def add_shapes(model, obs, proc, uncs, filename, hname, hname_with_systematics, include_uncertainties):
if filename not in add_shapes_rootfiles:
add_shapes_rootfiles[filename] = rootfile(filename)
rf = add_shapes_rootfiles[filename]
theta_obs = transform_name_to_theta(obs)
theta_proc = transform_name_to_theta(proc)
hname = hname.replace('$CHANNEL', obs)
hname_with_systematics = hname_with_systematics.replace('$CHANNEL', obs)
if proc == 'DATA':
hname_tmp = hname.replace('$PROCESS', 'DATA')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None:
hname_tmp = hname.replace('$PROCESS', 'data_obs')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None: raise RuntimeError, "did not find data histogram in rootfile"
model.set_data_histogram(theta_obs, histo, reset_binning = True)
return
hf = model.get_histogram_function(theta_obs, theta_proc)
assert hf is not None, "model has no process '%s' in channel '%s'" % (theta_proc, theta_obs)
assert len(hf.get_parameters())==0, "model has non-trivial shape uncertainty already"
old_nominal_histogram = hf.get_nominal_histo()
assert len(old_nominal_histogram[2])==1, "expected a counting-only histogram with only one bin"
hname = hname.replace('$PROCESS', proc)
hname_with_systematics = hname_with_systematics.replace('$PROCESS', proc)
nominal_histogram = rf.get_histogram(hname, include_uncertainties = include_uncertainties)
if utils.reldiff(sum(old_nominal_histogram[2]), sum(nominal_histogram[2])) > 0.01 and abs(sum(old_nominal_histogram[2]) - sum(nominal_histogram[2])) > 1e-4:
raise RuntimeError, "add_shapes: histogram normalisation given in datacard and from root file differ by more than >1% (and absolute difference is > 1e-4)"
hf.set_nominal_histo(nominal_histogram, reset_binning = True)
model.reset_binning(theta_obs, nominal_histogram[0], nominal_histogram[1], len(nominal_histogram[2]))
if len(uncs) == 0: return
for u in uncs:
theta_unc = transform_name_to_theta(u)
if '$DIRECTION_' in hname_with_systematics:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u)
hname_minus = hname_plus
hname_plus = hname_plus.replace('$DIRECTION_plusminus', 'plus')
hname_minus = hname_plus.replace('$DIRECTION_plusminus', 'minus')
hname_plus = hname_plus.replace('$DIRECTION_updown', 'up')
hname_minus = hname_plus.replace('$DIRECTION_updown', 'down')
else:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Up')
hname_minus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Down')
histo_plus = rf.get_histogram(hname_plus, include_uncertainties = include_uncertainties, fail_with_exception = True)
histo_minus = rf.get_histogram(hname_minus, include_uncertainties = include_uncertainties, fail_with_exception = True)
# make the rate uncertainty part of the coefficient function, i.e., normalize plus and minus histograms
# to nominal and add a lognormal uncertainty to the coefficient function:
lambda_plus = math.log(sum(histo_plus[2]) / sum(nominal_histogram[2])) * uncs[u]
lambda_minus = -math.log(sum(histo_minus[2]) / sum(nominal_histogram[2])) * uncs[u]
model.get_coeff(theta_obs, theta_proc).add_factor('exp', parameter = u, lambda_plus = lambda_plus, lambda_minus = lambda_minus)
f_plus = sum(nominal_histogram[2]) / sum(histo_plus[2])
utils.mul_list(histo_plus[2], f_plus)
f_minus = sum(nominal_histogram[2]) / sum(histo_minus[2])
utils.mul_list(histo_minus[2], f_minus)
hf.set_syst_histos(u, histo_plus, histo_minus, uncs[u])
hf.normalize_to_nominal = True
def add_shapes(model, obs, proc, uncs, filename, hname, hname_with_systematics, include_uncertainties, searchpaths = ['.'], variables = {}):
if filename not in add_shapes_rootfiles:
path = None
for s in searchpaths:
if os.path.isfile(os.path.join(s, filename)):
path = s
break
if path is None: raise RuntimeError, "did not find file '%s' in the paths %s" % (filename, str(searchpaths))
add_shapes_rootfiles[filename] = rootfile(os.path.join(path, filename))
rf = add_shapes_rootfiles[filename]
theta_obs = transform_name_to_theta(obs)
theta_proc = transform_name_to_theta(proc)
hname = hname.replace('$CHANNEL', obs)
hname_with_systematics = hname_with_systematics.replace('$CHANNEL', obs)
for varname, value in variables.iteritems():
hname = hname.replace('$%s' % varname, value)
hname_with_systematics = hname_with_systematics.replace('$%s' % varname, value)
if proc == 'DATA':
hname_tmp = hname.replace('$PROCESS', 'DATA')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None:
hname_tmp = hname.replace('$PROCESS', 'data_obs')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None:
if _verbose: print "note: did not find data histogram in %s" % rf.get_filename()
raise RuntimeError, "did not find histo"
model.set_data_histogram(theta_obs, histo, reset_binning = True)
return
hf = model.get_histogram_function(theta_obs, theta_proc)
assert hf is not None, "model has no process '%s' in channel '%s'" % (theta_proc, theta_obs)
assert len(hf.get_parameters())==0, "model has non-trivial shape uncertainty already"
old_nominal_histogram = hf.get_nominal_histo()
assert len(old_nominal_histogram[2])==1, "expected a counting-only histogram with only one bin"
hname = hname.replace('$PROCESS', proc)
hname_with_systematics = hname_with_systematics.replace('$PROCESS', proc)
nominal_histogram = rf.get_histogram(hname, include_uncertainties = include_uncertainties)
if nominal_histogram is None:
if _verbose: print "note: did not find histogram %s in %s" % (hname, rf.get_filename())
raise RuntimeError, "did not find histo"
if utils.reldiff(old_nominal_histogram.get_value_sum(), nominal_histogram.get_value_sum()) > 0.01 and abs(old_nominal_histogram.get_value_sum() - nominal_histogram.get_value_sum()) > 1e-4:
raise RuntimeError, "add_shapes: histogram normalisation given in datacard and from root file differ by more than >1% (and absolute difference is > 1e-4)"
hf.set_nominal_histo(nominal_histogram, reset_binning = True)
model.reset_binning(theta_obs, nominal_histogram[0], nominal_histogram[1], len(nominal_histogram[2]))
if len(uncs) == 0: return
for u in uncs:
theta_unc = transform_name_to_theta(u)
if '$DIRECTION_' in hname_with_systematics:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u)
hname_minus = hname_plus
hname_plus = hname_plus.replace('$DIRECTION_plusminus', 'plus')
hname_minus = hname_plus.replace('$DIRECTION_plusminus', 'minus')
hname_plus = hname_plus.replace('$DIRECTION_updown', 'up')
hname_minus = hname_plus.replace('$DIRECTION_updown', 'down')
else:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Up')
hname_minus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Down')
histo_plus = rf.get_histogram(hname_plus, include_uncertainties = include_uncertainties)
if histo_plus is None:
if _verbose: print "note: did not find histogram %s in %s" % (hname_plus, rf.get_filename())
raise RuntimeError, "did not find histo"
histo_minus = rf.get_histogram(hname_minus, include_uncertainties = include_uncertainties)
if histo_minus is None:
if _verbose: print "note: did not find histogram %s in %s" % (hname_minus, rf.get_filename())
raise RuntimeError, "did not find histo"
# make the rate uncertainty part of the coefficient function, i.e., normalize plus and minus histograms
# to nominal and add a lognormal uncertainty to the coefficient function:
lambda_plus = math.log(histo_plus.get_value_sum() / nominal_histogram.get_value_sum()) * uncs[u]
lambda_minus = -math.log(histo_minus.get_value_sum() / nominal_histogram.get_value_sum()) * uncs[u]
model.get_coeff(theta_obs, theta_proc).add_factor('exp', parameter = u, lambda_plus = lambda_plus, lambda_minus = lambda_minus)
f_plus = nominal_histogram.get_value_sum() / histo_plus.get_value_sum()
histo_plus = histo_plus.scale(f_plus)
f_minus = nominal_histogram.get_value_sum() / histo_minus.get_value_sum()
histo_minus = histo_minus.scale(f_minus)
hf.set_syst_histos(u, histo_plus, histo_minus, uncs[u])
hf.normalize_to_nominal = True
def add_shapes(model, obs, proc, uncs, filename, hname, hname_with_systematics, include_uncertainties, searchpaths = ['.'], variables = {}, rhandling = 'renormalize-lognormal'):
assert rhandling in ('renormalize-lognormal', 'morph')
if filename not in add_shapes_rootfiles:
path = None
for s in searchpaths:
if os.path.isfile(os.path.join(s, filename)):
path = s
break
if path is None: raise NotFoundException, "did not find file '%s' in the paths %s" % (filename, str(searchpaths))
add_shapes_rootfiles[filename] = rootfile(os.path.join(path, filename))
rf = add_shapes_rootfiles[filename]
theta_obs = transform_name_to_theta(obs)
theta_proc = transform_name_to_theta(proc)
hname = hname.replace('$CHANNEL', obs)
hname_with_systematics = hname_with_systematics.replace('$CHANNEL', obs)
for varname, value in variables.iteritems():
hname = hname.replace('$%s' % varname, value)
hname_with_systematics = hname_with_systematics.replace('$%s' % varname, value)
if proc == 'DATA':
hname_tmp = hname.replace('$PROCESS', 'DATA')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None:
hname_tmp = hname.replace('$PROCESS', 'data_obs')
histo = rf.get_histogram(hname_tmp, include_uncertainties = False)
if histo is None:
if _debug: print "note: did not find data histogram in %s" % rf.get_filename()
raise NotFoundException, "did not find histo"
model.set_data_histogram(theta_obs, histo, reset_binning = True)
return
hf = model.get_histogram_function(theta_obs, theta_proc)
assert hf is not None, "model has no process '%s' in channel '%s'" % (theta_proc, theta_obs)
assert len(hf.get_parameters())==0, "model has non-trivial shape uncertainty already"
old_nominal_histogram = hf.get_nominal_histo()
assert len(old_nominal_histogram[2])==1, "expected a counting-only histogram with only one bin"
hname = hname.replace('$PROCESS', proc)
hname_with_systematics = hname_with_systematics.replace('$PROCESS', proc)
nominal_histogram = rf.get_histogram(hname, include_uncertainties = include_uncertainties)
if nominal_histogram is None:
if _debug: print "note: did not find histogram %s in %s" % (hname, rf.get_filename())
raise NotFoundException, "did not find histo"
if _debug:
print "norm(%s) = %.3f" % (hname, nominal_histogram.get_value_sum())
# check that histogram in rootfile matches definition in datacard (allow deviations up to 1% / 1e-4 absolute):
nominal_is_zero = False
if old_nominal_histogram.get_value_sum() > 0.0 or nominal_histogram.get_value_sum() > 0.0:
if old_nominal_histogram.get_value_sum() != -1.0 and utils.reldiff(old_nominal_histogram.get_value_sum(), nominal_histogram.get_value_sum()) > 0.01 and abs(old_nominal_histogram.get_value_sum() - nominal_histogram.get_value_sum()) > 1e-4:
raise InconsistentDataException("add_shapes: histogram normalisation given in datacard and from root file differ by more than 1%% "
"(and absolute difference is > 1e-4) for channel %s, process %s (histogram name '%s')" % (obs, proc, hname))
else:
print "WARNING: channel '%s' process '%s': yield is <=0. Process will ALWAYS have 0 contribution; please delete it from the datacard." % (obs, proc)
nominal_is_zero = True
# even for nominal_is_zero, make sure to set the histogram to ensure that the binning is correct:
hf.set_nominal_histo(nominal_histogram, reset_binning = True)
model.reset_binning(theta_obs, nominal_histogram[0], nominal_histogram[1], len(nominal_histogram[2]))
if len(uncs) == 0: return
if nominal_is_zero: return
for u in uncs:
theta_unc = transform_name_to_theta(u)
if '$DIRECTION_' in hname_with_systematics:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u)
hname_minus = hname_plus
hname_plus = hname_plus.replace('$DIRECTION_plusminus', 'plus')
hname_minus = hname_plus.replace('$DIRECTION_plusminus', 'minus')
hname_plus = hname_plus.replace('$DIRECTION_updown', 'up')
hname_minus = hname_plus.replace('$DIRECTION_updown', 'down')
else:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Up')
hname_minus = hname_with_systematics.replace('$SYSTEMATIC', u + 'Down')
histo_plus = rf.get_histogram(hname_plus, include_uncertainties = include_uncertainties)
if histo_plus is None:
if _debug: print "note: did not find histogram %s in %s" % (hname_plus, rf.get_filename())
raise NotFoundException, "did not find histo"
histo_minus = rf.get_histogram(hname_minus, include_uncertainties = include_uncertainties)
if histo_minus is None:
if _debug: print "note: did not find histogram %s in %s" % (hname_minus, rf.get_filename())
raise NotFoundException, "did not find histo"
if _debug:
print "norm(%s) = %.3f" % (hname_plus, histo_plus.get_value_sum())
print "norm(%s) = %.3f" % (hname_minus, histo_minus.get_value_sum())
if rhandling == 'renormalize-lognormal':
# make the rate uncertainty part of the coefficient function, i.e., normalize plus and minus histograms
# to nominal and add a lognormal uncertainty to the coefficient function:
lambda_plus = math.log(histo_plus.get_value_sum() / nominal_histogram.get_value_sum()) * uncs[u]
lambda_minus = -math.log(histo_minus.get_value_sum() / nominal_histogram.get_value_sum()) * uncs[u]
model.get_coeff(theta_obs, theta_proc).add_factor('exp', parameter = u, lambda_plus = lambda_plus, lambda_minus = lambda_minus)
f_plus = nominal_histogram.get_value_sum() / histo_plus.get_value_sum()
histo_plus = histo_plus.scale(f_plus)
f_minus = nominal_histogram.get_value_sum() / histo_minus.get_value_sum()
histo_minus = histo_minus.scale(f_minus)
hf.set_syst_histos(u, histo_plus, histo_minus, uncs[u])
hf.normalize_to_nominal = True
else:
hf.set_syst_histos(u, histo_plus, histo_minus, uncs[u])
def add_shapes(model, obs, proc, uncs, filename, hname, hname_with_systematics,
include_uncertainties):
if filename not in add_shapes_rootfiles:
add_shapes_rootfiles[filename] = rootfile(filename)
rf = add_shapes_rootfiles[filename]
theta_obs = transform_name_to_theta(obs)
theta_proc = transform_name_to_theta(proc)
hname = hname.replace('$CHANNEL', obs)
hname_with_systematics = hname_with_systematics.replace('$CHANNEL', obs)
if proc == 'DATA':
hname_tmp = hname.replace('$PROCESS', 'DATA')
histo = rf.get_histogram(hname_tmp, include_uncertainties=False)
if histo is None:
hname_tmp = hname.replace('$PROCESS', 'data_obs')
histo = rf.get_histogram(hname_tmp, include_uncertainties=False)
if histo is None:
raise RuntimeError, "did not find data histogram in rootfile"
model.set_data_histogram(theta_obs, histo, reset_binning=True)
return
hf = model.get_histogram_function(theta_obs, theta_proc)
assert hf is not None, "model has no process '%s' in channel '%s'" % (
theta_proc, theta_obs)
assert len(hf.get_parameters()
) == 0, "model has non-trivial shape uncertainty already"
old_nominal_histogram = hf.get_nominal_histo()
assert len(old_nominal_histogram[2]
) == 1, "expected a counting-only histogram with only one bin"
hname = hname.replace('$PROCESS', proc)
hname_with_systematics = hname_with_systematics.replace('$PROCESS', proc)
nominal_histogram = rf.get_histogram(
hname, include_uncertainties=include_uncertainties)
if utils.reldiff(sum(old_nominal_histogram[2]), sum(
nominal_histogram[2])) > 0.01 and abs(
sum(old_nominal_histogram[2]) -
sum(nominal_histogram[2])) > 1e-4:
raise RuntimeError, "add_shapes: histogram normalisation given in datacard and from root file differ by more than >1% (and absolute difference is > 1e-4)"
hf.set_nominal_histo(nominal_histogram, reset_binning=True)
model.reset_binning(theta_obs, nominal_histogram[0], nominal_histogram[1],
len(nominal_histogram[2]))
if len(uncs) == 0: return
for u in uncs:
theta_unc = transform_name_to_theta(u)
if '$DIRECTION_' in hname_with_systematics:
hname_plus = hname_with_systematics.replace('$SYSTEMATIC', u)
hname_minus = hname_plus
hname_plus = hname_plus.replace('$DIRECTION_plusminus', 'plus')
hname_minus = hname_plus.replace('$DIRECTION_plusminus', 'minus')
hname_plus = hname_plus.replace('$DIRECTION_updown', 'up')
hname_minus = hname_plus.replace('$DIRECTION_updown', 'down')
else:
hname_plus = hname_with_systematics.replace(
'$SYSTEMATIC', u + 'Up')
hname_minus = hname_with_systematics.replace(
'$SYSTEMATIC', u + 'Down')
histo_plus = rf.get_histogram(
hname_plus,
include_uncertainties=include_uncertainties,
fail_with_exception=True)
histo_minus = rf.get_histogram(
hname_minus,
include_uncertainties=include_uncertainties,
fail_with_exception=True)
# make the rate uncertainty part of the coefficient function, i.e., normalize plus and minus histograms
# to nominal and add a lognormal uncertainty to the coefficient function:
lambda_plus = math.log(
sum(histo_plus[2]) / sum(nominal_histogram[2])) * uncs[u]
lambda_minus = -math.log(
sum(histo_minus[2]) / sum(nominal_histogram[2])) * uncs[u]
model.get_coeff(theta_obs,
theta_proc).add_factor('exp',
parameter=u,
lambda_plus=lambda_plus,
lambda_minus=lambda_minus)
f_plus = sum(nominal_histogram[2]) / sum(histo_plus[2])
utils.mul_list(histo_plus[2], f_plus)
f_minus = sum(nominal_histogram[2]) / sum(histo_minus[2])
utils.mul_list(histo_minus[2], f_minus)
hf.set_syst_histos(u, histo_plus, histo_minus, uncs[u])
hf.normalize_to_nominal = True
