def convert_vector(yaml_config, parameters, labels, parameters_df, labels_df, parameters_val, labels_val):
"""
Return: Numpy array (Events, Features)
Input: yaml_config | Dictionary of config options
parameters | List of parameters
labels | List of labels
parameters_df | Parameters dataframe
labels_df | Labels dataframe
parameters_val | Parameters array
labels_val | Labels array
"""
print('Converting data to vector...')
max_particles = yaml_config['MaxParticles']
vec_length = max_particles * (len(parameters) - 1)
# Allocate space
parameters_vec = np.zeros((len(labels_df), vec_length))
# Progress
event_num = 0. ; total_num = len(labels_df)
for i in range(len(labels_df)):
parameters_df_i = parameters_df[parameters_df['jet_index'] == labels_df['jet_index'].iloc[i]]
index_values = parameters_df_i.index.values
parameters_val_i = parameters_val[index_values, :-1]
num_particles = len(parameters_val_i)
if num_particles > max_particles:
parameters_val_i = parameters_val_i[0:max_particles, :]
else:
parameters_val_i = np.concatenate([parameters_val_i, np.zeros((max_particles - num_particles,
len(parameters) - 1))])
if yaml_config['Shuffle']: np.random.shuffle(parameters_val_i)
vector = []
for idx in range(len(parameters_val_i)):
vector.extend(parameters_val_i[idx])
parameters_vec[i,:] = vector
event_num += 1.
progress.update_progress(event_num/total_num)
parameters_val = parameters_vec
return parameters_val, labels_val
def convert_image(yaml_config, parameters, labels, parameters_df, labels_df, parameters_val, labels_val):
"""
Return: Numpy array (Events, Eta, Phi, Features)
Input: yaml_config | Dictionary of config options
parameters | List of parameters
labels | List of labels
parameters_df | Parameters dataframe
labels_df | Labels dataframe
parameters_val | Parameters array
labels_val | Labels array
"""
print('Converting data to image...')
BinsX = yaml_config['BinsX']
BinsY = yaml_config['BinsY']
xbins = np.linspace(yaml_config['MinX'], yaml_config['MaxX'], BinsX + 1)
ybins = np.linspace(yaml_config['MinY'], yaml_config['MaxY'], BinsY + 1)
parameters.remove('eta'); parameters.remove('phi')
parameters.remove('jet_eta'); parameters.remove('jet_phi'); parameters.remove('jet_index')
# Allocate space
parameters_image = np.zeros((len(labels_df), BinsX, BinsY, len(parameters)))
# Progress
event_num = 0. ; total_num = len(labels_df)
for i in range(len(labels_df)):
parameters_df_i = parameters_df[parameters_df['jet_index'] == labels_df['jet_index'].iloc[i]]
eta = np.asarray(parameters_df_i['eta'])
phi = np.asarray(parameters_df_i['phi'])
scaled_eta = np.asarray(parameters_df_i['jet_eta']) - eta
scaled_phi = np.asarray(parameters_df_i['jet_phi']) - phi
for param_idx, param in enumerate(parameters):
w = np.asarray(parameters_df_i[param])
hist, _, _ = np.histogram2d(scaled_eta, scaled_phi, weights=w, bins=(xbins, ybins))
for ix in range(BinsX):
for iy in range(BinsY):
parameters_image[i, ix, iy, param_idx] = hist[ix, iy]
event_num += 1.
progress.update_progress(event_num/total_num)
parameters_val = parameters_image
return parameters_val, labels_val
def convert_data(tree, number=None, regression=False, puppi=False):
"""
Particle flow inputs => Numpy structured array
"""
event_num = 0. # Event counter
jet_num = 0 # Jet counter
particle_num = 0 # Particle counter
if number: total_num = number
else: total_num = int(tree.GetEntries())
# Parameter lists
pt = []
eta = []
phi = []
energy = []
photon_ID = []
electron_ID = []
muon_ID = []
neutral_hadron_ID = []
charged_hadron_ID = []
jet_pt = []
jet_eta = []
jet_phi = []
jet_index = []
if regression:
tau_pt = []
tau_eta = []
tau_phi = []
classification = []
for event in tree:
if event_num == total_num: break
jet_candidates = create_jets(event, puppi=puppi)
tau_candidates = create_taus(event)
jets = match_taus(jet_candidates, tau_candidates)
# Fill parameters lists
for jet, seed, tau in jets:
if regression and (not tau): continue
for part, ID in jet:
# ID Parameters
if ID == 22:
photon_ID.append(1)
electron_ID.append(0)
muon_ID.append(0)
neutral_hadron_ID.append(0)
charged_hadron_ID.append(0)
elif abs(ID) == 11:
photon_ID.append(0)
electron_ID.append(1)
muon_ID.append(0)
neutral_hadron_ID.append(0)
charged_hadron_ID.append(0)
elif abs(ID) == 13:
photon_ID.append(0)
electron_ID.append(0)
muon_ID.append(1)
neutral_hadron_ID.append(0)
charged_hadron_ID.append(0)
elif ID == 130:
photon_ID.append(0)
electron_ID.append(0)
muon_ID.append(0)
neutral_hadron_ID.append(1)
charged_hadron_ID.append(0)
elif ID == 211:
photon_ID.append(0)
electron_ID.append(0)
muon_ID.append(0)
neutral_hadron_ID.append(0)
charged_hadron_ID.append(1)
else:
continue
# particle parameters
eta_val = seed.Eta() - part.Eta()
phi_val = delta_phi(seed.Phi(), part.Phi())
pt.append(part.Pt())
eta.append(eta_val)
phi.append(phi_val)
energy.append(part.E())
# jet parameters
jet_pt.append(seed.Pt())
jet_eta.append(seed.Eta())
jet_phi.append(seed.Phi())
jet_index.append(jet_num)
# tau parameters
if tau:
if regression:
tau_pt.append(tau.Pt())
tau_eta.append(tau.Eta())
tau_phi.append(tau.Phi())
classification.append(1)
else:
classification.append(0)
particle_num += 1
jet_num += 1
event_num += 1
progress.update_progress(event_num / total_num)
fields = [('pt', 'f8'), ('eta', 'f8'), ('phi', 'f8'), ('energy', 'f8'),
('photon_ID', 'i4'), ('electron_ID', 'i4'), ('muon_ID', 'i4'),
('neutral_hadron_ID', 'i4'), ('charged_hadron_ID', 'i4'),
('jet_pt', 'f8'), ('jet_eta', 'f8'), ('jet_phi', 'f8'),
('jet_index', 'i8'), ('classification', 'i4')]
if regression:
fields.extend([('tau_pt', 'f8'), ('tau_eta', 'f8'), ('tau_phi', 'f8')])
data = np.zeros(particle_num, dtype=fields)
data['pt'] = pt
data['eta'] = eta
data['phi'] = phi
data['energy'] = energy
data['photon_ID'] = photon_ID
data['electron_ID'] = electron_ID
data['muon_ID'] = muon_ID
data['neutral_hadron_ID'] = neutral_hadron_ID
data['charged_hadron_ID'] = charged_hadron_ID
data['jet_pt'] = jet_pt
data['jet_eta'] = jet_eta
data['jet_phi'] = jet_phi
if regression:
data['tau_pt'] = tau_pt
data['tau_eta'] = tau_eta
data['tau_phi'] = tau_phi
data['jet_index'] = jet_index
data['classification'] = classification
return data
def plot_data(tree, number=None):
"""
Plot matching efficiencies
"""
event_num = 0. # Event counter
if number: total_num = number
else: total_num = int(tree.GetEntries())
taus_matched = 0
taus_reconstructed = 0
# Initialize histograms
hist_pt1, hist_pt2 = initialize_hist(25, 0, 500)
hist_eta1, hist_eta2 = initialize_hist(15, -2.5, 2.5)
hist_lpt1, hist_lpt2 = initialize_hist(25, 0, 500)
hist_leta1, hist_leta2 = initialize_hist(15, -2.5, 2.5)
hist_slpt1, hist_slpt2 = initialize_hist(25, 0, 500)
hist_sleta1, hist_sleta2 = initialize_hist(15, -2.5, 2.5)
cnt = Counter()
for event in tree:
if event_num == total_num: break
jet_candidates = create_jets(event)
tau_candidates = create_taus(event)
jets = [] # List of jets
used_candidates = []
for seed, jet in jet_candidates:
tau = None
for vec in tau_candidates:
if (seed.DeltaR(vec) < 0.4) and (vec not in used_candidates):
tau = vec
used_candidates.append(vec)
break
jets.append((jet, seed, tau))
# Fill histograms
taus_matched += len(used_candidates)
taus_reconstructed += len(tau_candidates)
for vec in used_candidates:
hist_pt1.Fill(vec.Pt())
hist_eta1.Fill(vec.Eta())
for vec in tau_candidates:
hist_pt2.Fill(vec.Pt())
hist_eta2.Fill(vec.Eta())
tau_candidates = sorted(tau_candidates, key=lambda x: x.Pt())[::-1]
if len(tau_candidates) >= 1:
hist_lpt2.Fill(tau_candidates[0].Pt())
hist_leta2.Fill(tau_candidates[0].Eta())
if tau_candidates[0] in used_candidates:
hist_lpt1.Fill(tau_candidates[0].Pt())
hist_leta1.Fill(tau_candidates[0].Eta())
if len(tau_candidates) >= 2:
hist_slpt2.Fill(tau_candidates[1].Pt())
hist_sleta2.Fill(tau_candidates[1].Eta())
if tau_candidates[1] in used_candidates:
hist_slpt1.Fill(tau_candidates[1].Pt())
hist_sleta1.Fill(tau_candidates[1].Eta())
event_num += 1
progress.update_progress(event_num / total_num)
# Plot histograms
print 'Taus matched: ', taus_matched
print 'Taus reconstructed: ', taus_reconstructed
plot_hist(hist_pt1, hist_pt2, 'matching_efficiency_pt', 'pt', 'efficiency')
plot_hist(hist_lpt1, hist_lpt2, 'matching_efficiency_leadingpt', 'pt',
'efficiency')
plot_hist(hist_slpt1, hist_slpt2, 'matching_efficiency_subleadingpt', 'pt',
'efficiency')
plot_hist(hist_eta1, hist_eta2, 'matching_efficiency_eta', 'eta',
'efficiency')
plot_hist(hist_leta1, hist_leta2, 'matching_efficiency_leadingeta', 'eta',
'efficiency')
plot_hist(hist_sleta1, hist_sleta2, 'matching_efficiency_subleadingeta',
'eta', 'efficiency')
return None
def convert_data(tree, number=None):
"""
Convert data for binary classification
"""
event_num = 0.
jet_num = 0
particle_num = 0
if number: total_num = number
else: total_num = int(tree.GetEntries())
# Parameter lists
pt = []
eta = []
phi = []
et = []
photon_ID = []
electron_ID = []
hadron_ID = []
jet_pt = []
jet_eta = []
jet_phi = []
jet_et = []
jet_index = []
classification = []
for event in tree:
if event_num == total_num: break
for jet_idx, jet_id in enumerate(
event.genjetid): # iterate through jet
for k, _ in enumerate(
event.genindex): # iterate through jet particles
if (event.genindex[k] == jet_idx):
if (event.genid[k] == 22):
photon_ID.append(1)
electron_ID.append(0)
hadron_ID.append(0)
elif (abs(event.genid[k]) == 11):
photon_ID.append(0)
electron_ID.append(1)
hadron_ID.append(0)
elif (abs(event.genid[k]) > 40):
photon_ID.append(0)
electron_ID.append(0)
hadron_ID.append(1)
else:
continue
# particle parameters
eta_val = event.genjeteta[jet_idx] - event.geneta[k]
phi_val = delta_phi(event.genjetphi[jet_idx],
event.genphi[k])
pt.append(event.genpt[k])
eta.append(eta_val)
phi.append(phi_val)
et.append(event.genet[k])
# jet parameters
jet_pt.append(event.genjetpt[jet_idx])
jet_eta.append(event.genjeteta[jet_idx])
jet_phi.append(event.genjetphi[jet_idx])
jet_et.append(event.genjetet[jet_idx])
jet_index.append(jet_num)
if (abs(jet_id) >= 4):
classification.append(1)
else:
classification.append(0)
particle_num += 1
jet_num += 1
event_num += 1.
progress.update_progress(event_num / total_num)
fields = [('pt', 'f8'), ('eta', 'f8'), ('phi', 'f8'), ('et', 'f8'),
('photon_ID', 'i4'), ('electron_ID', 'i4'), ('hadron_ID', 'i4'),
('jet_pt', 'f8'), ('jet_eta', 'f8'), ('jet_phi', 'f8'),
('jet_et', 'f8'), ('jet_index', 'i8'), ('classification', 'i4')]
data = np.zeros(particle_num, dtype=fields)
data['pt'] = pt
data['eta'] = eta
data['phi'] = phi
data['et'] = et
data['photon_ID'] = photon_ID
data['electron_ID'] = electron_ID
data['hadron_ID'] = hadron_ID
data['jet_pt'] = jet_pt
data['jet_eta'] = jet_eta
data['jet_phi'] = jet_phi
data['jet_et'] = jet_et
data['jet_index'] = jet_index
data['classification'] = classification
return data
def convert_regression_data(tree, number=None):
"""
Convert data for regression
"""
event_num = 0.
jet_num = 0
particle_num = 0
if number: total_num = number
else: total_num = int(tree.GetEntries())
# Parameter lists
pt = []
eta = []
phi = []
et = []
photon_ID = []
electron_ID = []
hadron_ID = []
jet_pt = []
jet_eta = []
jet_phi = []
jet_et = []
jet_index = []
tau_pt = []
tau_eta = []
tau_phi = []
tau_energy = []
for event in tree:
if event_num == total_num: break
for jet_idx, jet_id in enumerate(
event.genjetid): # iterate through jet
if (abs(jet_id) >= 4): # consider tau jets
particle_vec = []
for k, _ in enumerate(
event.genindex): # iterate through jet particles
if (event.genindex[k] == jet_idx):
if (event.genstatus[k]
== 1) and ((event.genid[k] in [11, -11, 22]) or
(abs(event.genid[k]) > 40)):
index = k
while ((event.genparent[index] != -2)
and (abs(event.genid[index] != 15))):
index = event.genparent[index]
if (abs(event.genid[index]) == 15):
vec_pt = event.genpt[k]
vec_eta = event.geneta[k]
vec_phi = event.genphi[k]
vec_energy = event.genet[k] * np.cosh(vec_eta)
vec = TLorentzVector()
vec.SetPtEtaPhiE(vec_pt, vec_eta, vec_phi,
vec_energy)
particle_vec.append(vec)
if particle_vec:
vec_sum = TLorentzVector()
vec_sum.SetPtEtaPhiE(0., 0., 0., 0.)
for vec in particle_vec:
vec_sum += vec
tau_pt_val = vec_sum.Pt()
tau_energy_val = vec_sum.E()
tau_eta_val = vec_sum.Eta()
tau_phi_val = vec_sum.Phi()
for k, _ in enumerate(
event.genindex): # iterate through jet particles
if (event.genindex[k] == jet_idx):
if (event.genid[k] == 22):
photon_ID.append(1)
electron_ID.append(0)
hadron_ID.append(0)
elif (abs(event.genid[k]) == 11):
photon_ID.append(0)
electron_ID.append(1)
hadron_ID.append(0)
elif (abs(event.genid[k]) > 40):
photon_ID.append(0)
electron_ID.append(0)
hadron_ID.append(1)
else:
continue
# particle parameters
eta_val = event.genjeteta[jet_idx] - event.geneta[k]
phi_val = delta_phi(event.genjetphi[jet_idx],
event.genphi[k])
pt.append(event.genpt[k])
eta.append(eta_val)
phi.append(phi_val)
et.append(event.genet[k])
# jet parameters
jet_pt.append(event.genjetpt[jet_idx])
jet_eta.append(event.genjeteta[jet_idx])
jet_phi.append(event.genjetphi[jet_idx])
jet_et.append(event.genjetet[jet_idx])
jet_index.append(jet_num)
# tau parameters
tau_pt.append(tau_pt_val)
tau_eta.append(tau_eta_val)
tau_phi.append(tau_phi_val)
tau_energy.append(tau_energy_val)
particle_num += 1
jet_num += 1
event_num += 1.
progress.update_progress(event_num / total_num)
fields = [('pt', 'f8'), ('eta', 'f8'), ('phi', 'f8'), ('et', 'f8'),
('photon_ID', 'i4'), ('electron_ID', 'i4'), ('hadron_ID', 'i4'),
('jet_pt', 'f8'), ('jet_eta', 'f8'), ('jet_phi', 'f8'),
('jet_et', 'f8'), ('jet_index', 'i8'), ('tau_pt', 'f8'),
('tau_eta', 'f8'), ('tau_phi', 'f8'), ('tau_energy', 'f8')]
data = np.zeros(particle_num, dtype=fields)
data['pt'] = pt
data['eta'] = eta
data['phi'] = phi
data['et'] = et
data['photon_ID'] = photon_ID
data['electron_ID'] = electron_ID
data['hadron_ID'] = hadron_ID
data['jet_pt'] = jet_pt
data['jet_eta'] = jet_eta
data['jet_phi'] = jet_phi
data['jet_et'] = jet_et
data['tau_pt'] = tau_pt
data['tau_eta'] = tau_eta
data['tau_phi'] = tau_phi
data['tau_energy'] = tau_energy
data['jet_index'] = jet_index
return data