def runReco(inf):
#Set the channel, load in the models and normalization factors
if '3l' in inf:
channel = '3l'
ptDict = ptDictHiggsTop3lS
topModel = load_model(
"/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top3l.h5"
)
topNormFactors = np.load(
"/data_ceph/afwebb/higgs_diff/topMatching/models/top3l_normFactors.npy",
allow_pickle=True)
model3lF = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lF.h5"
)
normFactors3lF = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lF_normFactors.npy",
allow_pickle=True)
model3lS = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lS.h5"
)
normFactors3lS = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lS_normFactors.npy",
allow_pickle=True)
elif '2lSS' in inf:
channel = '2lSS'
ptDict = ptDictHiggsTop2lSS
topModel = load_model(
"/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top2lSS.h5"
)
topNormFactors = np.load(
"/data_ceph/afwebb/higgs_diff/topMatching/models/top2lSS_normFactors.npy",
allow_pickle=True)
model2lSS = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop2lSS.h5"
)
normFactors2lSS = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop2lSS_normFactors.npy",
allow_pickle=True)
else:
print(f'Channel {channel} is invalid. Should be 2lSS or 3l')
return
#Open the root file
f = TFile.Open(inf)
nom = f.Get('nominal')
if hasattr(nom, "recoHiggsPt_2lSS") or hasattr(nom, "recoHiggsPt_3lS"):
print(f'{inf} already has score')
return
#initialize output dicts
events = []
higgsRecoScores = []
topRecoScores = []
if channel == '3l':
events3lF = []
eventsDecay = []
higgsRecoScoresF = []
#Loop over all entries
nEntries = nom.GetEntries()
for idx in range(nEntries):
if idx % 10000 == 0:
print(str(idx) + '/' + str(nEntries))
#Get the events
nom.GetEntry(idx)
#Find the best top combination, top reco score
topRes = findBestTopKeras(nom, channel, topModel, topNormFactors)
if not topRes:
topIdx0, topIdx1 = 0, 0
topScore = np.float32(-10)
else:
topIdx0, topIdx1 = topRes['bestComb']
topScore = topRes['topScore']
topRecoScores.append(topScore) # add the top reco score
#Find the higgs decay products, higgs reco score for 3lF model
if channel == '3l':
res3lF = findBestHiggsTop(nom, '3lF', model3lF, normFactors3lF,
topIdx0, topIdx1, topScore)
#if not res3lF: continue
higgsTopScoreF = res3lF['higgsTopScore']
lepIdx = res3lF['bestComb'][0]
events3lF.append(
ptDictHiggsTop3lF(nom, lepIdx, higgsTopScoreF, topIdx0,
topIdx1, topScore))
higgsRecoScoresF.append(higgsTopScoreF)
#Find the higgs decay products, higgs reco score for 3lS, 2lSS (same final state)
if channel == '3l':
res = findBestHiggsTop(nom, '3lS', model3lS, normFactors3lS,
topIdx0, topIdx1, topScore)
else:
res = findBestHiggsTop(nom, '2lSS', model2lSS, normFactors2lSS,
topIdx0, topIdx1, topScore)
if not res:
higgsTopScore = np.float32(-10)
lepIdx, jetIdx0, jetIdx1 = 1, 0, 0
else:
higgsTopScore = res['higgsTopScore']
lepIdx, jetIdx0, jetIdx1 = res['bestComb']
#add the pt prediction dictionary
events.append(
ptDict(nom, jetIdx0, jetIdx1, lepIdx, higgsTopScore, topIdx0,
topIdx1, topScore))
higgsRecoScores.append(higgsTopScore)
#add decay mode dicts
if channel == '3l':
eventsDecay.append(
decayDict(nom, higgsTopScoreF, higgsTopScore, topIdx0, topIdx1,
topScore))
if channel == '3l':
return events, events3lF, eventsDecay, higgsRecoScores, higgsRecoScoresF, topRecoScores
else:
return events, higgsRecoScores, topRecoScores
def runReco(inf):
#Set the channel, load in the top model
if '3l' in inf:
channel='3l'
topModel = load_model("/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top3l.h5")
topNormFactors = np.load("/data_ceph/afwebb/higgs_diff/topMatching/models/top3l_normFactors.npy")
flatDict = higgsTopDict3lS
is3l = True
elif '2lSS' in inf:
channel='2lSS'
topModel = load_model("/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top2lSS.h5")
topNormFactors = np.load("/data_ceph/afwebb/higgs_diff/topMatching/models/top2lSS_normFactors.npy")
flatDict = higgsTopDict2lSS
is3l = False
else:
print(f'Channel {channel} is invalid. Should be 2lSS or 3l')
exit()
print('loaded')
topMaxVals = topNormFactors[0]
topMinVals = topNormFactors[1]
topDiff = topMaxVals - topMinVals
f = TFile.Open(inf)
nom = f.Get('nominal')
#initialize output dicts
events = {}
events3lF = []
#Loop over all entries
nEntries = nom.GetEntries()
for idx in range(nEntries):
if idx%10000==0:
print(str(idx)+'/'+str(nEntries))
nom.GetEntry(idx)
#Check if the Higgs decay products are reconstructed - first leptons
if channel=='2lSS' and nom.lep_Parent_0!=25 and nom.lep_Parent_1!=25:
continue
if is3l: # Check for lepton, decide if event is 3lF or 3lS
if nom.lep_Parent_1!=25 and nom.lep_Parent_2!=25:
continue
if nom.lep_Parent_0 == 25:
channel='3lF'
else:
channel='3lS'
#Check if Higgs jets are reconstructed
if channel!='3lF' and sum([x==25 for x in nom.jet_parents])!=2: continue
#if sum([x==25 for x in nom.jet_parents])!=2: continue
#Find the b-jets from tops
topRes = findBestTopKeras(nom, channel, topModel, topNormFactors)
if not topRes:
continue
topIdx0, topIdx1 = topRes['bestComb']
topScore = topRes['topScore']
#Get all possible combinations
combos = higgsTopCombos(channel, nom, topIdx0, topIdx1, topScore, 1)
if not combos or len(combos['higgsDicts'])==0:
continue
if channel=='3lF':
if events3lF=={}:
events3lF=combos['higgsDicts']
else:
for k in events3lF:
events3lF[k].extend(combos['higgsDicts'][k])
else:
if events=={}:
events=combos['higgsDicts']
else:
for k in events:
events[k].extend(combos['higgsDicts'][k])
dfFlat = pd.DataFrame.from_dict(events)
dfFlat = shuffle(dfFlat)
outF = '/'.join(inf.split("/")[-2:]).replace('.root','.csv')
if channel=='2lSS':
dfFlat.to_csv('csvFiles/higgsTop2lSS/'+outF, index=False, float_format='%.3f')
else:# channel=='3l':
dfFlat.to_csv('csvFiles/higgsTop3lS/'+outF, index=False, float_format='%.3f')
df3lF = pd.DataFrame.from_dict(events3lF)
df3lF = shuffle(df3lF)
df3lF.to_csv('csvFiles/higgsTop3lF/'+outF, index=False, float_format='%.3f')
n1bCorrect, n2bCorrect, n3bCorrect = 0,0,0
#Loop over each entry, add to events dict
for idx in range(nEntries):
if idx%1000==0:
print(str(idx)+'/'+str(nEntries))
if idx==5000:
break
nom.GetEntry(idx)
if '1b' in sys.argv[2] and nom.nJets_OR_DL1r_70!=1: continue
if '2b' in sys.argv[2] and nom.nJets_OR_DL1r_70!=2: continue
if '3b' in sys.argv[2] and nom.nJets_OR_DL1r_70!=3: continue
topRes = findBestTopKeras(nom, channel, topModel, topNormFactors)
if not topRes:
continue
topMatches, truthBs, topScore = topRes['bestComb'], topRes['truthComb'], topRes['topScore']
#print(topRes['bestComb'], topRes['truthComb'], topRes['topScore'])
#topMatches, truthBs = findBestTopKeras(nom, channel, topModel, topNormFactors, 1)
if len(truthBs)!=2: continue
if topScore>0.3:
nGood+=1
if topMatches[0] in truthBs and topMatches[1] in truthBs:
nGoodCorrect+=1
if topMatches[0] in truthBs or topMatches[1] in truthBs:
nGoodOne+=1
if topScore<0.3:
lepCorrect = 0
oneCorrect = 0
#Loop over each entry, add to events dict
for idx in range(nEntries):
if idx % 1000 == 0:
print(str(idx) + '/' + str(nEntries))
if idx == 5000:
break
nom.GetEntry(idx)
if '3l' in channel:
if channel == '3lF' and nom.lep_Parent_0 != 25: continue
if channel == '3lS' and nom.lep_Parent_0 == 25: continue
topRes = findBestTopKeras(nom, '3l', topModel, topNormFactors)
else:
topRes = findBestTopKeras(nom, '2lSS', topModel, topNormFactors)
if not topRes:
continue
topIdx0, topIdx1 = topRes['bestComb']
topScore = topRes['topScore']
#Get dict of all possible jet combinations
higgsRes = findBestHiggsTop(nom, channel, higgsModel, higgsNormFactors,
topIdx0, topIdx1, topScore)
if not higgsRes: continue
higgsMatches = higgsRes['bestComb']
truthPair = higgsRes['truthComb']
def runReco(inf):
#load in the top model - not picklable, can't do outside the function
topModel = load_model(
"/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top3l.h5")
topNormFactors = np.load(
"/data_ceph/afwebb/higgs_diff/topMatching/models/top3l_normFactors.npy"
)
topMaxVals = topNormFactors[0]
topMinVals = topNormFactors[1]
topDiff = topMaxVals - topMinVals
model3lF = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lF.h5"
)
normFactors3lF = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lF_normFactors.npy"
)
maxVals3lF, minVals3lF = normFactors3lF
diff3lF = maxVals3lF - minVals3lF
model3lS = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lS.h5"
)
normFactors3lS = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lS_normFactors.npy"
)
maxVals3lS, minVals3lS = normFactors3lS
diff3lS = maxVals3lS - minVals3lS
f = TFile.Open(inf)
nom = f.Get('nominal')
#initialize output dicts
events = []
#Loop over all entries
nEntries = nom.GetEntries()
for idx in range(nEntries):
if idx % 10000 == 0:
print(str(idx) + '/' + str(nEntries))
nom.GetEntry(idx)
#Perform top matching. Get top candidates, topScore
topRes = findBestTopKeras(nom, '3l', topModel, topNormFactors)
if not topRes: continue
topIdx0, topIdx1 = topRes['bestComb']
topScore = topRes['topScore']
#Perform higgs matching. Get 3lF, 3lS scores
res3lF = findBestHiggsTop(nom, '3lF', model3lF, normFactors3lF,
topIdx0, topIdx1, topScore)
res3lS = findBestHiggsTop(nom, '3lS', model3lS, normFactors3lS,
topIdx0, topIdx1, topScore)
if not res3lF or not res3lS: continue
#identify which lepton came from the Higgs
lepIdx = -1
if nom.lep_Parent_0 == 25:
isF = True
else:
isF = False
if nom.lep_Parent_1 == 25: lepIdx = 1
if nom.lep_Parent_2 == 25: lepIdx = 2
if lepIdx == -1: continue
if isF:
events.append(
decayDict(nom, res3lF['higgsTopScore'],
res3lS['higgsTopScore'], topIdx0, topIdx1, topScore,
0)) #Correct combination
else:
events.append(
decayDict(nom, res3lF['higgsTopScore'],
res3lS['higgsTopScore'], topIdx0, topIdx1, topScore,
1)) #Incorrect combination - swaps 2 and 1
dfFlat = pd.DataFrame.from_dict(events)
dfFlat = shuffle(dfFlat)
outF = '/'.join(inf.split("/")[-2:]).replace('.root', '.csv')
dfFlat.to_csv('csvFiles/' + outF, index=False)
def runReco(inf):
#Set the channel, load in the top model
if '3l' in inf:
channel = '3l'
is3l = True
ptDict = ptDictHiggsTop3lS
topModel = load_model(
"/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top3l.h5"
)
topNormFactors = np.load(
"/data_ceph/afwebb/higgs_diff/topMatching/models/top3l_normFactors.npy"
)
model3lF = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lF.h5"
)
normFactors3lF = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lF_normFactors.npy"
)
model3lS = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop3lS.h5"
)
normFactors3lS = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop3lS_normFactors.npy"
)
elif '2lSS' in inf:
channel = '2lSS'
ptDict = ptDictHiggsTop2lSS
is3l = False
topModel = load_model(
"/data_ceph/afwebb/higgs_diff/topMatching/models/keras_model_top2lSS.h5"
)
topNormFactors = np.load(
"/data_ceph/afwebb/higgs_diff/topMatching/models/top2lSS_normFactors.npy"
)
model2lSS = load_model(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/keras_model_higgsTop2lSS.h5"
)
normFactors2lSS = np.load(
"/data_ceph/afwebb/higgs_diff/higgsMatching/models/higgsTop2lSS_normFactors.npy"
)
else:
print(f'Channel {channel} is invalid. Should be 2lSS or 3l')
exit()
f = TFile.Open(inf)
nom = f.Get('nominal')
#initialize output dicts
events = []
events3lF = []
#Loop over all entries
nEntries = nom.GetEntries()
for idx in range(nEntries):
if idx % 10000 == 0:
print(str(idx) + '/' + str(nEntries))
nom.GetEntry(idx)
#Get the Higgs Pt
for i, pdgId in enumerate(nom.m_truth_pdgId):
if pdgId == 25:
higgs_pt = nom.m_truth_pt[i]
break
if not higgs_pt:
continue
topRes = findBestTopKeras(nom, channel, topModel, topNormFactors)
if not topRes: continue
topIdx0, topIdx1 = topRes['bestComb']
topScore = topRes['topScore']
isF = False
if is3l and nom.lep_Parent_0 == 25:
isF = True
if isF:
res3lF = findBestHiggsTop(nom, '3lF', model3lF, normFactors3lF,
topIdx0, topIdx1, topScore)
if not res3lF: continue
higgsTopScore = res3lF['higgsTopScore']
lepIdx = res3lF['bestComb'][0]
events3lF.append(
ptDictHiggsTop3lF(nom, lepIdx, higgsTopScore, topIdx0, topIdx1,
topScore, higgs_pt))
else:
if is3l:
res = findBestHiggsTop(nom, '3lS', model3lS, normFactors3lS,
topIdx0, topIdx1, topScore)
else:
res = findBestHiggsTop(nom, '2lSS', model2lSS, normFactors2lSS,
topIdx0, topIdx1, topScore)
if not res: continue
higgsTopScore = res['higgsTopScore']
lepIdx, jetIdx0, jetIdx1 = res['bestComb']
events.append(
ptDict(nom, jetIdx0, jetIdx1, lepIdx, higgsTopScore, topIdx0,
topIdx1, topScore, higgs_pt))
dfFlat = pd.DataFrame.from_dict(events)
dfFlat = shuffle(dfFlat)
outF = '/'.join(inf.split("/")[-2:]).replace('.root', '.csv')
if channel == '2lSS':
dfFlat.to_csv('inputFiles/higgsTop2lSS/' + outF, index=False)
elif channel == '3l':
dfFlat.to_csv('inputFiles/higgsTop3lS/' + outF, index=False)
df3lF = pd.DataFrame.from_dict(events3lF)
df3lF = shuffle(df3lF)
df3lF.to_csv('inputFiles/higgsTop3lF/' + outF, index=False)
