def BuildTransferFactorModel(data_3b, data_4b, category, tag):
print "[INFO] Processing predicted model for %s" % (tag)
#######################################
##Prepare data to create the model
#######################################
variablescr = ['HH_m']
originalcr, targetcr, originalcr_weights, targetcr_weights, transferfactor = data.preparedataformodel(
data_3b, data_4b, variablescr)
data.fittransferfactormodel(originalcr, targetcr, originalcr_weights,
targetcr_weights, transferfactor)
#plotter.Draw1DHistosComparison(originalcr, targetcr, variablescr, originalcr_weights,True,"%s_model"%tag)
########################################
## Save training information
########################################
SaveTransferFactorModel(variablescr, transferfactor, tag)
return originalcr_weights, transferfactor
def BuildReweightingModel(data_3b, data_4b, category, tag):
print "[INFO] Processing predicted model for %s" % (tag)
############################################################################
##Let's slice data one more time to have the inputs for the bdt reweighting#
############################################################################
if category == 'GGF':
variablescr = [
'H1_pt', 'H2_pt', 'H1_eta', 'H2_eta', 'nJet', 'H1_bb_deltaR',
'H2_bb_deltaR', 'H1_bb_deltaPhi', 'H2_bb_deltaPhi'
]
modelargs = [100, 0.1, 5, 300, 0.5]
elif category == 'VBF':
variablescr = [
'H1_pt', 'H2_pt', 'H1_eta', 'H2_eta', 'nJet', 'JJ_j1_qgl',
'JJ_j2_qgl', 'JJ_m', 'j1j2_deltaEta'
]
modelargs = [80, 0.1, 5, 200, 0.5]
else:
variablescr = ['H1_pt', 'H2_pt', 'H1_eta', 'H2_eta', 'nJet']
originalcr, targetcr, originalcr_weights, targetcr_weights = data.preparedataformodel(
data_3b, data_4b, variablescr)
#plotter.Draw1DHistosComparison(originalcr, targetcr, variablescr, originalcr_weights,True,"original_%s"%tag)
#######################################
##Folding Gradient Boosted Reweighter
#######################################
foldingcr_weights, reweightermodel, transferfactor = data.fitreweightermodel(
originalcr, targetcr, originalcr_weights, targetcr_weights, modelargs)
#plotter.Draw1DHistosComparison(originalcr, targetcr, variablescr, foldingcr_weights,True,"model_%s"%tag)
########################################
##GB ROC AUC
########################################
bdtreweighter.roc_auc_measurement(originalcr, targetcr, originalcr_weights,
foldingcr_weights)
########################################
##Update 3b dataframe for modeling
########################################
return foldingcr_weights, reweightermodel, transferfactor
def BuildReweightingModel(data_3b, data_4b, category, tag, bkgclassifierparams,
bkgparams):
print "[INFO] Processing predicted model for %s" % (tag)
modelargs = bkgparams
############################################################################
##Let's slice data one more time to have the inputs for the bdt reweighting
############################################################################
if category == 'GGF' or category == 'GGF1' or category == 'GGF2':
variablescr = [
'H1_b1_ptRegressed', 'H1_b2_ptRegressed', 'H2_b1_ptRegressed',
'H2_b2_ptRegressed', 'H1_m', 'H2_m', 'HH_m', 'H1_pt', 'H2_pt',
'h1h2_deltaEta', 'H1_bb_deltaR', 'H2_bb_deltaR',
'abs_costh_H1_ggfcm', 'HH_btag_b3_bres', 'abs_costh_H1_b1_h1cm',
'sum_3b_bscore', 'sum_4b_pt', 'HH_pt'
]
elif category == 'VBF' or category == 'VBF1' or category == 'VBF2':
variablescr = [
'H1_b1_ptRegressed', 'H1_b2_ptRegressed', 'H2_b1_ptRegressed',
'H2_b2_ptRegressed', 'H1_m', 'H2_m', 'HH_m', 'H1_pt', 'H2_pt',
'h1h2_deltaEta', 'h1h2_deltaPhi', 'JJ_m', 'j1j2_deltaEta',
'GGFKiller'
]
else:
variablescr = ['HH_m']
#######################################
##Prepare data to create the model
#######################################
originalcr, targetcr, originalcr_weights, targetcr_weights, transferfactor = data.preparedataformodel(
data_3b, data_4b, variablescr)
#plotter.Draw1DHistosComparison(originalcr, targetcr, variablescr, originalcr_weights,True,"%s_original"%tag)
#######################################
##Grid search or main fit
#######################################
print "--------------------------Best-Hyperparameters-------------------------------------"
print "-----------------------------------------------------------------------------------"
print "Parameters: ", modelargs
print "-----------------------------------------------------------------------------------"
#######################################
##Folding Gradient Boosted Reweighter (3-fold BDT reweighter)
#######################################
foldingcr_weights, reweightermodel = data.fitreweightermodel(
originalcr, targetcr, originalcr_weights, targetcr_weights,
transferfactor, modelargs)
#plotter.Draw1DHistosComparison(originalcr, targetcr, variablescr, foldingcr_weights,True,"%s_model"%tag)
########################################
## KS Test (as the developers of the method do), currently used for optimization/check of the parameters
########################################
ksresult_original = bdtreweighter.ks_test(originalcr, targetcr,
variablescr, originalcr_weights)
ksresult_model = bdtreweighter.ks_test(originalcr, targetcr, variablescr,
foldingcr_weights)
passedksresult = bdtreweighter.ks_comparison(variablescr,
ksresult_original,
ksresult_model)
###########################################################################################
## GB ROC AUC Test Study (Very slow test, needs to train a classifier in cross-validation)
############################################################################################
roc_auc_nowts = bdtreweighter.discrimination_test(originalcr, targetcr,
originalcr_weights,
bkgclassifierparams,
"%s" % tag, "original",
False)
roc_auc_wts = bdtreweighter.discrimination_test(originalcr, targetcr,
foldingcr_weights,
bkgclassifierparams,
"%s" % tag, "model", False)
passedresult = bdtreweighter.discrimination_comparison(
roc_auc_nowts, roc_auc_wts)
return foldingcr_weights, reweightermodel, transferfactor