a
dictionary which maps the category labels (phipi etc) to the
names of the data samples in the ROOT file/workspace
The config dict key 'DataSetVarNameMapping' contains a useful feature:
Instead of providing a one-to-one-mapping of observables to tuple/data set
names, an observable can be calculated from more than one tuple column.
This is useful e.g. to convert a tuple that's stored the tagging decision
as untagged/mixed/unmixed, or to sum up sweights for the different
samples. Most simple formulae should be supported, but constants in
scientific notation (1.0E+00) are not for now (until someone write a
better parser for this).
Example:
@code
seed = 42 # it's easy to modify the filename depending on the seed number
configdict = {
# file to read from
'DataFileName': '/some/path/to/file/with/toy_%04d.root' % seed,
# data set is in a workspace already
'DataWorkSpaceName': 'FitMeToolWS',
# name of data set inside workspace
'DataSetNames': 'combData',
# mapping between observables and variable name in data set
'DataSetVarNameMapping': {
'sample': 'sample',
'mass': 'lab0_MassFitConsD_M',
'pidk': 'lab1_PIDK',
'dsmass': 'lab2_MM',
'time': 'lab0_LifetimeFit_ctau',
'timeerr': 'lab0_LifetimeFit_ctauErr',
'mistag': 'tagOmegaComb',
'qf': 'lab1_ID',
'qt': ' tagDecComb',
# sweights need to be combined from different branches in this
# case, only one of the branches is ever set to a non-zero value,
# depending on which subsample the event is in
'weight': ('nSig_both_nonres_Evts_sw+nSig_both_phipi_Evts_sw+'
'nSig_both_kstk_Evts_sw+nSig_both_kpipi_Evts_sw+'
'nSig_both_pipipi_Evts_sw')
}
}
# define all observables somewhere, and put the into a RooArgSet called obs
# import observables in to a workspace saved in ws
# now read the data set
data = readDataSet(configdict, ws, observables)
@endcode
"""
from ROOT import ( TFile, RooWorkspace, RooRealVar, RooCategory,
RooBinningCategory, RooUniformBinning, RooMappedCategory,
RooDataSet, RooArgSet, RooArgList )
import sys, math
# local little helper routine
def round_to_even(x):
xfl = int(math.floor(x))
rem = x - xfl
if rem < 0.5: return xfl
elif rem > 0.5: return xfl + 1
else:
if xfl % 2: return xfl + 1
else: return xfl
# another small helper routine
def tokenize(s, delims = '+-*/()?:'):
# FIXME: this goes wrong for numerical constants like 1.4e-3
# proposed solution: regexp for general floating point constants,
# replace occurences of matches with empty string
delims = [ c for c in delims ]
delims.insert(0, None)
for delim in delims:
tmp = s.split(delim)
tmp = list(set(( s + ' ' for s in tmp)))
s = ''.join(tmp)
tmp = list(set(s.split(None)))
return tmp
# figure out which names from the mapping we need - look at the observables
names = ()
for n in config['DataSetVarNameMapping'].keys():
if None != observables.find(n):
names += (n,)
# build RooArgSets and maps with source and destination variables
dmap = { }
for k in names: dmap[k] = observables.find(k)
if None in dmap.values():
raise NameError('Some variables not found in destination: %s' % str(dmap))
dset = RooArgSet()
for v in dmap.values(): dset.add(v)
if None != dset.find('weight'):
# RooFit insists on weight variable being first in set
tmpset = RooArgSet()
tmpset.add(dset.find('weight'))
it = dset.fwdIterator()
while True:
obj = it.next()
if None == obj: break
if 'weight' == obj.GetName(): continue
tmpset.add(obj)
dset = tmpset
del tmpset
ddata = RooDataSet('agglomeration', 'of positronic circuits', dset, 'weight')
else:
ddata = RooDataSet('agglomeration', 'of positronic circuits', dset)
# open file with data sets
f = TFile(config['DataFileName'], 'READ')
# get workspace
fws = f.Get(config['DataWorkSpaceName'])
ROOT.SetOwnership(fws, True)
if None == fws or not fws.InheritsFrom('RooWorkspace'):
# ok, no workspace, so try to read a tree of the same name and
# synthesize a workspace
from ROOT import RooWorkspace, RooDataSet, RooArgList
fws = RooWorkspace(config['DataWorkSpaceName'])
iset = RooArgSet()
addiset = RooArgList()
it = observables.fwdIterator()
while True:
obj = it.next()
if None == obj: break
name = config['DataSetVarNameMapping'][obj.GetName()]
vnames = tokenize(name)
if len(vnames) > 1 and not obj.InheritsFrom('RooAbsReal'):
print 'Error: Formulae not supported for categories'
return None
if obj.InheritsFrom('RooAbsReal'):
if 1 == len(vnames):
# simple case, just add variable
var = WS(fws, RooRealVar(name, name, -sys.float_info.max,
sys.float_info.max))
iset.addClone(var)
else:
# complicated case - add a bunch of observables, and
# compute something in a RooFormulaVar
from ROOT import RooFormulaVar
args = RooArgList()
for n in vnames:
try:
# skip simple numerical factors
float(n)
except:
var = iset.find(n)
if None == var:
var = WS(fws, RooRealVar(n, n, -sys.float_info.max,
sys.float_info.max))
iset.addClone(var)
args.add(iset.find(n))
var = WS(fws, RooFormulaVar(name, name, name, args))
addiset.addClone(var)
else:
for dsname in ((config['DataSetNames'], )
if type(config['DataSetNames']) == str else
config['DataSetNames']):
break
leaf = f.Get(dsname).GetLeaf(name)
if None == leaf:
leaf = f.Get(dsname).GetLeaf(name + '_idx')
if leaf.GetTypeName() in (
'char', 'unsigned char', 'Char_t', 'UChar_t',
'short', 'unsigned short', 'Short_t', 'UShort_t',
'int', 'unsigned', 'unsigned int', 'Int_t', 'UInt_t',
'long', 'unsigned long', 'Long_t', 'ULong_t',
'Long64_t', 'ULong64_t', 'long long',
'unsigned long long'):
var = WS(fws, RooCategory(name, name))
tit = obj.typeIterator()
ROOT.SetOwnership(tit, True)
while True:
tobj = tit.Next()
if None == tobj: break
var.defineType(tobj.GetName(), tobj.getVal())
else:
var = WS(fws, RooRealVar(name, name, -sys.float_info.max,
sys.float_info.max))
iset.addClone(var)
for dsname in ((config['DataSetNames'], )
if type(config['DataSetNames']) == str else
config['DataSetNames']):
tmpds = WS(fws, RooDataSet(dsname, dsname,f.Get(dsname), iset), [])
if 0 != addiset.getSize():
# need to add columns with RooFormulaVars
tmpds.addColumns(addiset)
del tmpds
# local data conversion routine
def doIt(config, rangeName, dsname, sname, names, dmap, dset, ddata, fws):
sdata = fws.obj(dsname)
if None == sdata: return 0
if None != config['DataSetCuts']:
# apply any user-supplied cuts
newsdata = sdata.reduce(config['DataSetCuts'])
ROOT.SetOwnership(newsdata, True)
del sdata
sdata = newsdata
del newsdata
sset = sdata.get()
smap = { }
for k in names:
smap[k] = sset.find(config['DataSetVarNameMapping'][k])
if 'sample' in smap.keys() and None == smap['sample'] and None != sname:
smap.pop('sample')
dmap['sample'].setLabel(sname)
if None in smap.values():
raise NameError('Some variables not found in source: %s' % str(smap))
# # additional complication: toys save decay time in ps, data is in nm
# # figure out which time conversion factor to use
# timeConvFactor = 1e9 / 2.99792458e8
# meantime = sdata.mean(smap['time'])
# if ((dmap['time'].getMin() <= meantime and
# meantime <= dmap['time'].getMax() and config['IsToy']) or
# not config['IsToy']):
# timeConvFactor = 1.
# print 'DEBUG: Importing data sample meantime = %f, timeConvFactor = %f' % (
# meantime, timeConvFactor)
timeConvFactor = 1.
# loop over all entries of data set
ninwindow = 0
if None != sname:
sys.stdout.write('Dataset conversion and fixup: %s: progress: ' % sname)
else:
sys.stdout.write('Dataset conversion and fixup: progress: ')
for i in xrange(0, sdata.numEntries()):
sdata.get(i)
if 0 == i % 128:
sys.stdout.write('*')
vals = { }
for vname in smap.keys():
obj = smap[vname]
if obj.InheritsFrom('RooAbsReal'):
val = obj.getVal()
vals[vname] = val
else:
val = obj.getIndex()
vals[vname] = val
# first fixup: apply time/timeerr conversion factor
if 'time' in dmap.keys():
vals['time'] *= timeConvFactor
if 'timeerr' in dmap.keys():
vals['timeerr'] *= timeConvFactor
# second fixup: only sign of qf is important
if 'qf' in dmap.keys():
vals['qf'] = 1 if vals['qf'] > 0.5 else (-1 if vals['qf'] <
-0.5 else 0.)
# third fixup: untagged events are forced to 0.5 mistag
if ('qt' in dmap.keys() and 'mistag' in dmap.keys() and 0 ==
vals['qt']):
vals['mistag'] = 0.5
# apply cuts
inrange = True
for vname in dmap.keys():
if not dmap[vname].InheritsFrom('RooAbsReal'): continue
# no need to cut on untagged events
if 'mistag' == vname and 0 == vals['qt']: continue
if None != rangeName and dmap[vname].hasRange(rangeName):
if (dmap[vname].getMin(rangeName) > vals[vname] or
vals[vname] >= dmap[vname].getMax(rangeName)):
inrange = False
break
else:
if (dmap[vname].getMin() > vals[vname] or
vals[vname] >= dmap[vname].getMax()):
inrange = False
break
# skip cuts which are not within the allowed range
if not inrange: continue
# copy values over, doing real-category conversions as needed
for vname in smap.keys():
dvar, svar = dmap[vname], vals[vname]
if dvar.InheritsFrom('RooAbsRealLValue'):
if float == type(svar): dvar.setVal(svar)
elif int == type(svar): dvar.setVal(svar)
elif dvar.InheritsFrom('RooAbsCategoryLValue'):
if int == type(svar): dvar.setIndex(svar)
elif float == type(svar):
dvar.setIndex(round_to_even(svar))
if 'weight' in dmap:
ddata.add(dset, vals['weight'])
else:
ddata.add(dset)
ninwindow = ninwindow + 1
del sdata
sys.stdout.write(', done - %d events\n' % ninwindow)
return ninwindow
ninwindow = 0
if type(config['DataSetNames']) == str:
ninwindow += doIt(config, rangeName, config['DataSetNames'],
None, names, dmap, dset, ddata, fws)
else:
for sname in config['DataSetNames'].keys():
ninwindow += doIt(config, rangeName, config['DataSetNames'][sname],
sname, names, dmap, dset, ddata, fws)
# free workspace and close file
del fws
f.Close()
del f
# put the new dataset into our proper workspace
ddata = WS(ws, ddata, [])
# for debugging
if config['Debug']:
ddata.Print('v')
if 'qt' in dmap.keys():
data.table(dmap['qt']).Print('v')
if 'qf' in dmap.keys():
data.table(dmap['qf']).Print('v')
if 'qf' in dmap.keys() and 'qt' in dmap.keys():
data.table(RooArgSet(dmap['qt'], dmap['qf'])).Print('v')
if 'sample' in dmap.keys():
data.table(dmap['sample']).Print('v')
# all done, return Data to the bridge
return ddata
def buildSplineAcceptance(
ws, # workspace into which to import
time, # time variable
pfx, # prefix to be used in names
knots, # knots
coeffs, # acceptance coefficients
floatParams = False, # float acceptance parameters
debug = False # debug printout
):
"""
build a spline acceptance function
ws -- workspace into which to import acceptance functions
time -- time observable
pfx -- prefix (mode name) from which to build object names
knots -- list of knot positions
coeffs -- spline coefficients
floatParams -- if True, spline acceptance parameters will be floated
debug -- if True, print some debugging output
returns a pair of acceptance functions, first the unnormalised one for
fitting, then the normalised one for generation
The minimum and maximum of the range of the time variable implicitly
defines the position of the first and last knot. The other knot positions
are passed in knots. Conversely, the coeffs parameter records the height
of the sline at all but the last two knot positions. The next to last knot
coefficient is fixed to 1.0, thus fixing the overall scale of the
acceptance function. The spline coefficient for the last knot is fixed by
extrapolating linearly from the two knots before; this prevents
statistical fluctuations at the low stats high lifetime end of the
spectrum to curve the spline.
"""
# build acceptance function
from copy import deepcopy
myknots = deepcopy(knots)
mycoeffs = deepcopy(coeffs)
from ROOT import (RooBinning, RooArgList, RooPolyVar, RooCubicSplineFun,
RooConstVar, RooProduct, RooRealVar)
if (len(myknots) != len(mycoeffs) or 0 >= min(len(myknots), len(mycoeffs))):
raise ValueError('ERROR: Spline knot position list and/or coefficient'
'list mismatch')
one = WS(ws, RooConstVar('one', '1', 1.0))
# create the knot binning
knotbinning = WS(ws, RooBinning(time.getMin(), time.getMax(),
'%s_knotbinning' % pfx))
for v in myknots:
knotbinning.addBoundary(v)
knotbinning.removeBoundary(time.getMin())
knotbinning.removeBoundary(time.getMax())
knotbinning.removeBoundary(time.getMin())
knotbinning.removeBoundary(time.getMax())
oldbinning, lo, hi = time.getBinning(), time.getMin(), time.getMax()
time.setBinning(knotbinning, '%s_knotbinning' % pfx)
time.setBinning(oldbinning)
time.setRange(lo, hi)
del knotbinning
del oldbinning
del lo
del hi
# create the knot coefficients
coefflist = RooArgList()
i = 0
for v in mycoeffs:
if floatParams:
coefflist.add(WS(ws, RooRealVar('%s_SplineAccCoeff%u' % (pfx, i),
'v_{%u}' % (i+1), v, 0., 3.)))
else:
coefflist.add(WS(ws, RooConstVar('%s_SplineAccCoeff%u' % (pfx, i),
'v_{%u}' % (i+1), v)))
i = i + 1
del mycoeffs
coefflist.add(one)
i = i + 1
myknots.append(time.getMax())
myknots.reverse()
fudge = (myknots[0] - myknots[1]) / (myknots[2] - myknots[1])
lastmycoeffs = RooArgList(
WS(ws, RooConstVar('%s_SplineAccCoeff%u_coeff0' % (pfx, i),
'%s_SplineAccCoeff%u_coeff0' % (pfx, i), 1. - fudge)),
WS(ws, RooConstVar('%s_SplineAccCoeff%u_coeff1' % (pfx, i),
'%s_SplineAccCoeff%u_coeff1' % (pfx, i), fudge)))
del myknots
coefflist.add(WS(ws, RooPolyVar(
'%s_SplineAccCoeff%u' % (pfx, i), 'v_{%u}' % (i+1),
coefflist.at(coefflist.getSize() - 2), lastmycoeffs)))
del i
if debug:
print 'DEBUG: Spline Coeffs: %s' % str([
coefflist.at(i).getVal() for i in xrange(0, coefflist.getSize())
])
# create the spline itself
tacc = WS(ws, RooCubicSplineFun('%s_SplineAcceptance' % pfx,
'%s_SplineAcceptance' % pfx, time, '%s_knotbinning' % pfx,
coefflist))
del lastmycoeffs
if not floatParams:
# make sure the acceptance is <= 1 for generation
m = max([coefflist.at(j).getVal() for j in
xrange(0, coefflist.getSize())])
c = WS(ws, RooConstVar('%s_SplineAccNormCoeff' % pfx,
'%s_SplineAccNormCoeff' % pfx, 0.99 / m))
tacc_norm = WS(ws, RooProduct('%s_SplineAcceptanceNormalised' % pfx,
'%s_SplineAcceptanceNormalised' % pfx, RooArgList(tacc, c)))
del c
del m
else:
tacc_norm = None # not supported when floating
del coefflist
return tacc, tacc_norm