class BTagCorrector:
def __init__(self, tagger, year, workingpoint):
self._year = year
common = load('data/common.coffea')
self._wp = common['btagWPs'][tagger][year][workingpoint]
files = {
'deepflav': {
'2016': 'DeepJet_2016LegacySF_V1.csv',
'2017': 'DeepFlavour_94XSF_V4_B_F.csv',
'2018': 'DeepJet_102XSF_V1.csv'
},
'deepcsv': {
'2016': 'DeepCSV_2016LegacySF_V1.csv',
'2017': 'DeepCSV_94XSF_V5_B_F.csv',
'2018': 'DeepCSV_102XSF_V1.csv'
}
}
filename = 'data/' + files[tagger][year]
self.sf = BTagScaleFactor(filename, workingpoint)
files = {
'2016': 'btag2017.merged',
'2017': 'btag2017.merged',
'2018': 'btag2018.merged',
}
filename = 'hists/' + files[year]
btag = load(filename)
bpass = btag[tagger].integrate('dataset').integrate(
'wp', workingpoint).integrate('btag', 'pass').values()[()]
ball = btag[tagger].integrate('dataset').integrate(
'wp', workingpoint).integrate('btag').values()[()]
nom = bpass / np.maximum(ball, 1.)
self.eff = lookup_tools.dense_lookup.dense_lookup(
nom, [ax.edges() for ax in btag[tagger].axes()[3:]])
def btag_weight(self, pt, eta, flavor, tag):
abseta = abs(eta)
#https://twiki.cern.ch/twiki/bin/viewauth/CMS/BTagSFMethods#1b_Event_reweighting_using_scale
def zerotag(eff):
return (1 - eff).prod()
eff = self.eff(flavor, pt, abseta)
sf_nom = self.sf.eval('central', flavor, abseta, pt)
sf_up = self.sf.eval('up', flavor, abseta, pt)
sf_down = self.sf.eval('down', flavor, abseta, pt)
eff_data_nom = np.minimum(1., sf_nom * eff)
eff_data_up = np.minimum(1., sf_up * eff)
eff_data_down = np.minimum(1., sf_down * eff)
nom = zerotag(eff_data_nom) / zerotag(eff)
up = zerotag(eff_data_up) / zerotag(eff)
down = zerotag(eff_data_down) / zerotag(eff)
if '-1' in tag:
nom = (1 - zerotag(eff_data_nom)) / (1 - zerotag(eff))
up = (1 - zerotag(eff_data_up)) / (1 - zerotag(eff))
down = (1 - zerotag(eff_data_down)) / (1 - zerotag(eff))
return np.nan_to_num(nom), np.nan_to_num(up), np.nan_to_num(down)
def validate_btag(filename, btagtype, etamax):
btagData = ROOT.BTagCalibration(btagtype, filename)
npts = 10000
flavor = numpy.full(npts, 5)
abseta = numpy.random.uniform(0, etamax, size=npts)
pt = numpy.random.exponential(50, size=npts) + numpy.random.exponential(20, size=npts)
pt = numpy.maximum(20.1, pt)
discr = numpy.random.rand(npts)
coffea_sf = BTagScaleFactor(filename, BTagScaleFactor.RESHAPE, 'iterativefit', keep_df=True)
btagReader = ROOT.BTagCalibrationReader(ROOT.BTagEntry.OP_RESHAPING, 'central', stdvec(['up_jes', 'down_jes']))
btagReader.load(btagData, ROOT.BTagEntry.FLAV_B, 'iterativefit')
btv_sf = makesf(btagReader)
for syst in ['central', 'up_jes', 'down_jes']:
csf = coffea_sf.eval(syst, flavor, abseta, pt, discr)
bsf = btv_sf(syst, flavor, abseta, pt, discr)
print(abs(csf - bsf).max())
flavor = numpy.random.choice([0, 4, 5], size=npts)
coffea_sf = BTagScaleFactor(filename, BTagScaleFactor.TIGHT, 'comb,mujets,incl', keep_df=True)
btagReader = ROOT.BTagCalibrationReader(ROOT.BTagEntry.OP_TIGHT, 'central', stdvec(['up', 'down']))
btagReader.load(btagData, ROOT.BTagEntry.FLAV_B, 'comb')
btagReader.load(btagData, ROOT.BTagEntry.FLAV_C, 'mujets')
btagReader.load(btagData, ROOT.BTagEntry.FLAV_UDSG, 'incl')
btv_sf = makesf(btagReader)
for syst in ['central', 'up', 'down']:
csf = coffea_sf.eval(syst, flavor, abseta, pt, discr)
bsf = btv_sf(syst, flavor, abseta, pt, discr)
print(abs(csf - bsf).max())
class btag_scalefactor:
def __init__(self, year, UL=True):
self.year = year
if self.year == 2016:
pass
elif self.year == 2017:
SF_file = os.path.expandvars('$TWHOME/data/btag/DeepJet_106XUL17SF_WPonly_V2.csv')
self.btag_sf = BTagScaleFactor(SF_file, "medium", keep_df=False)
# and load the efficiencies
self.effs = {
'b': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL17_b_eff_deepJet.json")),
'c': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL17_c_eff_deepJet.json")),
'light': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL17_light_eff_deepJet.json")),
}
elif self.year == 2018 and UL:
SF_file = os.path.expandvars('$TWHOME/data/btag/DeepJet_106XUL18SF_WPonly.csv')
self.btag_sf = BTagScaleFactor(SF_file, "medium", keep_df=False)
# and load the efficiencies
self.effs = {
'b': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL18_b_eff_deepJet.json")),
'c': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL18_c_eff_deepJet.json")),
'light': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Summer20UL18_light_eff_deepJet.json")),
}
elif self.year == 2018 and not UL:
SF_file = os.path.expandvars('$TWHOME/data/btag/DeepJet_102XSF_V2.csv')
self.btag_sf = BTagScaleFactor(SF_file, "medium", keep_df=False)
# and load the efficiencies
self.effs = {
'b': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Autumn18_b_eff_deepJet.json")),
'c': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Autumn18_c_eff_deepJet.json")),
'light': Hist1D.from_json(os.path.expandvars("$TWHOME/data/btag/Autumn18_light_eff_deepJet.json")),
}
def Method1a(self, tagged, untagged, b_direction='central', c_direction='central'):
import numpy as np
'''
tagged: jet collection of tagged jets
untagged: jet collection untagged jets
effs: dictionary of the tagging efficiencies (1D yahist objects)
btag_sf: coffea b-tag SF object
'''
tagged_b = yahist_1D_lookup(self.effs['b'], tagged.pt)*(tagged.hadronFlavour==5)
tagged_c = yahist_1D_lookup(self.effs['c'], tagged.pt)*(tagged.hadronFlavour==4)
tagged_light = yahist_1D_lookup(self.effs['light'], tagged.pt)*(tagged.hadronFlavour==0)
tagged_SFs_b = self.btag_sf.eval(b_direction, tagged.hadronFlavour, abs(tagged.eta), tagged.pt )
tagged_SFs_c = self.btag_sf.eval(c_direction, tagged.hadronFlavour, abs(tagged.eta), tagged.pt )
tagged_SFs_light = self.btag_sf.eval(c_direction, tagged.hadronFlavour, abs(tagged.eta), tagged.pt )
SFs_c = ((tagged_c/tagged_c)*tagged_SFs_c)
SFs_b = ((tagged_b/tagged_b)*tagged_SFs_b)
SFs_light = ((tagged_light/tagged_light)*tagged_SFs_light)
SFs_c = np.where(np.isnan(SFs_c), 0, SFs_c)
SFs_b = np.where(np.isnan(SFs_b), 0, SFs_b)
SFs_light = np.where(np.isnan(SFs_light), 0, SFs_light)
tagged_SFs = SFs_b+SFs_c+SFs_light
untagged_b = yahist_1D_lookup(self.effs['b'], untagged.pt)*(untagged.hadronFlavour==5)
untagged_c = yahist_1D_lookup(self.effs['c'], untagged.pt)*(untagged.hadronFlavour==4)
untagged_light = yahist_1D_lookup(self.effs['light'], untagged.pt)*(untagged.hadronFlavour==0)
untagged_SFs_b = self.btag_sf.eval(b_direction, untagged.hadronFlavour, abs(untagged.eta), untagged.pt )
untagged_SFs_c = self.btag_sf.eval(c_direction, untagged.hadronFlavour, abs(untagged.eta), untagged.pt )
untagged_SFs_light = self.btag_sf.eval(c_direction, untagged.hadronFlavour, abs(untagged.eta), untagged.pt )
SFs_c = ((untagged_c/untagged_c)*untagged_SFs_c)
SFs_b = ((untagged_b/untagged_b)*untagged_SFs_b)
SFs_light = ((untagged_light/untagged_light)*untagged_SFs_light)
SFs_c = np.where(np.isnan(SFs_c), 0, SFs_c)
SFs_b = np.where(np.isnan(SFs_b), 0, SFs_b)
SFs_light = np.where(np.isnan(SFs_light), 0, SFs_light)
untagged_SFs = SFs_b+SFs_c+SFs_light
tagged_all = (tagged_b+tagged_c+tagged_light)
untagged_all = (untagged_b+untagged_c+untagged_light)
denom = ak.prod(tagged_all, axis=1) * ak.prod((1-untagged_all), axis=1)
num = ak.prod(tagged_all*tagged_SFs, axis=1) * ak.prod((1-untagged_all*untagged_SFs), axis=1)
return num/denom
class BTagCorrector:
def __init__(self, year, workingpoint):
self._year = year
self._wp = BTagEfficiency.btagWPs[year][workingpoint]
files = {
'2016': 'DeepCSV_Moriond17_B_H.csv.gz',
'2017': 'DeepCSV_94XSF_V5_B_F.csv.gz',
'2018': 'DeepCSV_102XSF_V1.csv.gz',
}
filename = os.path.join(os.path.dirname(__file__), 'data', files[year])
self.sf = BTagScaleFactor(filename, workingpoint)
files = {
'2016': 'btagQCD2017.coffea',
'2017': 'btagQCD2017.coffea',
'2018': 'btagQCD2017.coffea',
}
filename = os.path.join(os.path.dirname(__file__), 'data', files[year])
btag = util.load(filename)
bpass = btag.integrate('btag', 'pass').values()[()]
ball = btag.integrate('btag').values()[()]
nom = bpass / numpy.maximum(ball, 1.)
dn, up = hist.clopper_pearson_interval(bpass, ball)
self.eff = dense_lookup(nom, [ax.edges() for ax in btag.axes()[1:]])
self.eff_statUp = dense_lookup(up,
[ax.edges() for ax in btag.axes()[1:]])
self.eff_statDn = dense_lookup(dn,
[ax.edges() for ax in btag.axes()[1:]])
def addBtagWeight(self, weights, jets):
abseta = abs(jets.eta)
passbtag = jets.btagDeepB > self._wp
# https://twiki.cern.ch/twiki/bin/viewauth/CMS/BTagSFMethods#1a_Event_reweighting_using_scale
def combine(eff, sf):
# tagged SF = SF*eff / eff = SF
tagged_sf = sf[passbtag].prod()
# untagged SF = (1 - SF*eff) / (1 - eff)
untagged_sf = ((1 - sf * eff) / (1 - eff))[~passbtag].prod()
return tagged_sf * untagged_sf
eff_nom = self.eff(jets.hadronFlavour, jets.pt, abseta)
eff_statUp = self.eff_statUp(jets.hadronFlavour, jets.pt, abseta)
eff_statDn = self.eff_statDn(jets.hadronFlavour, jets.pt, abseta)
sf_nom = self.sf.eval('central', jets.hadronFlavour, abseta, jets.pt)
sf_systUp = self.sf.eval('up', jets.hadronFlavour, abseta, jets.pt)
sf_systDn = self.sf.eval('down', jets.hadronFlavour, abseta, jets.pt)
nom = combine(eff_nom, sf_nom)
weights.add('btagWeight',
nom,
weightUp=combine(eff_nom, sf_systUp),
weightDown=combine(eff_nom, sf_systDn))
weights.add('btagEffStat',
numpy.ones_like(nom),
weightUp=combine(eff_statUp, sf_nom) / nom,
weightDown=combine(eff_statDn, sf_nom) / nom)
for i in numpy.where((nom < 0.01) | (nom > 10)
| numpy.isnan(nom))[0][:4]:
jet = jets[i]
logger.info("Strange weight for event: %r", nom[i])
logger.info(" jet pts: %r", jet.pt)
logger.info(" jet etas: %r", jet.eta)
logger.info(" jet flavors: %r", jet.hadronFlavour)
logger.info(" jet btags: %r", jet.btagDeepB)
logger.info(" result eff: %r up %r down %r", eff_nom[i],
eff_statUp[i], eff_statDn[i])
logger.info(" result sf: %r", sf_nom[i])
return nom
def test_BTagScalefactor():
sf1 = BTagScaleFactor('tests/samples/testBTagSF.btag.csv', 'medium')
sf2 = BTagScaleFactor('tests/samples/DeepCSV_102XSF_V1.btag.csv.gz',
BTagScaleFactor.RESHAPE, 'iterativefit')
sf3 = BTagScaleFactor('tests/samples/DeepCSV_102XSF_V1.btag.csv.gz',
BTagScaleFactor.TIGHT)
sf4 = BTagScaleFactor(
'tests/samples/DeepCSV_2016LegacySF_V1_TuneCP5.btag.csv.gz',
BTagScaleFactor.RESHAPE,
'iterativefit',
keep_df=True)
# import pdb; pdb.set_trace()
counts, test_eta, test_pt = dummy_jagged_eta_pt()
test_flavor = numpy.random.choice([0, 4, 5], size=len(test_eta))
test_allb = numpy.ones_like(test_flavor) * 5
test_discr = numpy.random.rand(len(test_eta))
offsets = numpy.zeros(len(counts) + 1)
offsets[1:] = numpy.cumsum(counts)
test_jets = ak.Array(
ak.layout.ListOffsetArray64(
ak.layout.Index64(offsets),
ak.zip(
{
"pt": test_pt,
"eta": test_eta,
"flavor": test_flavor,
"btag": test_discr,
},
highlevel=False)))
expected = numpy.array([
0.93724101, 0.89943609, 1.0671185, 1.06846618, 0.94530984, 1.06645614,
0.91862676, 1.06645614, 0.94372127, 0.94505261, 1.06645614, 1.06645614,
1.06645614, 1.06645614, 0.91385676, 1.06738093, 0.89943609, 0.92593492,
1.06960044, 0.89943609, 1.06645614, 1.06645614, 0.94290361, 1.06892548,
0.92440686, 0.92046542, 1.06645614, 0.93676041, 0.93392431, 0.91694353,
0.89943609, 0.89943609, 0.89943609, 0.89943609, 0.89943609, 0.89943609,
0.89943609, 0.89943609, 0.93371251, 0.89943609, 0.89943609, 0.89943609,
0.94767034, 1.06645614, 1.0670672, 1.07136352, 0.89943609, 0.90445481,
0.89943609, 1.06645614, 0.89943609, 0.89943609, 0.93745389, 0.90949125,
0.91778825, 1.06645614, 1.06645614, 0.89943609, 0.89943609, 1.06645614,
1.06645614, 1.06645614
])
result = sf1.eval('central', test_flavor, test_eta, test_pt, test_discr)
assert numpy.allclose(result, expected)
sf1.eval('up', test_flavor, test_eta, test_pt)
sf2.eval('central', test_allb, test_eta, test_pt, test_discr)
with pytest.raises(ValueError):
sf2.eval('up', test_allb, test_eta, test_pt)
sf3.eval('central', test_flavor, test_eta, test_pt, test_discr)
sf3.eval('up', test_flavor, test_eta, test_pt)
with pytest.raises(ValueError):
sf4.eval('central', test_flavor, test_eta, test_pt, test_discr)
expected = numpy.array([
1.2185781, 1.03526095, 1.14997077, 0.91933821, 1.2185781, 1.08865945,
0.99422718, 1.01943199, 1.01025089, 1.20312875, 0.84198391, 0.91726759,
0.93501452, 1.31649974, 1.14997077, 1.02107876, 1.06150099, 1.06063444,
0.90508972, 1.20768481, 0.8484613, 0.99217259, 0.98333802, 1.31302575,
1.0104926, 1.00474285, 1.24375693, 1.20949677, 0.91714979, 0.99533782,
1.14997077, 1.02871797, 0.99619147, 0.97543142, 1.31518787, 1.30700837,
1.14997077, 0.99879282, 0.98961045, 1.14997077, 0.88343516, 0.9930647,
1.17767042, 1.14997077, 1.30594256, 0.91888068, 1.04737201, 1.03583147,
1.02833176, 0.99527427, 0.98546895, 1.14997077, 1.04815223, 1.28007547,
1.1970858, 1.12892238, 1.14997077, 1.14997077, 1.01656481, 0.84198391,
1.2996388, 1.14997077
])
result = sf4.eval('central', test_allb, test_eta, test_pt, test_discr)
assert numpy.allclose(result, expected)
sf1.eval('down', test_jets.flavor, test_jets.eta, test_jets.pt)
systs = [
"jes", "lfstats1", "lfstats2", "hfstats1", "hfstats2", "cferr1",
"cferr2", "lf", "hf"
]
#BTagScaleFactor from coffea
t0 = time.time()
sf = BTagScaleFactor(sf_file,
BTagScaleFactor.RESHAPE,
'iterativefit,iterativefit,iterativefit',
keep_df=True)
for tsys in systs:
for sdir in ["up", "down"]:
tsys_name = sdir + '_' + tsys
sf.eval(tsys_name, arr_flav[:1], arr_abs_eta[:1], arr_pt[:1],
arr_discr[:1], True)
t1 = time.time()
print("init_py", t1 - t0)
#BTagCalibrationStandalone from POG
t0 = time.time()
libhmm = LibHMuMu()
print("loading BTagCalibration")
systs_sdir = []
for sdir in ["up", "down"]:
for syst in systs:
systs_sdir += [sdir + "_" + syst]
b = BTagCalibration(libhmm, "DeepCSV", sf_file, systs_sdir)
t1 = time.time()
print("init_C", t1 - t0)
class btag_scalefactor:
def __init__(self, year):
self.year = year
if self.year == 2016:
pass
elif self.year == 2017:
pass
elif self.year == 2018:
SF_file = os.path.expandvars(
'$TWHOME/data/btag/DeepJet_102XSF_V2.csv')
self.btag_sf = BTagScaleFactor(SF_file, "medium", keep_df=False)
# and load the efficiencies
self.effs = {
'b':
Hist1D.from_json(
os.path.expandvars(
"$TWHOME/data/btag/Autumn18_b_eff_deepJet.json")),
'c':
Hist1D.from_json(
os.path.expandvars(
"$TWHOME/data/btag/Autumn18_c_eff_deepJet.json")),
'light':
Hist1D.from_json(
os.path.expandvars(
"$TWHOME/data/btag/Autumn18_light_eff_deepJet.json")),
}
def Method1a(self, tagged, untagged):
'''
tagged: jet collection of tagged jets
untagged: jet collection untagged jets
effs: dictionary of the tagging efficiencies (1D yahist objects)
btag_sf: coffea b-tag SF object
'''
tagged_b = yahist_1D_lookup(self.effs['b'],
tagged.pt) * (tagged.hadronFlavour == 5)
tagged_c = yahist_1D_lookup(self.effs['c'],
tagged.pt) * (tagged.hadronFlavour == 4)
tagged_light = yahist_1D_lookup(
self.effs['light'], tagged.pt) * (tagged.hadronFlavour == 0)
tagged_SFs = self.btag_sf.eval('central', tagged.hadronFlavour,
abs(tagged.eta), tagged.pt)
untagged_b = yahist_1D_lookup(
self.effs['b'], untagged.pt) * (untagged.hadronFlavour == 5)
untagged_c = yahist_1D_lookup(
self.effs['c'], untagged.pt) * (untagged.hadronFlavour == 4)
untagged_light = yahist_1D_lookup(
self.effs['light'], untagged.pt) * (untagged.hadronFlavour == 0)
untagged_SFs = self.btag_sf.eval('central', untagged.hadronFlavour,
abs(untagged.eta), untagged.pt)
tagged_all = (tagged_b + tagged_c + tagged_light)
untagged_all = (untagged_b + untagged_c + untagged_light)
denom = ak.prod(tagged_all, axis=1) * ak.prod(
(1 - untagged_all), axis=1)
num = ak.prod(tagged_all * tagged_SFs, axis=1) * ak.prod(
(1 - untagged_all * untagged_SFs), axis=1)
return num / denom
out_events["mask_topmunu2b"] = cms_events["mask_topmunu2b"]
out_events["mask_topmunu1b"] = cms_events["mask_topmunu1b"]
out_events["mask_wenu1b"] = cms_events["mask_wenu1b"]
out_events["mask_wenu2b"] = cms_events["mask_wenu2b"]
out_events["mask_wmunu1b"] = cms_events["mask_wmunu1b"]
out_events["mask_wmunu2b"] = cms_events["mask_wmunu2b"]
#### #### #### #### #### #### #### #### #### #### #### #### #### #### #### #### ####
#### #### the following code to read the SFs can be moved in another python file #### #### ####
#### #### #### #### #### #### #### #### #### #### #### #### #### #### #### #### ####
## btagging SFs
from coffea.btag_tools import BTagScaleFactor
btag_sf = BTagScaleFactor("data/DeepCSV_2016LegacySF_V1.csv.gz", "medium")
out_events["btagsf0"] = btag_sf.eval("central", out_events.jetflav0,
abs(out_events.jeteta0),
out_events.jetpt0)
out_events["btagsf1"] = btag_sf.eval("central", out_events.jetflav1,
abs(out_events.jeteta1),
out_events.jetpt1)
out_events["btagsf2"] = btag_sf.eval("central", out_events.jetflav2,
abs(out_events.jeteta2),
out_events.jetpt2)
out_events["btagsf3"] = btag_sf.eval("central", out_events.jetflav3,
abs(out_events.jeteta3),
out_events.jetpt3)
out_events["btagsf4"] = btag_sf.eval("central", out_events.jetflav4,
abs(out_events.jeteta4),
out_events.jetpt4)
out_events["btagsf5"] = btag_sf.eval("central", out_events.jetflav5,
abs(out_events.jeteta5),