def main():
# Parse command-line arguments
args = parser.parse_args()
# Check(s)
if (not args.data) and args.subtractWZMC:
warning(
"Requesting to subtract W/Z MC from MC background which contains no contamination. Exiting."
)
return
if (not args.data) and args.subtractWZdata:
warning(
"Requesting to subtract W/Z data from MC background which contains no contamination. Exiting."
)
return
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Load data
if args.data:
files_data = glob.glob(tf.config['base_path'] + 'objdef_data_*.root')
else:
files_data = glob.glob(tf.config['base_path'] + 'objdef_MC_3610*.root')
pass
files_WZ = glob.glob(tf.config['base_path'] + 'objdef_MC_30543*.root') + \
glob.glob(tf.config['base_path'] + 'objdef_MC_30544*.root')
if len(files_data) == 0:
warning("No files found. Try to run:")
warning(" $ source getSomeData.sh")
return
data = loadData(files_data, tf.config['tree'], prefix=tf.config['prefix'])
WZ = loadData(files_WZ, tf.config['tree'], prefix=tf.config['prefix'])
info_data = loadData(files_data, tf.config['outputtree'], stop=1)
info_WZ = loadData(files_WZ, tf.config['outputtree'], stop=1)
# Scaling by cross section
xsec = loadXsec(tf.config['xsec_file'])
# Append new DSID field
data = append_fields(data, 'DSID', np.zeros((data.size, )), dtypes=int)
for idx, id in enumerate(info_data['id']):
msk = (
data['id'] == id
) # Get mask of all 'data' entries with same id, i.e. from same file
tmp_DSID = info_data['DSID'][idx] # Get DSID for this file
if not args.data:
data['weight'][msk] *= xsec[
tmp_DSID] # Scale by cross section x filter eff. for this DSID
data['DSID'][msk] = tmp_DSID # Store DSID
pass
pass
if not args.data:
data['weight'] *= tf.config[
'lumi'] # Scale all events (MC) by luminosity
pass
WZ = append_fields(WZ, 'DSID', np.zeros((WZ.size, )), dtypes=int)
for idx in info_WZ['id']:
msk = (
WZ['id'] == idx
) # Get mask of all 'data' entries with same id, i.e. from same file
tmp_DSID = info_WZ['DSID'][idx] # Get DSID for this file
WZ['weight'][msk] *= xsec[
tmp_DSID] # Scale by cross section x filter eff. for this DSID
WZ['DSID'][msk] = tmp_DSID # Store DSID
pass
WZ['weight'] *= tf.config['lumi'] # Scale all events (MC) by luminosity
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
WZ = append_fields(WZ, 'logpt', np.log(WZ['pt']))
# Transfer factor
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Pass/fail masks
msk_pass = tf.config['pass'](data)
msk_fail = ~msk_pass
msk_WZ_pass = tf.config['pass'](WZ)
msk_WZ_fail = ~msk_WZ_pass
# Transfer factor calculator instance
calc = tf.calculator(data=data,
config=tf.config,
subtract=WZ if
(args.subtractWZMC and args.data) else None)
# GBS mass bins
masses = np.linspace(100, 270, 34 + 1, endpoint=True) # GBS mass bins
# Weight and counter arrays
weights_bkg_nom = np.zeros((np.sum(msk_fail), ), dtype=float)
weights_bkg_up = np.zeros((np.sum(msk_fail), ), dtype=float)
weights_bkg_down = np.zeros((np.sum(msk_fail), ), dtype=float)
counter_bkg = np.zeros((np.sum(msk_fail), ), dtype=float)
weights_WZ_nom = np.zeros((np.sum(msk_WZ_fail), ), dtype=float)
weights_WZ_up = np.zeros((np.sum(msk_WZ_fail), ), dtype=float)
weights_WZ_down = np.zeros((np.sum(msk_WZ_fail), ), dtype=float)
counter_WZ = np.zeros((np.sum(msk_WZ_fail), ), dtype=float)
#ctemp = ap.canvas(batch=True)
for mass in masses:
print " --", mass
# Fit TF profile
calc.mass = mass
calc.fullfit()
if args.show or args.save:
calc.plot(show=args.show,
save=args.save,
prefix='plots/tf_gbs_%s_%dGeV_' %
('data' if args.data else 'MC', mass),
MC=not args.data)
# Get TF weights
w_nom, w_up, w_down = calc.fullweights(data[msk_fail])
w_WZ_nom, w_WZ_up, w_WZ_down = calc.fullweights(WZ[msk_WZ_fail])
# Compute mask for which jets to use in GBS computation
msk_gbs = ~(np.abs(data[msk_fail]['m'] - mass) < 0.2 * mass)
msk_gbs_WZ = ~(np.abs(WZ[msk_WZ_fail]['m'] - mass) < 0.2 * mass)
# Store weights and increment counter for masked jets
weights_bkg_nom[msk_gbs] += w_nom[msk_gbs]
weights_bkg_up[msk_gbs] += w_up[msk_gbs]
weights_bkg_down[msk_gbs] += w_down[msk_gbs]
counter_bkg[msk_gbs] += 1.
weights_WZ_nom[msk_gbs_WZ] += w_WZ_nom[msk_gbs_WZ]
weights_WZ_up[msk_gbs_WZ] += w_WZ_up[msk_gbs_WZ]
weights_WZ_down[msk_gbs_WZ] += w_WZ_down[msk_gbs_WZ]
counter_WZ[msk_gbs_WZ] += 1.
pass
# Take average of jets in signal regions
msk = (counter_bkg > 0)
weights_bkg_nom[msk] /= counter_bkg[msk]
weights_bkg_up[msk] /= counter_bkg[msk]
weights_bkg_down[msk] /= counter_bkg[msk]
msk = (counter_WZ > 0)
weights_WZ_nom[msk] /= counter_WZ[msk]
weights_WZ_up[msk] /= counter_WZ[msk]
weights_WZ_down[msk] /= counter_WZ[msk]
# Computing data-driven background estimate
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
check_make_dir('output')
DSID = 400000 if args.data else 400001
# Write TF-scaled failing data to file
output = ROOT.TFile(
'output/objdef_GBS{MC}_{DSID}.root'.format(
DSID=DSID, MC='' if args.data else 'MC'), 'RECREATE')
for shift, w, w_WZ in zip(
[0, 1, -1], [weights_bkg_nom, weights_bkg_up, weights_bkg_down],
[weights_WZ_nom, weights_WZ_up, weights_WZ_down]):
# -- Get branch name for current variation
var_name = 'Nominal' if shift == 0 else (
'TF_UP' if shift == 1 else 'TF_DOWN')
# -- Prepare mass- and weight vectors
vector_m = data['m'][msk_fail]
vector_w = data['weight'][msk_fail] * w
if args.subtractWZdata and args.data:
if WZ is not None and WZ.size > 0:
print " Subtracting TF-scaled W/Z MC from background estimate"
vector_m = np.concatenate((vector_m, WZ['m'][msk_WZ_fail]))
vector_w = np.concatenate(
(vector_w, -WZ['weight'][msk_WZ_fail] * w_WZ))
else:
warning(
" Could not subtract failed, TF-scale W/Z MC component")
pass
pass
# Note: Don't subtract the signal component; that's output as a separate histogram to be used in the simultaneous fit
# -- Prepare DISD and isMC vectors
vector_DSID = np.ones_like(vector_w) * DSID
vector_isMC = np.ones_like(vector_w).astype(bool)
array1 = np.array(zip(vector_m, vector_w),
dtype=[(tf.config['prefix'] + 'm', np.float64),
('weight', np.float64)])
array2 = np.array(zip(vector_DSID, vector_isMC),
dtype=[('DSID', np.uint32), ('isMC', np.bool_)])
# Mass and weight branch
print " Writing arrays to file: %s" % var_name
treename1 = tf.config['tree'].replace('NumLargeRadiusJets',
'Jet_tau21DDT').replace(
'Nominal', var_name)
make_directories('/'.join(treename1.split('/')[:-1]), fromDir=output)
tree1 = ROOT.TTree(treename1.split('/')[-1], "")
array2tree(array1, tree=tree1)
# outputTree
treename2 = tf.config['outputtree'].replace('Nominal', var_name)
make_directories('/'.join(treename2.split('/')[:-1]), fromDir=output)
tree2 = ROOT.TTree(treename2.split('/')[-1], "")
array2tree(array2, tree=tree2)
output.Write()
pass
output.Close()
# Save configuration
check_make_dir('logs')
# -- Turn numpy arrays into lists, in order to make them JSON serializable
cfg = make_serializable(tf.config)
json.dump([cfg, vars(args)],
open(
'logs/gbs_config_%s_%d.log' %
('data' if args.data else 'MC', DSID), 'w'))
return
def main():
# Parse command-line arguments
args = parser.parse_args()
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Get signal file
sig_DSID = get_signal_DSID(args.mass, tolerance=10)
if sig_DSID is None:
return
sig_file = 'objdef_MC_{DSID:6d}.root'.format(DSID=sig_DSID)
# Load data
files = glob.glob(tf.config['base_path'] + 'objdef_MC_3610*.root') + [
tf.config['base_path'] + sig_file
]
if len(files) == 0:
warning("No files found.")
return
data = loadData(files, tf.config['tree'], prefix=tf.config['prefix'])
info = loadData(files, tf.config['outputtree'], stop=1)
# Scaling by cross section
xsec = loadXsec(tf.config['xsec_file'])
# Append new DSID field # @TODO: Make more elegant?
data = append_fields(data, 'DSID', np.zeros((data.size, )), dtypes=int)
for idx in info['id']:
msk = (
data['id'] == idx
) # Get mask of all 'data' entries with same id, i.e. from same file
DSID = info['DSID'][idx] # Get DSID for this file
data['weight'][msk] *= xsec[
DSID] # Scale by cross section x filter eff. for this DSID
data['DSID'][msk] = DSID # Store DSID
pass
data['weight'] *= tf.config['lumi'] # Scale all events (MC) by luminosity
# Check output.
if data.size == 0:
warning("No data was loaded.")
return
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
# Separate out signal MC
msk_sig = (data['DSID'] == sig_DSID)
msk_data = ~msk_sig
print "DATA STATISTICS:", np.sum(data[msk_data]['weight'])
signal = data[msk_sig]
if not args.inject:
# If we're not injecting signal, explicitly remove it from the 'data' array
data = data[~msk_sig]
pass
# Toys
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if args.toys:
# Get masks
msk_pass = tf.config['pass'](data)
msk_fail = ~msk_pass
# Create histograms
if args.inject:
pdf_pass = get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_pass & ~msk_sig)
pdf_fail = get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_fail & ~msk_sig)
else:
pdf_pass = get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_pass)
pdf_fail = get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_fail)
pass
# Smooth (only leading background)
for _ in range(2):
pdf_pass.Smooth()
pdf_fail.Smooth()
pass
# Inject afterwards
if args.inject:
pdf_pass.Add(
get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_pass & msk_sig))
pdf_fail.Add(
get_histogram(data,
tf.config['params'],
tf.config['axes'],
mask=msk_fail & msk_sig))
# Create p.d.f.s
# -- Define variables
rhoDDT = ROOT.RooRealVar('rhoDDT', 'rhoDDT', tf.config['axes'][0][0],
tf.config['axes'][0][-1])
logpt = ROOT.RooRealVar('logpt', 'logpt', tf.config['axes'][1][0],
tf.config['axes'][1][-1])
rhoDDT.setBins(len(tf.config['axes'][0]) - 1)
logpt.setBins(len(tf.config['axes'][1]) - 1)
# -- Define histograms
rdh_pass = ROOT.RooDataHist('rdh_pass', 'rdh_pass',
ROOT.RooArgList(rhoDDT, logpt), pdf_pass)
rdh_fail = ROOT.RooDataHist('rdh_fail', 'rdh_fail',
ROOT.RooArgList(rhoDDT, logpt), pdf_fail)
# -- Turn histograms into pdf's
rhp_pass = ROOT.RooHistPdf('rhp_pass', 'rhp_pass',
ROOT.RooArgSet(rhoDDT, logpt), rdh_pass)
rhp_fail = ROOT.RooHistPdf('rhp_fail', 'rhp_fail',
ROOT.RooArgSet(rhoDDT, logpt), rdh_fail)
# Generate toys
mult = 1.
N_pass = int(np.sum(data['weight'][msk_pass]) * mult)
N_fail = int(np.sum(data['weight'][msk_fail]) * mult)
dtype = ['rhoDDT', 'logpt', 'tau21DDT', 'pt', 'm', 'weight']
dtype = [(var, 'f8') for var in dtype]
toys_pass = np.zeros(N_pass, dtype=dtype)
toys_fail = np.zeros(N_fail, dtype=dtype)
print "Generating toys (pass: %d, fail: %d)" % (N_pass, N_fail)
rds_pass = rhp_pass.generate(ROOT.RooArgSet(rhoDDT, logpt), N_pass,
True, False)
rds_fail = rhp_fail.generate(ROOT.RooArgSet(rhoDDT, logpt), N_fail,
True, False)
for idx in range(N_pass):
toys_pass['rhoDDT'][idx] = rds_pass.get(idx).getRealValue('rhoDDT')
toys_pass['logpt'][idx] = rds_pass.get(idx).getRealValue('logpt')
toys_pass['pt'][idx] = np.exp(toys_pass['logpt'][idx])
toys_pass['m'][idx] = np.sqrt(
np.exp(toys_pass['rhoDDT'][idx]) * toys_pass['pt'][idx] * 1.)
toys_pass['weight'][idx] = 1. / float(mult)
toys_pass['tau21DDT'][idx] = 0.
pass
for idx in range(N_fail):
toys_fail['rhoDDT'][idx] = rds_fail.get(idx).getRealValue('rhoDDT')
toys_fail['logpt'][idx] = rds_fail.get(idx).getRealValue('logpt')
toys_fail['pt'][idx] = np.exp(toys_fail['logpt'][idx])
toys_fail['m'][idx] = np.sqrt(
np.exp(toys_fail['rhoDDT'][idx]) * toys_fail['pt'][idx] * 1.)
toys_fail['weight'][idx] = 1. / float(mult)
toys_fail['tau21DDT'][idx] = 1.
pass
data = np.concatenate((toys_pass, toys_fail)) # ???
pass
# Transfer factor
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
calc = tf.calculator(data=data,
config=tf.config) # Using default configuration
calc.mass = args.mass
calc.fullfit()
# Pass/fail masks
msk_data_pass = tf.config['pass'](data)
msk_data_fail = ~msk_data_pass
msk_sig_pass = tf.config['pass'](signal)
msk_sig_fail = ~msk_sig_pass
print " -- Computing data weights"
w_nom, w_up, w_down = calc.fullweights(data[msk_data_fail])
print " -- Computing signal weights"
w_sig, _, _ = calc.fullweights(signal[msk_sig_fail])
print " -- Final fit done"
if args.show or args.save:
calc.plot(show=args.show,
save=args.save,
prefix='plots/new_signalinjection_%s%s_' %
("toys_" if args.toys else "",
"injected" if args.inject else "notinjected"))
# Performing signal injection test
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if True or args.show or args.save:
bestfit_mu = None
for mu, fit, prefit, subtract in zip([0, 1, 1, None],
[False, False, True, False],
[True, True, True, False],
[True, True, False, True]):
if not prefit:
mu = bestfit_mu[0]
pass
# Plotting
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
c = ap.canvas(num_pads=2, batch=not args.show)
p0, p1 = c.pads()
# -- Histograms: Main pad
bins = tf.config['massbins']
h_bkg = c.hist(data['m'][msk_data_fail],
bins=bins,
weights=data['weight'][msk_data_fail] * w_nom,
display=False)
h_bkg_up = c.hist(data['m'][msk_data_fail],
bins=bins,
weights=data['weight'][msk_data_fail] * w_up,
display=False)
h_bkg_down = c.hist(data['m'][msk_data_fail],
bins=bins,
weights=data['weight'][msk_data_fail] * w_down,
display=False)
h_sig = c.hist(signal['m'][msk_sig_pass],
bins=bins,
weights=signal['weight'][msk_sig_pass],
scale=mu,
display=False)
h_sfl = c.hist(signal['m'][msk_sig_fail],
bins=bins,
weights=signal['weight'][msk_sig_fail] * w_sig,
scale=mu,
display=False)
h_data = c.plot(data['m'][msk_data_pass],
bins=bins,
weights=data['weight'][msk_data_pass],
display=False)
for bin in range(1, h_bkg.GetXaxis().GetNbins() + 1):
width = float(h_bkg.GetBinWidth(bin))
h_bkg.SetBinContent(bin, h_bkg.GetBinContent(bin) / width)
h_bkg.SetBinError(bin, h_bkg.GetBinError(bin) / width)
h_bkg_up.SetBinContent(bin,
h_bkg_up.GetBinContent(bin) / width)
h_bkg_up.SetBinError(bin, h_bkg_up.GetBinError(bin) / width)
h_bkg_down.SetBinContent(bin,
h_bkg_down.GetBinContent(bin) / width)
h_bkg_down.SetBinError(bin,
h_bkg_down.GetBinError(bin) / width)
h_sig.SetBinContent(bin, h_sig.GetBinContent(bin) / width)
h_sig.SetBinError(bin, h_sig.GetBinError(bin) / width)
h_sfl.SetBinContent(bin, h_sfl.GetBinContent(bin) / width)
h_sfl.SetBinError(bin, h_sfl.GetBinError(bin) / width)
h_data.SetBinContent(bin, h_data.GetBinContent(bin) / width)
h_data.SetBinError(bin, h_data.GetBinError(bin) / width)
pass
if not fit:
h_bkg.Add(h_sfl, -1) # Subtracting signal
h_bkg_up.Add(h_sfl, -1) # --
h_bkg_down.Add(h_sfl, -1) # --
pass
c.hist(
h_bkg, option='HIST', linestyle=0, fillstyle=0, fillcolor=0
) # Staring with standard histogram, not THStack, just to get y-axis to coorperate
h_bkg = c.stack(h_bkg,
fillcolor=ROOT.kAzure + 7,
label='Background pred.')
h_sig = c.stack(h_sig,
fillcolor=ROOT.kRed - 4,
label="Z' (#mu = %s)" %
("%.0f" % mu if prefit else "%.2f #pm %.2f" %
(mu, bestfit_mu[1])))
h_sum = h_bkg
h_sum = c.hist(h_sum,
fillstyle=3245,
fillcolor=ROOT.kGray + 3,
option='E2',
label='Stat. uncert.')
h_bkg_up = c.hist(h_bkg_up,
linecolor=ROOT.kGreen + 1,
linestyle=2,
option='HIST',
label='Syst. uncert.')
h_bkg_down = c.hist(h_bkg_down,
linecolor=ROOT.kGreen + 1,
linestyle=2,
option='HIST')
h_data = c.plot(h_data, label='Pseudo-data')
c.hist(h_bkg, option='AXIS') # Re-draw axes
# -- Histograms: Ratio pad
c.ratio_plot((h_sig, h_sum), option='HIST', offset=1)
c.ratio_plot((h_sum, h_sum), option='E2')
c.ratio_plot((h_bkg_up, h_sum), option='HIST')
c.ratio_plot((h_bkg_down, h_sum), option='HIST')
c.ratio_plot((h_data, h_sum))
# -- Axis labels
c.xlabel('Large-#it{R} jet mass [GeV]')
c.ylabel('Events / GeV')
p1.ylabel('Data / Est.')
# -- Axis limits
c.ylim(1.0E+00, 1.0E+06)
p1.ylim(0.80, 1.20)
# -- Line(s)
p1.yline(1.0)
# -- Region(s)
c.region("SR", 0.8 * args.mass, 1.2 * args.mass)
# -- Text
c.text(
[
"#sqrt{s} = 13 TeV, %s fb^{-1}" % tf.config['lumi'],
"Incl. #gamma Monte Carlo",
"Photon channel",
#("Signal" if args.inject else "No signal") + " injected",
] + (["Using toys"] if args.toys else []),
qualifier='Simulation Internal')
# -- Log
c.log()
# -- Legend
c.legend()
if args.save and not fit:
c.save('plots/new_signalinjection_%s%dGeV_pm%d_%s_%s.pdf' %
("toys_" if args.toys else "", args.mass, 20.,
('prefit_mu%d' % mu if prefit else 'postfit'),
('injected' if args.inject else 'notinjected')))
if args.show and not fit: c.show()
# Fitting
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if fit:
bestfit_mu = list()
hs_save = [
h_bkg_down.Clone('h_save_down'),
h_bkg.Clone('h_save_nom'),
h_bkg_up.Clone('h_save_up'),
]
for variation in range(3):
print "Variation: " + ("Nominal" if variation == 1 else (
"Up" if variation == 0 else "Down"))
# Get correct histogram fore this variation
h_bkg_use = hs_save[variation]
# -- Define jet mass variable
mJ = ROOT.RooRealVar('mJ', 'mJ', 50, 300)
#mJ.setBins(50)
roobinning = ROOT.RooBinning(
len(tf.config['massbins']) - 1, tf.config['massbins'])
mJ.setBinning(roobinning)
# -- Define histograms
rdh_bkg = ROOT.RooDataHist('rdh_bkg', 'rdh_bkg',
ROOT.RooArgList(mJ), h_bkg_use)
rdh_sig = ROOT.RooDataHist('rdh_sig', 'rdh_sig',
ROOT.RooArgList(mJ), h_sig)
rdh_sfl = ROOT.RooDataHist('rdh_sfl', 'rdh_sfl',
ROOT.RooArgList(mJ), h_sfl)
# -- Turn histograms into pdf's
rhp_bkg = ROOT.RooHistPdf('rhp_bkg', 'rhp_bkg',
ROOT.RooArgSet(mJ), rdh_bkg)
rhp_sig = ROOT.RooHistPdf('rhp_sig', 'rhp_sig',
ROOT.RooArgSet(mJ), rdh_sig)
rhp_sfl = ROOT.RooHistPdf('rhp_sfl', 'rhp_sfl',
ROOT.RooArgSet(mJ), rdh_sfl)
# -- Define integrals as constants
n_bkg = ROOT.RooRealVar('n_bkg', 'n_bkg',
h_bkg_use.Integral())
n_sig = ROOT.RooRealVar('n_sig', 'n_sig', h_sig.Integral())
n_sfl = ROOT.RooRealVar('n_sfl', 'n_sfl', h_sfl.Integral())
# -- Define signal strength and constant(s)
mu = ROOT.RooRealVar('mu', 'mu', 1, 0, 5)
neg1 = ROOT.RooRealVar('neg1', 'neg1', -1)
# -- Define fittable normalisation factors
c_bkg = ROOT.RooFormulaVar('c_bkg', 'c_bkg', '@0',
ROOT.RooArgList(n_bkg))
c_sig = ROOT.RooFormulaVar('c_sig', 'c_sig', '@0 * @1',
ROOT.RooArgList(mu, n_sig))
c_sfl = ROOT.RooFormulaVar(
'c_sfl', 'c_sfl', '@0 * @1 * @2',
ROOT.RooArgList(neg1, mu, n_sfl))
# -- Construct combined pdf
pdf = ROOT.RooAddPdf(
'pdf', 'pdf', ROOT.RooArgList(rhp_bkg, rhp_sig,
rhp_sfl),
ROOT.RooArgList(c_bkg, c_sig, c_sfl))
# -- Construct data histogram
rdh_data = ROOT.RooDataHist('rdh_data', 'rdh_data',
ROOT.RooArgList(mJ), h_data)
# -- Fit pdf to data histogram
pdf.chi2FitTo(rdh_data, ROOT.RooLinkedList())
print "Best fit mu: %.3f +/- %.3f" % (mu.getValV(),
mu.getError())
bestfit_mu.append((mu.getValV(), mu.getError()))
pass
bestfit_mu = bestfit_mu[1][0], np.sqrt(
np.power(
abs(bestfit_mu[0][0] - bestfit_mu[2][0]) / 2., 2.) +
np.power(bestfit_mu[1][1], 2.))
pass
pass
pass
return
def main():
# Parse command-line arguments
args = parser.parse_args()
DSID = int("100%03d" % args.mass)
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Get signal file
sig_DSID = get_signal_DSID(args.mass, tolerance=10)
if sig_DSID is None:
warning("No signal file was found")
return
sig_file = 'objdef_MC_{DSID:6d}.root'.format(DSID=sig_DSID)
# Load data
files = {
'data': glob.glob(tf.config['base_path'] + 'objdef_MC_3610*.root'),
'gbs': glob.glob(tf.config['base_path'] + 'objdef_GBSMC_400001.root'),
'WZ': glob.glob(tf.config['base_path'] + 'objdef_MC_3054*.root')
}
if args.inject:
files['sig'] = glob.glob(tf.config['base_path'] + sig_file)
pass
if len(files) == 0:
warning("No files found. Try to run:")
warning(" $ source getSomeData.sh")
return
data = loadData(files['data'],
tf.config['tree'],
prefix=tf.config['prefix'])
gbs = loadData(files['gbs'],
tf.config['finaltree'],
prefix=tf.config['prefix'])
WZ = loadData(files['WZ'], tf.config['tree'], prefix=tf.config['prefix'])
if args.inject:
signal = loadData(files['sig'],
tf.config['tree'],
prefix=tf.config['prefix'])
else:
signal = None
pass
info = {
key: loadData(files[key], tf.config['outputtree'], stop=1)
for key in files
}
# Scaling by cross section
xsec = loadXsec(tf.config['xsec_file'])
# Append new DSID field
if args.inject:
signal = append_fields(signal,
'DSID',
np.zeros((signal.size, )),
dtypes=int)
for idx, id in enumerate(info['sig']['id']):
msk = (
signal['id'] == id
) # Get mask of all 'signal' entries with same id, i.e. from same file
DSID = info['sig']['DSID'][idx] # Get DSID for this file
signal['weight'][msk] *= xsec[
DSID] # Scale by cross section x filter eff. for this DSID
signal['DSID'][msk] = DSID # Store DSID
pass
signal['weight'] *= tf.config['lumi']
pass
WZ = append_fields(WZ, 'DSID', np.zeros((WZ.size, )), dtypes=int)
for idx, id in enumerate(info['WZ']['id']):
msk = (
WZ['id'] == id
) # Get mask of all 'WZ' entries with same id, i.e. from same file
DSID = info['WZ']['DSID'][idx] # Get DSID for this file
WZ['weight'][msk] *= xsec[
DSID] # Scale by cross section x filter eff. for this DSID
WZ['DSID'][msk] = DSID # Store DSID
pass
WZ['weight'] *= tf.config['lumi']
#if not args.data:
data = append_fields(data, 'DSID', np.zeros((data.size, )), dtypes=int)
for idx, id in enumerate(info['data']['id']):
msk = (
data['id'] == id
) # Get mask of all 'data' entries with same id, i.e. from same file
DSID = info['data']['DSID'][idx] # Get DSID for this file
data['weight'][msk] *= xsec[
DSID] # Scale by cross section x filter eff. for this DSID
data['DSID'][msk] = DSID # Store DSID
pass
data['weight'] *= tf.config['lumi']
#pass
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
WZ = append_fields(WZ, 'logpt', np.log(WZ['pt']))
if signal is not None:
signal = append_fields(signal, 'logpt', np.log(signal['pt']))
pass
# Inject signal into data
if args.inject:
data = np.array(np.concatenate((data, signal)), dtype=data.dtype)
pass
#if not args.data:
data = np.array(np.concatenate((data, WZ)), dtype=data.dtype)
#pass
""" @TODO: Not sure this script works for data input... But it's not used anyway. """
# Transfer factor
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Pass/fail masks
# -- Data (incl. signal)
msk_pass = tf.config['pass'](data)
msk_fail = ~msk_pass
# -- W/Z
msk_WZ_pass = tf.config['pass'](WZ)
msk_WZ_fail = ~msk_WZ_pass
# -- Signal
if args.inject:
msk_sig_pass = tf.config['pass'](signal)
msk_sig_fail = ~msk_sig_pass
pass
# Transfer factor calculator instance
calc = tf.calculator(data=data, config=tf.config, subtract=WZ)
# Nominal fit
calc.fit()
w_nom = calc.weights(data[msk_fail])
w_nom_WZ = calc.weights(WZ[msk_WZ_fail])
if args.show or args.save:
calc.plot(show=args.show,
save=args.save,
prefix='plots/globalbackground_%s_%s_' %
('injected' if args.inject else 'notinjected',
'data' if args.data else 'MC'))
# mass +/- 20% stripe fit
calc.mass = args.mass
calc.window = 0.2
calc.fit()
w_stripe = calc.weights(data[msk_fail])
w_stripe_WZ = calc.weights(WZ[msk_WZ_fail])
if args.inject:
w_stripe_sig = calc.weights(signal[msk_sig_fail])
pass
# Plotting
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
bins = np.linspace(100, 250, 30 + 1, endpoint=True)
# Setup canvas
c = ap.canvas(num_pads=2, batch=not args.show)
p0, p1 = c.pads()
# Add stacked backgrounds
h_bkg_nom = c.hist(data['m'][msk_fail],
bins=bins,
weights=data['weight'][msk_fail] * w_nom,
display=False)
h_bkg_stripe = c.hist(data['m'][msk_fail],
bins=bins,
weights=data['weight'][msk_fail] * w_stripe,
display=False)
h_WZfl_nom = c.hist(WZ['m'][msk_WZ_fail],
bins=bins,
weights=WZ['weight'][msk_WZ_fail] * w_nom_WZ,
display=False)
h_WZfl_stripe = c.hist(WZ['m'][msk_WZ_fail],
bins=bins,
weights=WZ['weight'][msk_WZ_fail] * w_stripe_WZ,
display=False)
if args.inject:
h_sig = c.hist(signal['m'][msk_sig_pass],
bins=bins,
weights=signal['weight'][msk_sig_pass],
display=False)
h_sfl = c.hist(signal['m'][msk_sig_fail],
bins=bins,
weights=signal['weight'][msk_sig_fail] * w_stripe_sig,
display=False)
pass
h_gbs = c.hist(gbs['m'], bins=bins, weights=gbs['weight'], display=False)
# -- Subtract (opt.)
if args.inject:
h_bkg_stripe.Add(h_sfl, -1)
h_gbs.Add(h_sfl, -1)
pass
h_bkg_nom.Add(h_WZfl_nom, -1)
h_bkg_stripe.Add(h_WZfl_stripe, -1)
# -- Actually draw
#if not args.data:
h_WZ = c.stack(WZ['m'][msk_WZ_pass],
bins=bins,
weights=WZ['weight'][msk_WZ_pass],
fillcolor=ROOT.kRed - 4,
label='W/Z + #gamma')
#pass
h_bkg_nom = c.stack(h_bkg_nom,
fillcolor=ROOT.kAzure + 7,
label="Bkg. (full)")
h_sum = c.getStackSum()
h_bkg_stripe.Add(h_WZ)
h_gbs.Add(h_WZ)
if args.inject:
h_sig = c.stack(h_sig,
fillcolor=ROOT.kViolet - 4,
label="Z' (%d GeV)" % args.mass)
pass
h_bkg_stripe = c.hist(h_bkg_stripe,
linecolor=ROOT.kGreen + 1,
label="Bkg. (window)") # % args.mass)
h_gbs = c.hist(h_gbs, linecolor=ROOT.kViolet + 1, label="Bkg. (GBS)")
# Draw stats. error of stacked sum
h_sum = c.hist(h_sum,
fillstyle=3245,
fillcolor=ROOT.kGray + 2,
linecolor=ROOT.kGray + 3,
label='Stats. uncert.',
option='E2')
# Add (pseudo-) data
h_data = c.plot(data['m'][msk_pass],
bins=bins,
weights=data['weight'][msk_pass],
markersize=0.8,
label='Data' if args.data else 'Pseudo-data')
# Axis limits
p1.ylim(0.8, 1.2)
c.padding(0.45)
c.log(True)
# Draw error- and ratio plots
if args.inject:
hr_sig = c.ratio_plot((h_sig, h_sum), option='HIST', offset=1)
pass
h_err = c.ratio_plot((h_sum, h_sum), option='E2')
h_ratio = c.ratio_plot((h_data, h_sum), oob=True)
h_rgbs = c.ratio_plot((h_gbs, h_sum),
linecolor=ROOT.kViolet + 1,
option='HIST ][')
h_rgbs = c.ratio_plot((h_bkg_stripe, h_sum),
linecolor=ROOT.kGreen + 1,
option='HIST ][')
# Add labels and text
c.xlabel('Signal jet mass [GeV]')
c.ylabel('Events')
p1.ylabel('Data / Nom.')
c.text([
"#sqrt{s} = 13 TeV, L = 36.1 fb^{-1}",
] + ([
"Sherpa incl. #gamma MC",
] if not args.data else []) + [
"Trimmed anti-k_{t}^{R=1.0} jets",
"ISR #gamma selection",
] + (["Signal injected"] if args.inject else []),
qualifier='%sInternal' % ("Simulation " if not args.data else ""))
# Add line(s)
p1.yline(1.0)
# Draw legend
c.legend()
c.region("SR", 0.8 * args.mass, 1.2 * args.mass)
# Save and show plot
if args.save:
c.save('plots/globalbackground_spectrum_%dGeV_%s_%s.pdf' %
(args.mass, 'injected' if args.inject else 'notinjected',
'data' if args.data else 'MC'))
if args.show: c.show()
# p0-plot
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Setup canvas
c2 = ap.canvas(batch=not args.show)
p_local = h_data.Clone('p_local')
p_global = h_data.Clone('p_global')
for bin in range(1, h_data.GetXaxis().GetNbins() + 1):
c_data = h_data.GetBinContent(bin)
e_data = h_data.GetBinError(bin)
c_loc = h_bkg_stripe.GetBinContent(bin)
e_loc = h_bkg_stripe.GetBinError(bin)
c_glb = h_gbs.GetBinContent(bin)
e_glb = e_loc # h_gbs .GetBinError (bin)
z_loc = (c_data -
c_loc) / np.sqrt(np.square(e_data) + np.square(e_loc))
z_glb = (c_data -
c_glb) / np.sqrt(np.square(e_data) +
np.square(e_glb)) if c_glb > 0 else 0
p_loc = min(ROOT.TMath.Erfc(z_loc / np.sqrt(2)), 1)
p_glb = min(ROOT.TMath.Erfc(z_glb / np.sqrt(2)), 1)
p_local.SetBinContent(bin, p_loc)
p_global.SetBinContent(bin, p_glb)
p_local.SetBinError(bin, 0)
p_global.SetBinError(bin, 0)
pass
c2.plot(p_local,
markercolor=ROOT.kGreen + 1,
linecolor=ROOT.kGreen + 1,
option='PL',
label="Local (20% window)")
c2.plot(p_global,
markercolor=ROOT.kViolet + 1,
linecolor=ROOT.kViolet + 1,
option='PL',
label="Global (GBS)")
c2.xlabel("Signal jet mass [GeV]")
c2.ylabel("p_{0}")
c2.log()
c2.ylim(1E-04, 1E+04)
for sigma in range(4):
c2.yline(ROOT.TMath.Erfc(sigma / np.sqrt(2)))
pass
c2.text([
"#sqrt{s} = 13 TeV, L = 36.1 fb^{-1}",
] + ([
"Sherpa incl. #gamma MC",
] if not args.data else []) + [
"Trimmed anti-k_{t}^{R=1.0} jets",
"ISR #gamma selection",
("Signal" if args.inject else "No signal") + " injected" +
(" at m = %d GeV" % args.mass if args.inject else ""),
],
qualifier='Simulation Internal')
c2.region("SR", 0.8 * args.mass, 1.2 * args.mass)
c2.legend()
if args.save:
c2.save('plots/globalbackground_p0_%dGeV_%s_%s.pdf' %
(args.mass, 'injected' if args.inject else 'notinjected',
'data' if args.data else 'MC'))
if args.show: c2.show()
return
def main():
# Parse command-line arguments
args = parser.parse_args()
DSID = int("1%02d%03d" %
(0 if args.window is None else args.window * 100, args.mass))
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Get signal file
sig_DSID = get_signal_DSID(args.mass)
# Load data
#files_data = glob.glob(tf.config['base_path'] + 'objdef_MC_3610*.root')
files_data = glob.glob(tf.config['base_path'] + 'objdef_data_*.root')
files_WZ = glob.glob(tf.config['base_path'] + 'objdef_MC_30543*.root') + \
glob.glob(tf.config['base_path'] + 'objdef_MC_30544*.root')
if sig_DSID is None:
if args.mass < 100.:
print "Assuming signal is W/Z"
files_sig = files_WZ
files_WZ = []
else:
files_sig = []
pass
else:
sig_file = 'objdef_MC_{DSID:6d}.root'.format(DSID=sig_DSID)
print "Using signal file: %s" % sig_file
files_sig = [tf.config['base_path'] + sig_file]
pass
if len(files_data) == 0 or (sig_DSID and len(files_sig) == 0):
warning("No files found.")
return
data = loadData(files_data, tf.config['tree'], prefix=tf.config['prefix'])
signal = loadData(files_sig, tf.config['tree'], prefix=tf.config['prefix'])
WZ = loadData(files_WZ, tf.config['tree'], prefix=tf.config['prefix'])
info_data = loadData(files_data, tf.config['outputtree'], stop=1)
info_sig = loadData(files_sig, tf.config['outputtree'], stop=1)
info_WZ = loadData(files_WZ, tf.config['outputtree'], stop=1)
# Scaling by cross section
xsec = loadXsec(tf.config['xsec_file'])
# ----------------------------------------------------
# Make more elegant!
# ----------------------------------------------------
# Append new DSID field # @TODO: Make more elegant?
#for arr, info in zip([signal, WZ], [info_sig, info_WZ]):
'''# @TEMP >>>
if data is not None:
data = append_fields(data, 'DSID', np.zeros((data.size,)), dtypes=int)
for idx in info_data['id']:
msk = (data['id'] == idx) # Get mask of all 'data' entries with same id, i.e. from same file
tmp_DSID = info_data['DSID'][idx] # Get DSID for this file
data['weight'][msk] *= xsec[tmp_DSID] # Scale by cross section x filter eff. for this DSID
data['DSID'] [msk] = tmp_DSID # Store DSID
pass
#data['weight'] *= tf.config['lumi'] # Scale all events (MC) by luminosity
pass
# @TEMP <<<'''
if signal is not None:
signal = append_fields(signal,
'DSID',
np.zeros((signal.size, )),
dtypes=int)
for idx in info_sig['id']:
msk = (
signal['id'] == idx
) # Get mask of all 'data' entries with same id, i.e. from same file
tmp_DSID = info_sig['DSID'][idx] # Get DSID for this file
signal['weight'][msk] *= xsec[
tmp_DSID] # Scale by cross section x filter eff. for this DSID
signal['DSID'][msk] = tmp_DSID # Store DSID
pass
signal['weight'] *= tf.config[
'lumi'] # Scale all events (MC) by luminosity
pass
if WZ is not None:
WZ = append_fields(WZ, 'DSID', np.zeros((WZ.size, )), dtypes=int)
for idx in info_WZ['id']:
msk = (
WZ['id'] == idx
) # Get mask of all 'data' entries with same id, i.e. from same file
tmp_DSID = info_WZ['DSID'][idx] # Get DSID for this file
WZ['weight'][msk] *= xsec[
tmp_DSID] # Scale by cross section x filter eff. for this DSID
WZ['DSID'][msk] = tmp_DSID # Store DSID
pass
# @TODO: k-factors?
WZ['weight'] *= tf.config[
'lumi'] # Scale all events (MC) by luminosity
pass
# Check output.
if data.size == 0 or ((signal is not None) and signal.size == 0):
warning("No data was loaded. Exiting.")
return
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
if signal is not None:
signal = append_fields(signal, 'logpt', np.log(signal['pt']))
pass
if WZ is not None:
WZ = append_fields(WZ, 'logpt', np.log(WZ['pt']))
pass
# Transfer factor
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Pass/fail masks
msk_data_pass = tf.config['pass'](data)
msk_data_fail = ~msk_data_pass
if signal is not None:
msk_sig_pass = tf.config['pass'](signal)
msk_sig_fail = ~msk_sig_pass
pass
if WZ is not None:
msk_WZ_pass = tf.config['pass'](WZ)
msk_WZ_fail = ~msk_WZ_pass
pass
calc = tf.calculator(data=data,
config=tf.config,
subtract=WZ if args.subtractWZMC else
None) # Using default configuration
calc.mass = args.mass
# Perform full fit
if args.window is None:
calc.fullfit()
print " -- Computing data weights"
w_nom, w_up, w_down = calc.fullweights(data[msk_data_fail])
if signal is not None:
print " -- Computing signal weights"
w_sig_nom, w_sig_up, w_sig_down = calc.fullweights(
signal[msk_sig_fail])
w_sig_pass, _, _ = calc.fullweights(signal[msk_sig_pass])
print "Scale factors for signal strength:"
print " pass / (pass + fail) = %.3e / (%.3e + %.3e) = %.4f" % (
np.sum(w_sig_pass), np.sum(w_sig_pass), np.sum(w_sig_nom),
np.sum(w_sig_pass) / (np.sum(w_sig_nom) + np.sum(w_sig_pass)))
pass
if WZ is not None:
print " -- Computing W/Z weights"
w_WZ_nom, w_WZ_up, w_WZ_down = calc.fullweights(WZ[msk_WZ_fail])
else:
w_WZ_nom, w_WZ_up, w_WZ_down = None, None, None
pass
print " -- Final fit done"
if args.show or args.save:
calc.plot(show=args.show,
save=args.save,
prefix='plots/tf_',
MC=False)
# Perform fit with manually-set window size
else:
# @TODO: - Forcing the fit to use same length scale as 20% window fit. Improve?
calc.window = 0.2
calc.fit()
theta = calc.theta()
calc.window = args.window
calc.fit(theta=theta)
print " -- Computing data weights"
w_nom = calc.weights(data[msk_data_fail])
w_up = calc.weights(data[msk_data_fail], shift=+1)
w_down = calc.weights(data[msk_data_fail], shift=-1)
if signal is not None:
print " -- Computing signal weights"
w_sig_nom = calc.weights(signal[msk_sig_fail])
w_sig_up = calc.weights(signal[msk_sig_fail], shift=+1)
w_sig_down = calc.weights(signal[msk_sig_fail], shift=-1)
pass
if WZ is not None:
print " -- Computing W/Z weights"
w_WZ_nom, = calc.weights(WZ[msk_WZ_fail])
w_WZ_up = calc.weights(WZ[msk_WZ_fail], shift=+1)
w_WZ_down = calc.weights(WZ[msk_WZ_fail], shift=-1)
else:
w_WZ_nom, w_WZ_up, w_WZ_down = None, None, None
pass
print " -- Manual fit done"
if args.show or args.save:
calc.plot(show=args.show,
save=args.save,
prefix='plots/tf_',
MC=False)
pass
# Computing data-driven background estimate
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
check_make_dir('output')
# Write TF-scaled failing data to file
if args.save:
output = ROOT.TFile(
'output/objdef_TF_{DSID:6d}.root'.format(DSID=DSID), 'RECREATE')
pass
for shift, w, w_WZ in zip([0, 1, -1], [w_nom, w_up, w_down],
[w_WZ_nom, w_WZ_up, w_WZ_down]):
# -- Get branch name for current variation
var_name = 'Nominal' if shift == 0 else (
'TF_UP' if shift == 1 else 'TF_DOWN')
# -- Prepare mass- and weight vectors
vector_m = data['m'][msk_data_fail]
vector_w = data['weight'][msk_data_fail] * w
if args.subtractWZdata:
if WZ is not None and WZ.size > 0:
print " Subtracting TF-scaled W/Z MC from background estimate"
vector_m = np.concatenate((vector_m, WZ['m'][msk_WZ_fail]))
vector_w = np.concatenate(
(vector_w, -WZ['weight'][msk_WZ_fail] * w_WZ))
else:
warning(
" Could not subtract failed, TF-scale W/Z MC component")
pass
pass
# Note: Don't subtract the signal component; that's output as a separate histogram to be used in the simultaneous fit
# -- Prepare DISD and isMC vectors
vector_DSID = np.ones_like(vector_w) * DSID
vector_isMC = np.ones_like(vector_w).astype(bool)
array1 = np.array(zip(vector_m, vector_w),
dtype=[(tf.config['prefix'] + 'm', np.float64),
('weight', np.float64)])
array2 = np.array(zip(vector_DSID, vector_isMC),
dtype=[('DSID', np.uint32), ('isMC', np.bool_)])
if args.save:
# Mass and weight branch
print " Writing arrays to file: %s" % var_name
treename1 = tf.config['tree'].replace('NumLargeRadiusJets',
'Jet_tau21DDT').replace(
'Nominal', var_name)
make_directories('/'.join(treename1.split('/')[:-1]),
fromDir=output)
tree1 = ROOT.TTree(treename1.split('/')[-1], "")
array2tree(array1, tree=tree1)
# outputTree
treename2 = tf.config['outputtree'].replace('Nominal', var_name)
make_directories('/'.join(treename2.split('/')[:-1]),
fromDir=output)
tree2 = ROOT.TTree(treename2.split('/')[-1], "")
array2tree(array2, tree=tree2)
output.Write()
pass
pass
if args.save:
output.Close()
pass
# Write TF-scaled failing signal MC to file
if signal is not None:
if args.save:
output = ROOT.TFile(
'output/objdef_TF_{DSID:6d}_signalfail.root'.format(DSID=DSID),
'RECREATE')
pass
for shift, w_sig in zip([0, 1, -1], [w_sig_nom, w_sig_up, w_sig_down]):
# -- Get branch name for current variation
var_name = 'Nominal' if shift == 0 else (
'TF_UP' if shift == 1 else 'TF_DOWN')
# -- Prepare mass- and weight vectors
vector_m = signal['m'][msk_sig_fail]
vector_w = signal['weight'][msk_sig_fail] * w_sig
# -- Prepare DISD and isMC vectors
vector_DSID = np.ones_like(vector_w) * (DSID + 1E+05)
vector_isMC = np.ones_like(vector_w).astype(bool)
array1 = np.array(zip(vector_m, vector_w),
dtype=[(tf.config['prefix'] + 'm', np.float64),
('weight', np.float64)])
array2 = np.array(zip(vector_DSID, vector_isMC),
dtype=[('DSID', np.uint32), ('isMC', np.bool_)])
if args.save:
# Mass and weight branch
print " Writing arrays to file: %s" % var_name
treename1 = tf.config['tree'].replace('NumLargeRadiusJets',
'Jet_tau21DDT').replace(
'Nominal', var_name)
make_directories('/'.join(treename1.split('/')[:-1]),
fromDir=output)
tree1 = ROOT.TTree(treename1.split('/')[-1], "")
array2tree(array1, tree=tree1)
# outputTree
treename2 = tf.config['outputtree'].replace(
'Nominal', var_name)
make_directories('/'.join(treename2.split('/')[:-1]),
fromDir=output)
tree2 = ROOT.TTree(treename2.split('/')[-1], "")
array2tree(array2, tree=tree2)
output.Write()
pass
pass
if args.save:
output.Close()
pass
pass
# Save configuration
check_make_dir('logs')
# -- Turn numpy arrays into lists, in order to make them JSON serializable
cfg = make_serializable(tf.config)
json.dump([cfg, vars(args)], open('logs/tf_config_%d.log' % DSID, 'w'))
return
def main():
# Parse command-line arguments
args = parser.parse_args()
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Load data
files = glob.glob(tf.config['base_path'] + 'objdef_MC_3610*.root')
if len(files) == 0:
warning("No files found.")
return
data = loadData(files, tf.config['tree'], prefix=tf.config['prefix'])
info = loadData(files, tf.config['outputtree'], stop=1)
# Scaling by cross section
xsec = loadXsec(tf.config['xsec_file'])
# Append new DSID field # @TODO: Make more elegant?
data = append_fields(data, 'DSID', np.zeros((data.size, )), dtypes=int)
for idx in info['id']:
msk = (
data['id'] == idx
) # Get mask of all 'data' entries with same id, i.e. from same file
DSID = info['DSID'][idx] # Get DSID for this file
data['weight'][msk] *= xsec[
DSID] # Scale by cross section x filter eff. for this DSID
data['DSID'][msk] = DSID # Store DSID
pass
data['weight'] *= tf.config['lumi'] # Scale all events (MC) by luminosity
# Check output.
if data.size == 0:
warning("No data was loaded.")
return
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
# Transfer factor
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Pass/fail masks
msk_pass = tf.config['pass'](data)
msk_fail = ~msk_pass
# Transfer factor calculator instance
calc = tf.calculator(data=data,
config=tf.config) # Using default configuration
calc.mass = args.mass
calc.window = args.window
# ... calc.partialbins, calc.emptybins, ...
calc.fit() # ...(theta=0.5)
w_nom = calc.weights(data[msk_fail])
w_up = calc.weights(data[msk_fail], shift=+1)
w_down = calc.weights(data[msk_fail], shift=-1)
if args.show or args.save:
calc.plot(show=args.show, save=args.save, prefix='plots/new_closure_')
# Comparing jet mass distrbutions (closure)
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if args.show or args.save:
c = ap.canvas(num_pads=2, batch=not args.show)
p0, p1 = c.pads()
bins = tf.config['massbins']
h_bkg = c.hist(data['m'][msk_fail],
bins=bins,
weights=data['weight'][msk_fail] * w_nom,
display=False)
h_up = c.hist(data['m'][msk_fail],
bins=bins,
weights=data['weight'][msk_fail] * w_up,
display=False)
h_down = c.hist(data['m'][msk_fail],
bins=bins,
weights=data['weight'][msk_fail] * w_down,
display=False)
h_data = c.plot(data['m'][msk_pass],
bins=bins,
weights=data['weight'][msk_pass],
display=False)
for bin in range(1, h_bkg.GetXaxis().GetNbins() + 1):
width = float(h_bkg.GetBinWidth(bin))
h_bkg.SetBinContent(bin, h_bkg.GetBinContent(bin) / width)
h_bkg.SetBinError(bin, h_bkg.GetBinError(bin) / width)
h_up.SetBinContent(bin, h_up.GetBinContent(bin) / width)
h_up.SetBinError(bin, h_up.GetBinError(bin) / width)
h_down.SetBinContent(bin, h_down.GetBinContent(bin) / width)
h_down.SetBinError(bin, h_down.GetBinError(bin) / width)
h_data.SetBinContent(bin, h_data.GetBinContent(bin) / width)
h_data.SetBinError(bin, h_data.GetBinError(bin) / width)
pass
h_bkg = c.hist(h_bkg,
fillcolor=ROOT.kAzure + 7,
label='Background est.')
h_err = c.hist(h_bkg,
fillstyle=3245,
fillcolor=ROOT.kGray + 2,
linecolor=ROOT.kGray + 3,
label='Stat. uncert.',
option='E2')
h_up = c.hist(h_up,
linecolor=ROOT.kGreen + 1,
linestyle=2,
option='HIST',
label='Syst. uncert.')
h_down = c.hist(h_down,
linecolor=ROOT.kGreen + 1,
linestyle=2,
option='HIST')
h_data = c.plot(h_data, label='Pseudo-data')
c.ratio_plot((h_err, h_bkg), option='E2')
c.ratio_plot((h_up, h_bkg), option='HIST')
c.ratio_plot((h_down, h_bkg), option='HIST')
c.ratio_plot((h_data, h_bkg))
c.xlabel('Large-#it{R} jet mass [GeV]')
c.ylabel('Events / GeV')
p1.ylabel('Data / Est.')
c.ylim(1E+00, 1E+06)
p1.ylim(0.80, 1.20)
p1.yline(1.0)
c.region("SR", 0.8 * args.mass, 1.2 * args.mass)
#for x in [args.mass * (1 - args.window), args.mass * (1 + args.window)]:
# p0.line(x, 1E+01, x, 2E+04)
# pass
#p1.xlines([args.mass * (1 - args.window), args.mass * (1 + args.window)])
c.text([
"#sqrt{s} = 13 TeV, %s fb^{-1}" % tf.config['lumi'],
"Incl. #gamma Monte Carlo",
"Photon channel",
],
qualifier='Simulation Internal')
c.log()
c.legend()
if args.save:
c.save('plots/new_closure_%dGeV_pm%d.pdf' %
(args.mass, args.window * 100.))
if args.show: c.show()
pass
return
def main ():
# Parse command-line arguments
args = parser.parse_args()
# Setup.
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
# Load data
files = glob.glob(tf.config['base_path'] + 'objdef_data_*.root')
if len(files) == 0:
warning("No files found.")
return
data = loadData(files, tf.config['tree'], prefix=tf.config['prefix'])
info = loadData(files, tf.config['outputtree'], stop=1)
# Check output.
if data.size == 0:
warning("No data was loaded. Exiting.")
return
# Compute new variables
data = append_fields(data, 'logpt', np.log(data['pt']))
# Pass/fail masks
msk_pass = tf.config['pass'](data)
msk_fail = ~msk_pass
# Validating transfer factor fit using toys
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
#for mass in [85] + list(np.linspace(100, 250, 15 + 1, endpoint=True)):
for mass in list(np.linspace(110, 250, 14 + 1, endpoint=True)):
print "-------- MASS: %d GeV" % mass
# Set up transfer factor calculator instance
calc = tf.calculator(data=data, config=tf.config, verbose=False) # Using default configuration
calc.mass = mass
calc.window = 0.2 if (args.window is None) else args.window
# Get nomnial best-fit theta
calc.fit()
theta = calc.theta()
nominal_weights = calc.weights(data[msk_fail], shift=0), \
calc.weights(data[msk_fail], shift=+1), \
calc.weights(data[msk_fail], shift=-1)
# "Throw toys" from TF profile, fit N times
calc.toysfit(N=args.N, theta=theta)
# Get weights for each toys experiment fit
toys_weights = calc.toysweights(data[msk_fail])
# Plot variations
bins = tf.config['massbins']
c = ap.canvas(num_pads=2, batch=not args.show)
# -- Nominal background(s)
hist_nom = c.hist(data[msk_fail]['m'], bins=bins, weights=nominal_weights[0], fillcolor=ROOT.kAzure + 7, label='Nominal bkg.')
h_sum = c.hist(hist_nom,
fillstyle=3245, fillcolor=ROOT.kGray + 2, linecolor=ROOT.kGray + 3, option='E2',
label='Stat. uncert.')
# -- Toys backgrounds
toys_hists = list()
for idx, weights in enumerate(toys_weights):
h = c.hist(data[msk_fail]['m'], bins=bins, weights=weights[0], fillstyle=0, linecolor=ROOT.kRed + idx % 5, linestyle = 1 + idx // 5, label='Toys %d' % (idx + 1) if idx < 5 else None)
toys_hists.append(h)
pass
# -- Nominal variations
hist_up = c.hist(data[msk_fail]['m'], bins=bins, weights=nominal_weights[1], fillstyle=0, linecolor=ROOT.kGreen, linestyle=2, label='Syst. uncert.')
hist_down = c.hist(data[msk_fail]['m'], bins=bins, weights=nominal_weights[2], fillstyle=0, linecolor=ROOT.kGreen, linestyle=2)
# -- Data
hist_data = c.plot(data[msk_pass]['m'], bins=bins, label='Data')
# -- Ratio plots
c.ratio_plot((h_sum, hist_nom), option='E2')
for idx, h in enumerate(toys_hists):
c.ratio_plot((h, hist_nom), option='HIST')
pass
c.ratio_plot((hist_up, hist_nom), option='HIST')
c.ratio_plot((hist_down, hist_nom), option='HIST')
c.ratio_plot((hist_data, hist_nom), oob=True)
# -- Decorations
c.xlabel('Large-#it{R} jet mass [GeV]')
c.ylabel('Events / 5 GeV')
c.pads()[1].ylabel('Ratio wrt. nominal')
c.pads()[1].ylim(0.8, 1.2)
c.pads()[1].yline(1.)
c.text(["#sqrt{s} = 13 TeV, L = 36.1 fb^{-1}",
"Photon channel"],
qualifier="Internal")
c.region("SR", 0.8 * mass, 1.2*mass)
c.legend()
c.log()
if args.show: c.show()
if args.save: c.save('plots/validation_%dGeV_N%d.pdf' % (mass, args.N))
pass
return