def import_fitresult(inCard, fitResult, toDrop=[]):
# First convert the card and open workspace
execute_cmd(
'text2workspace.py -b %s -o morphedWorkspace.root --channel-masks --X-no-jmax'
% inCard)
w_f = ROOT.TFile.Open('morphedWorkspace.root', 'UPDATE')
w = w_f.Get('w')
# Open fit result we want to import
print('Importing %s...' % fitResult)
fr_f = ROOT.TFile.Open(fitResult)
fr = fr_f.Get('fit_b') # b-only fit result (fit_s for s+b)
myargs = fr.floatParsFinal()
outargs = ROOT.RooArgSet()
_cache = [] # to avoid seg faults
for i in range(myargs.getSize()):
var = myargs.at(i)
if var.GetName() in toDrop: continue
if not w.allVars().contains(var):
print('WARNING: Could not find %s' % var.GetName())
continue
outargs.add(var)
_cache.append(var)
w.saveSnapshot('morphedModel', outargs, True)
w_f.WriteTObject(w, 'w', "Overwrite")
w_f.Close()
def _runMLfit(cardOrW,
blinding,
verbosity,
rMin,
rMax,
setParams,
usePreviousFit=False,
extra=''):
if usePreviousFit: param_options = ''
else: param_options = '--text2workspace "--channel-masks" '
params_to_set = ','.join(['mask_%s_SIG=1' % r for r in blinding] +
['%s=%s' % (p, v)
for p, v in setParams.items()] + ['r=1'])
param_options += '--setParameters ' + params_to_set
fit_cmd = 'combine -M FitDiagnostics {card_or_w} {param_options} --saveWorkspace --cminDefaultMinimizerStrategy 0 --rMin {rmin} --rMax {rmax} -v {verbosity} {extra}'
fit_cmd = fit_cmd.format(
card_or_w='initifalFitWorkspace.root --snapshotName initialFit'
if usePreviousFit else cardOrW,
param_options=param_options,
rmin=rMin,
rmax=rMax,
verbosity=verbosity,
extra=extra)
with open('FitDiagnostics_command.txt', 'w') as out:
out.write(fit_cmd)
if os.path.isfile('fitDiagnosticsTest.root'):
execute_cmd('rm fitDiagnosticsTest.root')
execute_cmd(fit_cmd, out='FitDiagnostics.log')
def nuis_pulls():
diffnuis_cmd = 'python $CMSSW_BASE/src/HiggsAnalysis/CombinedLimit/test/diffNuisances.py fitDiagnosticsTest.root --abs -g nuisance_pulls.root'
execute_cmd(diffnuis_cmd)
# Make a PDF of the nuisance_pulls.root
if os.path.exists('nuisance_pulls.root'):
nuis_file = ROOT.TFile.Open('nuisance_pulls.root')
nuis_can = nuis_file.Get('nuisances')
nuis_can.Print('nuisance_pulls.pdf','pdf')
nuis_file.Close()
def gen_post_fit_shapes():
fit_result_file = ROOT.TFile.Open('fitDiagnosticsTest.root')
goodFitTags = _get_good_fit_results(fit_result_file)
for t in goodFitTags:
if os.path.exists('higgsCombineTest.FitDiagnostics.mH120.root'):
workspace_file = 'higgsCombineTest.FitDiagnostics.mH120.root'
else:
workspace_file = 'higgsCombineTest.FitDiagnostics.mH120.123456.root'
shapes_cmd = 'PostFit2DShapesFromWorkspace -w {w} -o postfitshapes_{t}.root -f fitDiagnosticsTest.root:fit_{t} --postfit --samples 100 --print 2> PostFitShapes2D_stderr_{t}.txt'.format(t=t,w=workspace_file)
execute_cmd(shapes_cmd)
fit_result_file.Close()
def _setupProjDir(self):
'''Create the directory structure where results will be stored.
'''
if not os.path.isdir(self.tag + '/'):
if self.options.overwrite:
execute_cmd('rm -rf ' + self.tag)
print('Making dir ' + self.tag + '/')
os.mkdir(self.tag + '/')
if self.options.plotTemplateComparisons and not os.path.isdir(
self.tag + '/UncertPlots/'):
os.mkdir(self.tag + '/UncertPlots/')
def _runLimit(blindData,
verbosity,
setParams,
card_or_w='card.txt',
condor=False):
# card_or_w could be `morphedWorkspace.root --snapshotName morphedModel`
param_options = ''
if len(setParams) > 0:
param_options = '--setParameters ' + ','.join(
'%s=%s' % (k, v) for k, v in setParams.items())
limit_cmd = 'combine -M AsymptoticLimits -d {card_or_w} --saveWorkspace --cminDefaultMinimizerStrategy 0 {param_opt} {blind_opt} -v {verb}'
limit_cmd = limit_cmd.format(card_or_w=card_or_w,
blind_opt='--run=blind' if blindData else '',
param_opt=param_options,
verb=verbosity)
# Run combine if not on condor
if not condor:
with open('Limit_command.txt', 'w') as out:
out.write(limit_cmd)
execute_cmd(limit_cmd)
return limit_cmd
def plot_signalInjection(tag, subtag, injectedAmount, seed=123456, condor=False):
with cd(tag+'/'+subtag):
if condor:
tmpdir = 'notneeded/tmp/'
execute_cmd('mkdir '+tmpdir)
execute_cmd('cat %s_%s_sigInj_r%s_output_*.tgz | tar zxvf - -i --strip-components 2 -C %s'%(tag,subtag,injectedAmount,tmpdir))
tree_fit_sb = ROOT.TChain('tree_fit_sb')
tree_fit_sb.Add(tmpdir+'fitDiagnostics_sigInj_r{rinj}*.root'.format(rinj=injectedAmount))
else:
toyOutput = ROOT.TFile.Open('fitDiagnostics_sigInj_r{rinj}_{seed}.root'.format(rinj=injectedAmount,seed=seed))
tree_fit_sb = toyOutput.Get('tree_fit_sb')
ROOT.gROOT.SetBatch(True)
ROOT.gStyle.SetOptStat(False)
# Final plotting
result_can = ROOT.TCanvas('sigpull_can','sigpull_can',800,700)
tree_fit_sb.Draw("(r-{rinj})/(rHiErr*(r<{rinj})+rLoErr*(r>{rinj}))>>sigpull(20,-5,5)".format(rinj=injectedAmount),"fit_status>=0")
hsigpull = ROOT.gDirectory.Get('sigpull')
tree_fit_sb.Draw("(r-{rinj})>>sigstrength(20,-1,1)".format(rinj=injectedAmount),"fit_status>=0")
hsignstrength = ROOT.gDirectory.Get('sigstrength')
hsigpull.Fit("gaus","L")
hsigpull.SetTitle('')
hsigpull.GetXaxis().SetTitle('(r-%s)/rErr'%injectedAmount)
result_can.cd()
hsigpull.Draw('pe')
result_can.Print('signalInjection_r%s_pull.png'%(str(injectedAmount).replace('.','p')),'png')
hsignstrength.Fit("gaus","L")
hsignstrength.SetTitle('')
hsignstrength.GetXaxis().SetTitle('r-%s'%injectedAmount)
result_can.cd()
hsignstrength.Draw('pe')
result_can.Print('signalInjection_r%s.png'%(str(injectedAmount).replace('.','p')),'png')
execute_cmd('rm -r '+tmpdir)
def plot_gof(tag, subtag, seed=123456, condor=False):
with cd(tag+'/'+subtag):
if condor:
tmpdir = 'notneeded/tmp/'
execute_cmd('mkdir '+tmpdir)
execute_cmd('cat %s_%s_gof_toys_output_*.tgz | tar zxvf - -i --strip-components 2 -C %s'%(tag,subtag,tmpdir))
toy_limit_tree = ROOT.TChain('limit')
if len(glob.glob(tmpdir+'higgsCombine_gof_toys.GoodnessOfFit.mH120.*.root')) == 0:
raise Exception('No files found')
toy_limit_tree.Add(tmpdir+'higgsCombine_gof_toys.GoodnessOfFit.mH120.*.root')
else:
toyOutput = ROOT.TFile.Open('higgsCombine_gof_toys.GoodnessOfFit.mH120.{seed}.root'.format(seed=seed))
toy_limit_tree = toyOutput.Get('limit')
# Now to analyze the output
# Get observation
ROOT.gROOT.SetBatch(True)
ROOT.gStyle.SetOptStat(False)
gof_data_file = ROOT.TFile.Open('higgsCombine_gof_data.GoodnessOfFit.mH120.root')
gof_limit_tree = gof_data_file.Get('limit')
gof_limit_tree.GetEntry(0)
gof_data = gof_limit_tree.limit
# Get toys
toy_limit_tree.Draw('limit>>hlimit','limit>1.0 && limit<%s && limit != %s'%(gof_data*2.0,gof_data))
htoy_gof = ROOT.gDirectory.Get('hlimit')
time.sleep(1) # if you don't sleep the code moves too fast and won't perform the fit
htoy_gof.Fit("gaus")
# Fit toys and derive p-value
gaus = htoy_gof.GetFunction("gaus")
pvalue = 1-(1/gaus.Integral(-float("inf"),float("inf")))*gaus.Integral(-float("inf"),gof_data)
# Write out for reference
with open('gof_results.txt','w') as out:
out.write('Test statistic in data = '+str(gof_data))
out.write('Mean from toys = '+str(gaus.GetParameter(1)))
out.write('Width from toys = '+str(gaus.GetParameter(2)))
out.write('p-value = '+str(pvalue))
# Extend the axis if needed
if htoy_gof.GetXaxis().GetXmax() < gof_data:
print ('Axis limit greater than GOF p value')
binwidth = htoy_gof.GetXaxis().GetBinWidth(1)
xmin = htoy_gof.GetXaxis().GetXmin()
new_xmax = int(gof_data*1.1)
new_nbins = int((new_xmax-xmin)/binwidth)
toy_limit_tree.Draw('limit>>hlimitrebin('+str(new_nbins)+', '+str(xmin)+', '+str(new_xmax)+')','limit>0.001 && limit<1500')
htoy_gof = ROOT.gDirectory.Get('hlimitrebin')
htoy_gof.Fit("gaus")
gaus = htoy_gof.GetFunction("gaus")
# Arrow for observed
arrow = ROOT.TArrow(gof_data,0.25*htoy_gof.GetMaximum(),gof_data,0)
arrow.SetLineWidth(2)
# Legend
leg = ROOT.TLegend(0.1,0.7,0.4,0.9)
leg.SetLineColor(ROOT.kWhite)
leg.SetLineWidth(0)
leg.SetFillStyle(0)
leg.SetTextFont(42)
leg.AddEntry(htoy_gof,"toy data","lep")
leg.AddEntry(arrow,"observed = %.1f"%gof_data,"l")
leg.AddEntry(0,"p-value = %.2f"%(pvalue),"")
# Draw
cout = ROOT.TCanvas('cout','cout',800,700)
htoy_gof.SetTitle('')
htoy_gof.Draw('pez')
arrow.Draw()
leg.Draw()
cout.Print('gof_plot.pdf','pdf')
cout.Print('gof_plot.png','png')
execute_cmd('rm -r '+tmpdir)
def Impacts(self,
subtag,
rMin=-15,
rMax=15,
cardOrW='initialFitWorkspace.root --snapshotName initialFit',
extra=''):
# param_str = '' if setParams == {} else '--setParameters '+','.join(['%s=%s'%(p,v) for p,v in setParams.items()])
with cd(self.tag + '/' + subtag):
subset = LoadLedger('')
impact_nuis_str = '--named=' + ','.join(subset.GetAllSystematics())
if cardOrW.endswith('.txt'):
execute_cmd(
'text2workspace.py -b %s -o prefitWorkspace.root --channel-masks --X-no-jmax'
% cardOrW)
card_or_w = 'prefitWorkspace.root'
else:
card_or_w = cardOrW
base_opts = [
'-M Impacts',
'--rMin %s' % rMin,
'--rMax %s' % rMax,
'-d %s' % card_or_w,
'--cminDefaultMinimizerStrategy 0 -m 0',
impact_nuis_str,
extra #param_str,
# '-t -1 --bypassFrequentistFit' if blindData else ''
]
# Remove old runs if they exist
execute_cmd('rm *_paramFit_*.root *_initialFit_*.root')
# Step 1
execute_cmd('combineTool.py %s --doInitialFit' %
(' '.join(base_opts)))
# Dumb hack - combineTool --doFits will go looking for the wrong file if you run on a toy
_combineTool_impacts_fix(
'higgsCombine_initialFit_Test.MultiDimFit.mH0.root')
# Step 2
execute_cmd('combineTool.py %s --doFits' % (' '.join(base_opts)))
# Dumb hack - combineTool next step will go looking for the wrong file if you run on a toy
_combineTool_impacts_fix(
'higgsCombine_paramFit_Test_*.MultiDimFit.mH0.root')
# Grab the output
execute_cmd('combineTool.py %s -o impacts.json' %
(' '.join(base_opts)))
execute_cmd('plotImpacts.py -i impacts.json -o impacts')
def SignalInjection(self,
subtag,
injectAmount,
ntoys,
blindData=True,
card_or_w='card.txt',
rMin=-5,
rMax=5,
seed=123456,
verbosity=0,
setParams={},
extra='',
condor=False,
eosRootfiles=None,
njobs=0):
run_dir = self.tag + '/' + subtag
_runDirSetup(run_dir)
rinj = str(injectAmount).replace('.', 'p')
param_str = '' if setParams == {} else '--setParameters ' + ','.join(
['%s=%s' % (p, v) for p, v in setParams.items()])
with cd(run_dir):
fit_cmd = [
'combine -M FitDiagnostics', '-d ' + card_or_w,
'--skipBOnlyFit', '--cminDefaultMinimizerStrategy 0',
'-t {ntoys}', '--toysFrequentist',
'--bypassFrequentistFit' if blindData else '', param_str,
'-s {seed}',
'--rMin %s' % rMin,
'--rMax %s' % rMax,
'-n _sigInj_r%s_{seed}' % rinj,
'-v %s' % verbosity, extra
]
fit_cmd = ' '.join(fit_cmd)
if not condor:
fit_cmd = fit_cmd.format(seed=seed, ntoys=ntoys)
execute_cmd(fit_cmd)
else:
per_job_toys = int(round(ntoys / njobs))
print(
'Running toys on condor... first cleaning potential duplicates...'
)
execute_cmd(
'rm fitDiagnostics_sigInj_r{rinj}_{seed}.root'.format(
rinj=rinj, seed=seed))
fit_cmds = [
fit_cmd.format(ntoys=per_job_toys,
seed=random.randint(0, 1000000))
for _ in range(njobs)
]
condor = CondorRunner(
name=self.tag + '_' + subtag + '_sigInj_r' + rinj,
primaryCmds=fit_cmds,
toPkg=self.tag + '/',
runIn=run_dir,
toGrab='{run_dir}/fitDiagnostics_sigInj_r{rinj}*.root'.
format(run_dir=run_dir, rinj=rinj),
eosRootfileTarball=eosRootfiles,
remakeEnv=False)
condor.submit()
def GoodnessOfFit(self,
subtag,
ntoys,
card_or_w='card.txt',
freezeSignal=False,
seed=123456,
verbosity=0,
extra='',
condor=False,
eosRootfiles=None,
njobs=0):
# NOTE: There's no way to blind data here - need to evaluate it to get the p-value
# param_str = '' if setParams == {} else '--setParameters '+','.join(['%s=%s'%(p,v) for p,v in setParams.items()])
run_dir = self.tag + '/' + subtag
_runDirSetup(run_dir)
with cd(run_dir):
gof_data_cmd = [
'combine -M GoodnessOfFit', '-d ' + card_or_w,
'--algo=saturated',
'' if not freezeSignal else '--fixedSignalStrength %s' %
freezeSignal, '-n _gof_data',
'-v %s' % verbosity, extra
]
gof_toy_cmd = copy.deepcopy(gof_data_cmd)
gof_toy_cmd.extend(
['--toysFrequentist', '-t {ntoys}', '-s {seed}'])
gof_data_cmd = ' '.join(gof_data_cmd)
gof_toy_cmd = ' '.join(gof_toy_cmd).replace(
'-n _gof_data', '-n _gof_toys')
execute_cmd(gof_data_cmd)
if not condor:
gof_toy_cmd = gof_toy_cmd.format(seed=seed, ntoys=ntoys)
execute_cmd(gof_toy_cmd)
else:
per_job_toys = int(round(ntoys / njobs))
print(
'Running toys on condor... first cleaning potential duplicates...'
)
execute_cmd(
'rm higgsCombine_gof_toys.GoodnessOfFit.mH120.*.root')
gof_toy_cmds = [
gof_toy_cmd.format(ntoys=per_job_toys,
seed=random.randint(0, 1000000))
for _ in range(njobs)
]
condor = CondorRunner(
name=self.tag + '_' + subtag + '_gof_toys',
primaryCmds=gof_toy_cmds,
toPkg=self.tag + '/',
runIn=run_dir,
toGrab=run_dir +
'/higgsCombine_gof_toys.GoodnessOfFit.mH120.*.root',
eosRootfileTarball=eosRootfiles,
remakeEnv=False)
condor.submit()
def GenerateToys(self,
name,
subtag,
card=None,
workspace=None,
ntoys=1,
seed=123456,
expectSignal=0,
setParams={},
freezeParams=[]):
if card == None and workspace == None:
raise IOError(
'Either card or workspace must be provided relative to the directory where the generation will be run.'
)
elif card != None and workspace != None:
raise IOError('Only one of card or workspace can be provided.')
if card:
if isinstance(card, str):
card_name = card
elif isinstance(card, bool) and card == True:
card_name = 'card.txt'
workspace_file = '%s_gen_workspace.root' % (name)
execute_cmd(
'text2workspace.py -b {0}/{1} -o {0}/{2} --channel-masks --X-no-jmax'
.format(self.tag + '/' + subtag, card_name, workspace_file))
input_opt = '-d %s' % workspace_file
elif workspace:
if isinstance(workspace, str):
workspace_file = workspace.split(':')
input_opt = '-d %s --snapshotName %s' % (workspace.split(':'))
elif isinstance(workspace, bool) and workspace == True:
workspace_file = 'initialFitWorkspace.root'
input_opt = '-d %s --snapshotName initialFit' % workspace_file
run_dir = self.tag + '/' + subtag
masks_off = [
'%s=0' % mask
for mask in self._getMasks(run_dir + '/' + workspace_file)
]
with cd(run_dir):
param_vals = ['%s=%s' % (k, v) for k, v in setParams.items()]
param_vals.extend(masks_off)
param_opt = ''
if len(param_vals) > 0:
param_opt = '--setParameters ' + ','.join(param_vals)
freeze_opt = ''
if len(freezeParams) > 0:
freeze_opt = '--freezeParameters ' + ','.join(freezeParams)
gen_command_pieces = [
'combine -M GenerateOnly', input_opt, param_opt,
'--toysFrequentist --bypassFrequentistFit',
'-t %s' % ntoys,
'--saveToys -s %s' % seed,
'--expectSignal %s' % expectSignal,
'-n _%s' % name, freeze_opt
]
execute_cmd(' '.join(gen_command_pieces))
toyfile_path = '%s/higgsCombine_%s.GenerateOnly.mH120.%s.root' % (
self.tag + '/' + subtag + '/', name, seed)
return toyfile_path