def customInit(self, initVars):
self.sampleTree = initVars['sampleTree']
self.isData = initVars['sample'].isData()
if not self.isData:
self.addCollection(
Collection('GenVbosons', ['pt', 'pdgId', 'GenPartIdx'],
maxSize=40))
self.addCollection(
Collection('GenTop', ['pt', 'GenPartIdx'], maxSize=4))
self.addCollection(
Collection('GenHiggsBoson', ['pt', 'GenPartIdx'], maxSize=4))
self.branchBuffers['VtypeSim'] = array.array('i', [0])
self.branches.append({
'name': 'VtypeSim',
'formula': self.getBranch,
'arguments': 'VtypeSim',
'type': 'i'
})
self.addCollection(
Collection('GenBs', ['pt', 'eta', 'phi', 'genPartIdx'],
maxSize=32))
self.addCollection(
Collection('GenDs', ['pt', 'eta', 'phi', 'genPartIdx'],
maxSize=32))
self.addVectorBranch("GenJetAK8_nBhadrons",
default=0,
branchType='i',
length=100,
leaflist="GenJetAK8_nBhadrons[nGenJetAK8]/i")
self.addVectorBranch(
"GenJetAK8_nBhadrons2p4",
default=0,
branchType='i',
length=100,
leaflist="GenJetAK8_nBhadrons2p4[nGenJetAK8]/i")
self.addVectorBranch("GenJet_nBhadrons",
default=0,
branchType='i',
length=100,
leaflist="GenJet_nBhadrons[nGenJet]/i")
self.addVectorBranch("GenJet_nBhadrons2p4",
default=0,
branchType='i',
length=100,
leaflist="GenJet_nBhadrons2p4[nGenJet]/i")
#Sum$(GenBs_pt>25&&abs(GenBs_eta)<2.6)
self.addIntegerBranch("nGenBpt25eta2p6")
self.addIntegerBranch("nGenBpt20eta2p6")
self.addIntegerBranch("nGenDpt25eta2p6")
self.addIntegerBranch("nGenDpt20eta2p6")
def customInit(self, initVars):
self.config = initVars['config']
self.sampleTree = initVars['sampleTree']
self.sample = initVars['sample']
self.tensorflowConfig = self.config.get(self.mvaName,
'tensorflowConfig')
self.scalerDump = self.config.get(self.mvaName, 'scalerDump')
self.checkpoint = self.config.get(self.mvaName, 'checkpoint')
self.branchName = self.config.get(self.mvaName, 'branchName')
# Jet systematics
self.systematics = self.config.get('systematics',
'systematics').split(' ')
# create output branches
self.dnnCollection = Collection(self.branchName,
self.systematics,
leaves=True)
self.addCollection(self.dnnCollection)
# create formulas for input variables
self.inputVariables = {}
for syst in self.systematics:
self.inputVariables[syst] = self.config.get(
self.config.get(self.mvaName, "treeVarSet"),
syst if self.sample.isMC() else 'Nominal').split(' ')
for var in self.inputVariables[syst]:
self.sampleTree.addFormula(var)
# create tensorflow graph
self.reloadModel()
def __init__(self, nano=False):
self.nano = nano
self.debug = False
super(FSR, self).__init__()
# corrected Higgs properties
self.addCollection(
Collection('HCMVAV2_reg_fsrCorr', ['pt', 'eta', 'phi', 'mass']))
self.addCollection(
Collection('fsrJet', ['pt', 'eta', 'phi', 'mass', 'deltaR'],
maxSize=4))
self.addCollection(
Collection('isrJet', ['pt', 'eta', 'phi', 'mass', 'deltaR'],
maxSize=4))
# test of new syntax
self.addCollection(
Collection('hJetFSRcorr', ['pt', 'eta', 'phi', 'mass'], maxSize=2))
def customInit(self, initVars):
v_sys = getattr(ROOT, 'vector<string>')()
for syst in self.systematics:
v_sys.push_back(syst)
print 'load BTagCalibrationStandalone...'
if os.path.isfile(self.btagCalibratorFileName):
ROOT.gSystem.Load(self.btagCalibratorFileName)
else:
print "\x1b[31m:ERROR: BTagCalibrationStandalone not found! Go to Xbb directory and run 'make'!\x1b[0m"
raise Exception("BTagCalibrationStandaloneNotFound")
print 'load bTag CSV files...'
calib = ROOT.BTagCalibration(self.calibName, self.calibFile)
self.btag_calibrators = {}
print "[btagSF]: Loading calibrator for algo:", self.calibName
self.btag_calibrators[self.calibName +
"_iterative"] = ROOT.BTagCalibrationReader(
3, "central", v_sys)
for fl in range(3):
self.btag_calibrators[self.calibName + "_iterative"].load(
calib, fl, self.method)
print 'INFO: bTag initialization done.'
sample = initVars['sample']
self.isData = sample.type == 'DATA'
if not self.isData:
self.systBranches = [""] #nominal
self.systBranches += [
syst + sdir for syst in self.systList for sdir in self.systVars
]
if self.decorrelatePtEta:
self.systBranches += [
syst + "_pt" + str(ipt) + "_eta" + str(ieta) + sdir
for syst in self.systList for sdir in self.systVars
for ipt in range(0, 5) for ieta in range(1, 4)
]
self.btagCollection = Collection(self.branchBaseName,
self.systBranches,
leaves=False)
self.addCollection(self.btagCollection)
def customInit(self, initVars):
sample = initVars['sample']
self.isData = sample.type == 'DATA'
if not self.isData:
self.systBranches = [""] #nominal
self.systBranches += [
syst + sdir for syst in self.systList for sdir in self.systVars
]
if self.includeFixPtEtaBins:
self.systBranches += [
syst + "_pt" + str(ipt) + "_eta" + str(ieta) + sdir
for syst in self.systList for sdir in self.systVars
for ipt in range(0, 5) for ieta in range(1, 4)
]
self.btagCollection = Collection(self.branchBaseName,
self.systBranches,
leaves=False)
self.addCollection(self.btagCollection)
print "sys:", self.systBranches
def customInit(self, initVars):
self.config = initVars['config']
self.sampleTree = initVars['sampleTree']
self.sample = initVars['sample']
if self.condition:
self.sampleTree.addFormula(self.condition)
self.hJidx = self.config.get('General',
'hJidx') if self.config.has_option(
'General', 'hJidx') else 'hJidx'
self.scalerDump = self.config.get(
self.mvaName, 'scalerDump') if self.config.has_option(
self.mvaName, 'scalerDump') else None
self.checkpoint = self.config.get(
self.mvaName, 'checkpoint') if self.config.has_option(
self.mvaName, 'checkpoint') else None
if self.config.has_option(self.mvaName, 'branchName'):
self.branchName = self.config.get(self.mvaName, 'branchName')
elif self.checkpoint is not None:
self.branchName = self.checkpoint.strip().replace(
'/model.ckpt', '').replace('/', '_')
if self.branchName[0] in [
'0', '1', '2', '3', '4', '5', '6', '7', '8', '9'
]:
self.branchName = 'DNN_' + self.branchName
self.addDebugVariables = eval(
self.config.get('Multi', 'evalAddDebugVariables')
) if self.config.has_section('Multi') and self.config.has_option(
'Multi', 'evalAddDebugVariables') else False
if self.checkpoint is None:
print(
"\x1b[31mERROR: 'checkpoint' option missing for MVA config section [%s]! .../model.ckpt has to be specified to be able to restore classifier.\x1b[0m"
% self.mvaName)
raise Exception("CheckpointError")
if self.scalerDump is not None and not os.path.isfile(self.scalerDump):
self.scalerDump = None
if self.config.has_option(self.mvaName, 'fixInputs'):
self.fixInputs = eval(self.config.get(self.mvaName, 'fixInputs'))
if os.path.isdir(self.checkpoint):
self.checkpoint += '/model.ckpt'
if not os.path.isfile(self.checkpoint + '.meta'):
print(
"\x1b[31mERROR: can't restore from graph! .meta file not found in checkpoint:",
self.checkpoint, "\x1b[0m")
raise Exception("CheckpointError")
# INFO file (with training parameters)
if os.path.isfile(self.checkpoint + '.info'):
with open(self.checkpoint + '.info', 'r') as infoFile:
self.info = json.load(infoFile)
else:
print("WARNING: (optional) .info file not found in checkpoint!")
self.info = {}
# CLASSES
self.classes = eval(self.config.get(
self.mvaName, 'classes')) if self.config.has_option(
self.mvaName, 'classes') else None
if self.classes:
self.nClasses = len(self.classes)
else:
try:
self.nClasses = eval(self.config.get(self.mvaName, 'nClasses'))
except Exception as e:
self.nClasses = len(self.info["labels"].keys())
if self.nClasses < 3:
self.nClasses = 1
if self.nClasses > 1:
print("INFO: multi-class checkpoint found! number of classes =",
self.nClasses)
else:
print("INFO: binary-classifier found!")
# FEATURES
self.featuresConfig = None
self.featuresCheckpoint = None
try:
self.featuresConfig = self.config.get(
self.config.get(self.mvaName, "treeVarSet"),
"Nominal").strip().split(" ")
except Exception as e:
print("WARNING: could not get treeVarSet from config:", e)
if 'variables' in self.info:
self.featuresCheckpoint = self.info['variables']
if self.featuresConfig is None and self.featuresCheckpoint is not None:
self.features = self.featuresCheckpoint
elif self.featuresConfig is not None and self.featuresCheckpoint is None:
self.features = self.featuresConfig
elif self.featuresConfig is None and self.featuresCheckpoint is None:
raise Exception("NoInputFeaturesDefined")
else:
self.features = self.featuresCheckpoint
if len(self.featuresConfig) != len(self.featuresCheckpoint):
print(
"\x1b[31mWARNING: number of input features does not match!"
)
print(" > classifier expects:", len(self.featuresCheckpoint))
print(" > configuration has:", len(self.featuresConfig),
"\x1b[0m")
print("INFO: => feature list from checkpoint will be used.")
else:
if self.config.has_option(
self.mvaName, 'forceInputFeaturesFromConfig') and eval(
self.config.get(self.mvaName,
'forceInputFeaturesFromConfig')):
print(
"INFO: forceInputFeaturesFromConfig is enabled, features from configuration will be used"
)
self.features = self.featuresConfig
print("INFO: list of input features:")
print("INFO:", "config".ljust(40), "---->", "checkpoint")
match = True
for i in range(
min(len(self.featuresConfig),
len(self.featuresCheckpoint))):
if self.featuresConfig[i] != self.featuresCheckpoint[i]:
print("\x1b[41m\x1b[37mINFO:",
self.featuresConfig[i].ljust(40), "---->",
self.featuresCheckpoint[i].ljust(40),
" => MISMATCH!\x1b[0m")
match = False
else:
print("INFO:\x1b[32m", self.featuresConfig[i].ljust(40),
"---->", self.featuresCheckpoint[i],
"(match)\x1b[0m")
if match:
print("INFO: => all input variables match!")
else:
print(
"INFO: some variables are not identically defined as for the training, please check!"
)
if self.config.has_option(
self.mvaName, 'forceInputFeaturesFromConfig') and eval(
self.config.get(self.mvaName,
'forceInputFeaturesFromConfig')):
print(
"\x1b[31mWARNING: forceInputFeaturesFromConfig is enabled, features from configuration will be used although they could be incompatible with the features used during training.\x1b[0m"
)
# fix input features at constant values: check if features given exist in checkpoint
if self.fixInputs:
for feature, value in self.fixInputs.items():
if feature not in self.featuresCheckpoint:
print("ERROR: can't fix input feature '", feature,
"' to value ", value,
" => feature not found in checkpoint.")
raise Exception("ConfigError")
self.featureList = XbbMvaInputsList(self.features, config=self.config)
self.nFeatures = self.featureList.length()
# SIGNAL definition
self.signalIndex = 0
if self.config.has_option(self.mvaName, 'signalIndex'):
self.signalIndex = eval(
self.config.get(self.mvaName, 'signalIndex'))
self.signalClassIds = [self.signalIndex]
if self.classes:
self.signalClassIds = [
x for x, y in enumerate(self.classes) if y[0].startswith('SIG')
]
print("INFO: signals:", self.signalClassIds)
print("INFO: number of classes:",
self.nClasses if self.nClasses > 1 else 2)
# systematics
self.systematics = self.config.get(
self.mvaName, 'systematics').split(' ') if self.config.has_option(
self.mvaName, 'systematics') else self.config.get(
'systematics', 'systematics').split(' ')
# create output branches
self.dnnCollections = []
for i in range(self.nClasses):
collectionName = self.branchName if self.nClasses == 1 else self.branchName + "_%d" % i
if self.nClasses == 1 or self.addDebugVariables:
self.dnnCollection = Collection(collectionName,
self.systematics,
leaves=True)
self.addCollection(self.dnnCollection)
self.dnnCollections.append(self.dnnCollection)
# create formulas for input variables
self.inputVariables = {}
for syst in self.systematics:
systBase, UD = XbbTools.splitSystVariation(syst,
sample=self.sample)
self.inputVariables[syst] = [
XbbTools.sanitizeExpression(self.featureList.get(i,
syst=systBase,
UD=UD),
self.config,
debug=self.debug)
for i in range(self.nFeatures)
]
for var in self.inputVariables[syst]:
self.sampleTree.addFormula(var)
# additional pre-computed values for multi-classifiers
if self.nClasses > 1:
self.dnnCollectionsMulti = {
'default':
Collection(self.branchName, self.systematics, leaves=True),
}
if self.addDebugVariables:
self.dnnCollectionsMulti.update({
'argmax':
Collection(self.branchName + "_argmax",
self.systematics,
leaves=True),
'max':
Collection(self.branchName + "_max",
self.systematics,
leaves=True),
'max2':
Collection(self.branchName + "_max2",
self.systematics,
leaves=True),
'signal':
Collection(self.branchName + "_signal",
self.systematics,
leaves=True),
})
for k, v in self.dnnCollectionsMulti.items():
self.addCollection(v)
# create tensorflow graph
self.ev = TensorflowDNNEvaluator(checkpoint=self.checkpoint,
scaler=self.scalerDump)