def _normalize(self, cutflows):
if self.__normalized:
return
self.__normalized = True
for fullname, hist in self.__hists.items():
if fullname.endswith('Up') or fullname.endswith('Down'):
name, _ = fullname.rsplit('_CMS', 1)
proc = Process.get(name)
else:
proc = Process.get(fullname)
logging.debug("normalizing histogram {0}, process {1}".format(self.__name, proc))
denom = float(cutflows[proc.cutflow][-3][proc])
factor = 0. if denom == 0. else cutflows[proc.cutflow][-1][proc] / denom
hist.Scale(factor)
def _add_legend(self, config, factor):
l = Legend(0.05, 3, 0.08)
for cfg in config['backgrounds']:
bkg, color = cfg.items()[0]
l.draw_box(1001, self._eval(color), Process.get(bkg).fullname)
l.draw_box(3654, r.kBlack, "Bkg. err.", True)
# TODO add collisions
l.new_row()
for cfg in config['signals']:
sig, color = cfg.items()[0]
label = Process.get(sig).fullname
if factor != 1:
label += " (#times {0:.1f})".format(factor)
l.draw_line(2, self._eval(color), label)
return l
def _add_legend(self, config, factor):
l = Legend(0.05, 3, 0.08)
for cfg in config['backgrounds']:
bkg, color = cfg.items()[0]
l.draw_box(1001, self._eval(color), Process.get(bkg).fullname)
l.draw_box(3654, r.kBlack, "Bkg. err.", True)
# TODO add collisions
l.new_row()
for cfg in config['signals']:
sig, color = cfg.items()[0]
label = Process.get(sig).fullname
if factor != 1:
label += " (#times {0:.1f})".format(factor)
l.draw_line(2, self._eval(color), label)
return l
def cutflow(cuts, procs, relative=False, f=sys.stdout):
expanded_proc = [Process.expand(proc) for proc in procs]
cutdata = [[sum(float(cut[p]) for p in ps) for ps in expanded_proc] for cut in cuts]
if relative:
for i in xrange(1, len(cutdata)):
cutdata[-i] = [float(b) / a for a, b in zip(cutdata[-(i + 1)], cutdata[-i])]
namelength = max(len(unicode(cut)) for cut in cuts)
fieldlengths = []
for proc, subprocs in zip(procs, expanded_proc):
val = sum(cuts[0][p] for p in subprocs)
length = max(len(proc), len("{:.2f}".format(float(val))))
fieldlengths.append(length)
header = u"{{:{0}}}".format(namelength) \
+ u"".join(u" {{:{0}}}".format(fl) for fl in fieldlengths) \
+ u"\n"
format = u"{{:{0}}}".format(namelength) \
+ "".join(" {{:{0}.2f}}".format(fl) for fl in fieldlengths) \
+ "\n"
f.write(header.format("Cut", *procs))
f.write("-" * namelength + "".join(" " + "-" * fl for fl in fieldlengths) + "\n")
for cut, data in zip(cuts, cutdata):
f.write(format.format(cut, *data))
def get_event_count(cls, f, proc, category, fmt, unweighed):
p = cls.__plots['Events']
p.clear()
p.read(f, category, Process.expand(proc), fmt=fmt)
if unweighed:
return p._get_histogram(proc).GetEntries()
return p._get_histogram(proc).GetBinContent(1)
def cutflow(cuts, processes, relative=False, weighed=False, f=sys.stdout):
expanded_proc = []
procs = []
for proc in processes:
subs = [p for p in Process.expand(proc) if str(p) in cuts[0].processes()]
if len(subs) > 0:
expanded_proc.append(subs)
procs.append(proc)
cutdata = [[sum(float(cut[p]) for p in ps) for ps in expanded_proc] for cut in cuts]
if weighed:
for n, c in enumerate(reversed(cuts)):
if not isinstance(c, StaticCut):
break
ratios = [a / (b if b != 0 else 1) for a, b in zip(cutdata[-1], cutdata[-(n + 1)])]
cutdata = cutdata[:-n]
for i in xrange(3, len(cutdata)):
cutdata[i] = [a * b for a, b in zip(cutdata[i], ratios)]
if relative:
for i in xrange(1, len(cutdata)):
cutdata[-i] = [(float(b) / a if a != 0 else 0)
for a, b in zip(cutdata[-(i + 1)], cutdata[-i])]
print_cuts(cuts, procs, cutdata, expanded_proc, "Cut", f, 5 if relative else 2)
def cutflow(cuts, procs, relative=False, f=sys.stdout):
expanded_proc = [Process.expand(proc) for proc in procs]
cutdata = [[sum(float(cut[p]) for p in ps) for ps in expanded_proc]
for cut in cuts]
if relative:
for i in xrange(1, len(cutdata)):
cutdata[-i] = [
float(b) / a for a, b in zip(cutdata[-(i + 1)], cutdata[-i])
]
namelength = max(len(unicode(cut)) for cut in cuts)
fieldlengths = []
for proc, subprocs in zip(procs, expanded_proc):
val = sum(cuts[0][p] for p in subprocs)
length = max(len(proc), len("{:.2f}".format(float(val))))
fieldlengths.append(length)
header = u"{{:{0}}}".format(namelength) \
+ u"".join(u" {{:{0}}}".format(fl) for fl in fieldlengths) \
+ u"\n"
format = u"{{:{0}}}".format(namelength) \
+ "".join(" {{:{0}.2f}}".format(fl) for fl in fieldlengths) \
+ "\n"
f.write(header.format("Cut", *procs))
f.write("-" * namelength + "".join(" " + "-" * fl
for fl in fieldlengths) + "\n")
for cut, data in zip(cuts, cutdata):
f.write(format.format(cut, *data))
def _get_histogram(self, process, systematic=None):
if isinstance(process, Process):
proc = process
process = str(process)
else:
proc = Process.get(process)
if isinstance(proc, BasicProcess):
scale = 1.
if systematic and systematic.startswith('Relative'):
scale = max(0, 1 + proc.relativesys() * (1. if systematic.endswith('Up') else -1.))
systematic = None
suffix = '_' + systematic if systematic else ''
hist = self.__hists[process + suffix].Clone()
if hist.ClassName().startswith('TH1'):
lastbin = hist.GetNbinsX()
overbin = lastbin + 1
err = math.sqrt(hist.GetBinError(lastbin) ** 2 + hist.GetBinError(overbin) ** 2)
val = hist.GetBinContent(lastbin) + hist.GetBinContent(overbin)
hist.SetBinContent(lastbin, val)
hist.SetBinError(lastbin, err)
if scale != 1.:
hist.Scale(scale)
return hist
hist = None
for p in proc.subprocesses:
h = self._get_histogram(p, systematic)
if hist:
hist.Add(h, proc.factor)
else:
hist = h.Clone()
if not hist:
raise KeyError(process)
return hist
def read_inputs(config, setup):
from ttH.TauRoast.processing import Process
fn = os.path.join(config.get("indir", config["outdir"]), "ntuple.root")
signal = None
signal_weights = None
for proc, weight in sum([cfg.items() for cfg in setup['signals']], []):
for p in sum([Process.expand(proc)], []):
logging.debug('reading {}'.format(p))
d = rec2array(root2array(fn, str(p), setup['variables']))
if isinstance(weight, float) or isinstance(weight, int):
w = np.array([weight] * len(d))
else:
w = rec2array(root2array(fn, str(p), [weight])).ravel()
w *= p.cross_section / p.events
if signal is not None:
signal = np.concatenate((signal, d))
signal_weights = np.concatenate((signal_weights, w))
else:
signal = d
signal_weights = w
background = None
background_weights = None
for proc, weight in sum([cfg.items() for cfg in setup['backgrounds']], []):
for p in sum([Process.expand(proc)], []):
logging.debug('reading {}'.format(p))
d = rec2array(root2array(fn, str(p), setup['variables']))
if isinstance(weight, float) or isinstance(weight, int):
w = np.array([weight] * len(d))
else:
w = rec2array(root2array(fn, str(p), [weight])).ravel()
w *= p.cross_section / p.events
if background is not None:
background = np.concatenate((background, d))
background_weights = np.concatenate((background_weights, w))
else:
background = d
background_weights = w
factor = np.sum(signal_weights) / np.sum(background_weights)
logging.info("renormalizing background events by factor {}".format(factor))
background_weights *= factor
return signal, signal_weights, background, background_weights
def add_mva(args, config):
fn = os.path.join(config["outdir"], "ntuple.root")
for proc in set(sum((Process.expand(p) for p in config['plot'] + config['limits']), [])):
systematics = ['NA']
if args.systematics:
weights = config.get(proc.cutflow + ' weights')
systematics = config.get(proc.cutflow + ' systematics', [])
systematics = set([s for s, w in expand_systematics(systematics, weights)])
for unc in systematics:
logging.info("using systematics: " + unc)
proc.add_mva(config, fn, unc)
def _add_legend(self, factor):
legend = Legend(0.05, 4, 0.03)
if len(self.__backgrounds_present) > 0:
legend.draw_marker(20, r.kBlack, "Data")
for cfg in self._plotconfig['backgrounds']:
props = {'SetFillStyle': 1001}
props.update(cfg)
bkg = props.pop('process')
if bkg not in self.__backgrounds_present:
continue
legend.draw_box({k: self._eval(v) for (k, v) in props.items()}, Process.get(bkg).fullname, centerline=False)
if len(self.__signals_present) > 0:
legend.new_row()
for cfg in self._plotconfig['signals']:
sig, color = cfg.items()[0]
if sig not in self.__signals_present:
continue
label = Process.get(sig).fullname
if factor != 1:
label += " (#times {0:.1f})".format(factor)
legend.draw_line(2, self._eval(color), label)
legend.new_row()
return legend
def write(self, file, cutflows, category, systematics=None, procs=None, fmt="{p}_{c}_{v}"):
"""Write histograms of the plot to `file`.
Use the specified `category` and normalize histograms via the
cutflows passed by `cutflows`. Can be limited to processes
specified in `procs`. Optionally, use `systematics`, and write with
the format specified in `fmt`, where `p` is the process name, `c`
the category, and `v` the limit name of the current plot.
Systematics always get appended to the format string (with an
underscore).
"""
self._normalize(cutflows)
if systematics is None:
systematics = []
systematics = set(systematics + ['NA'])
uncertainties = []
for systematic in systematics:
if systematic == 'NA':
uncertainties.append((None, ''))
else:
uncertainties.append((systematic + 'Up', '_{}Up'.format(systematic)))
uncertainties.append((systematic + 'Down', '_{}Down'.format(systematic)))
if procs is None:
procs = [Process.get(k) for k in self.__hists if
(not k.endswith('Up')) and (not k.endswith('Down'))]
else:
procs = map(Process.get, procs)
for proc in procs:
for uncertainty, suffix in uncertainties:
histname = fmt.format(p=proc.limitname, v=self.__limitname, c=category)
histname += suffix
logging.debug("writing histogram {0}".format(histname))
try:
hist = self._get_histogram(proc, uncertainty)
file.WriteObject(hist, histname, "WriteDelete")
except KeyError:
pass
def normalize(cuts, lumi, limit=None):
weights = None
processed = None
ntuplized = None
analyzed = None
for cut in cuts:
if str(cut).lower() == "dataset processed":
processed = cut
elif str(cut).lower() == "dataset event weights":
weights = cut
elif str(cut).lower() == "ntuple":
ntuplized = cut
elif str(cut).lower() == "ntuple analyzed":
analyzed = cut
elif processed and weights and ntuplized and analyzed:
break
dsetnorm = StaticCut("Dataset norm")
luminorm = StaticCut("Luminosity norm")
for proc in cuts[-1].processes():
if str(proc).startswith("collisions") or str(proc).startswith("fakes"):
dsetnorm[proc] = cuts[-1][proc]
luminorm[proc] = cuts[-1][proc]
else:
p = Process.get(proc)
scale = processed[proc] / float(weights[proc])
if ntuplized[proc] == 0 or analyzed[proc] == 0:
logging.warning("0 event count for {}".format(proc))
fraction = 1.
elif (not limit) or analyzed[proc] < limit:
fraction = 1.
else:
fraction = analyzed[proc] / float(ntuplized[proc])
logging.warning("scaling {} by {} to compensate for partially analyzed dataset".format(proc, 1. / fraction))
dsetnorm[proc] = cuts[-1][proc] * scale / fraction
luminorm[proc] = cuts[-1][proc] * scale / fraction * lumi * p.cross_section / float(p.events)
cuts.append(dsetnorm)
cuts.append(luminorm)
def analyze(args, config):
fn = os.path.join(config["outdir"], "ntuple.root")
if args.reuse:
cutflows = split_cuts(load_cutflows(config))
else:
if os.path.exists(fn):
os.unlink(fn)
cutflows = setup_cuts(config)
for proc in set(sum((Process.expand(p) for p in config['plot'] + config['limits']), [])):
uncertainties = ['NA']
if args.systematics:
weights = config.get(proc.cutflow + ' weights')
systematics = config.get(proc.cutflow + ' systematics', [])
uncertainties = [s for s, w in expand_systematics(systematics, weights)]
for unc in uncertainties:
suffix = '' if unc == 'NA' else '_' + unc
counts, cuts, weights = cutflows[proc.cutflow + suffix]
if len(counts) > 0 and str(proc) in counts[0].processes():
continue
logging.info("using systematics: " + unc)
local_cuts = list(cuts)
for cfg in proc.additional_cuts:
local_cuts.insert(0, Cut(*cfg))
proc.analyze(config, fn, counts, local_cuts, weights, unc, args.debug_cuts)
concatenated_cutflows = Cutflows()
for name, (counts, cuts, weights) in cutflows.items():
cuts = counts + cuts + weights
normalize(cuts, config["lumi"], config.get("event limit"))
concatenated_cutflows[name] = cuts
concatenated_cutflows.save(config)
def fill(args, config):
cutflows = load_cutflows(config)
for name, cuts in cutflows.items():
normalize(cuts, config["lumi"], config.get("event limit"))
categories, definitions = get_categories(config)
atomic_processes = set(sum(map(Process.expand, config['plot'] + config['limits']), []))
limit_processes = config["limits"]
all_processes = set(Process.get(n) for n in limit_processes + config['plot']) | atomic_processes
if len(all_processes) != len(set([p.limitname for p in all_processes])):
logging.error("the limit names of the processes are not unique and will lead to collisions!")
if 'indir' in config and config['indir'] != config['outdir']:
shutil.copy(
os.path.join(config['indir'], 'cutflow.pkl'),
os.path.join(config['outdir'], 'cutflow.pkl')
)
fn = os.path.join(config.get("indir", config["outdir"]), "ntuple.root")
forest = Forest(fn)
for category, definition in zip(categories, definitions):
logging.info("filling category: " + category)
Plot.reset()
for proc in atomic_processes:
logging.info("filling process: " + str(proc))
weights = config.get(proc.cutflow + ' weights')
systematics = config.get(proc.cutflow + ' systematics', [])
uncertainties = [('NA', weights)]
if args.systematics:
uncertainties = expand_systematics(systematics, weights)
for n, (systematic, weights) in enumerate(uncertainties):
logging.info("using systematics: " + systematic)
logging.info("using weights: " + ", ".join(weights))
for p in Plot.plots():
if (not args.essential and n == 0) or p.essential():
p.fill(proc, systematic, weights, definition)
uncertainties = None
if args.systematics:
uncertainties = list(set(sum((config.get(p.cutflow + ' systematics', []) for p in atomic_processes), [])))
logging.info("writing out plots for category: " + category)
fn = os.path.join(config["outdir"], "plots.root")
with open_rootfile(fn) as f:
for p in Plot.plots():
p.write(f, cutflows, category, uncertainties,
procs=all_processes, fmt=config["histformat"])
discriminants = config.get("discriminants", [])
fn = os.path.join(config["outdir"], "limits.root")
with open_rootfile(fn) as f:
for p in Plot.plots():
if p.limitname in discriminants:
p.write(f, cutflows, category, uncertainties,
procs=limit_processes, fmt=config["histformat"])
timing = sorted(Plot.plots(), key=lambda p: p._time)
for p in timing[:10] + timing[-10:]:
logging.debug("plot filling time for {0}: {1}".format(p, p._time))
logging.info("done filling category: " + category)
del forest
def _get_histogram(self, process):
procs = Process.expand(process)
h = self.__hists[procs[0]].Clone()
for proc in procs[1:]:
h.Add(self.__hists[proc])
return h
def _get_histogram(self, process):
procs = Process.expand(process)
h = self.__hists[procs[0]].Clone()
for proc in procs[1:]:
h.Add(self.__hists[proc])
return h
