def recalculate(eventWise_path):
eventWise = Components.EventWise.from_file(eventWise_path)
n_events = len(eventWise.X)
#n_events = 200
mean_distance = np.full((n_events, max_particles), np.nan)
dimension_remove = np.zeros((n_events, max_particles))
dimension_add = np.zeros((n_events, max_particles))
merged_change = np.zeros((n_events, max_particles))
eigenvalues = np.full((n_events, max_particles, max_particles), np.nan)
break_point = np.full(n_events, -1)
for event_n in range(n_events):
if event_n % 100 == 0:
print(f"{event_n/n_events:%} ", end='\r', flush=True)
eventWise.selected_event = event_n
# make the jets
jets = FormJets.SpectralFull(eventWise,
assign=False,
dict_jet_params=spectral_jet_params)
# get the initial embedding
embedding = np.copy(jets._eigenspace)
num_particles = embedding.shape[0]
num_dims = embedding.shape[1]
if len(jets._distances2):
mean = np.nanmean(
np.sqrt(jets._distances2[np.tril_indices_from(
jets._distances2, -1)]))
else:
continue
mean_distance[event_n, jets.currently_avalible] = mean
p = 1
while jets.currently_avalible:
p = jets.currently_avalible
#p+=1
np.fill_diagonal(jets._distances2, np.inf)
row, column = np.unravel_index(np.argmin(jets._distances2),
jets._distances2.shape)
jets._merge_pseudojets(row, column, jets._distances2[row, column])
new_embedding = np.copy(jets._eigenspace)
new_mean = np.nanmean(
np.sqrt(jets._distances2[np.tril_indices_from(
jets._distances2, -1)]))
if new_mean > R and break_point[event_n] < 0:
break_point[event_n] = p
mean_distance[event_n, p - 1] = new_mean
new_num_dims = jets._eigenspace.shape[1]
eigenvalues[event_n, p, :new_num_dims] = jets.eigenvalues[-1]
if new_num_dims > num_dims:
lower_dim = jets.eigenspace_distance2(
new_embedding[:, :num_dims], new_embedding[:, :num_dims],
[], [])
lower_mean = np.nanmean(
np.sqrt(lower_dim[np.tril_indices_from(lower_dim, -1)]))
dimension_shift = new_mean - lower_mean
dimension_add[event_n, p] = dimension_shift
elif new_num_dims < num_dims:
# old embedding had more particles
lower_dim = jets.eigenspace_distance2(
embedding[:, :new_num_dims], embedding[:, :new_num_dims],
[], [])
lower_mean = np.nanmean(
np.sqrt(lower_dim[np.tril_indices_from(lower_dim, -1)]))
dimension_shift = lower_mean - mean
dimension_remove[event_n, p] = dimension_shift
remaining_shift = new_mean - mean - dimension_shift
merged_change[event_n, p] = remaining_shift
mean = new_mean
embedding = new_embedding
num_dims = new_num_dims
return mean_distance, dimension_remove, dimension_add, merged_change, break_point, eigenvalues
def get_data(eventWise):
spectral_jet_params = dict(
ExpofPTMultiplier=0,
ExpofPTPosition='input',
ExpofPTFormat='Luclus',
NumEigenvectors=np.inf,
StoppingCondition='meandistance',
EigNormFactor=0.,
EigenvalueLimit=0.4,
# BaseJump=0.05,
# JumpEigenFactor=10,
# MaxCutScore=0.2,
Laplacien='symmetric',
DeltaR=1.26,
AffinityType='exponent',
AffinityExp=2.,
Sigma=.15,
CombineSize='sum',
EigDistance='abscos',
CutoffKNN=5,
CutoffDistance=None,
PhyDistance='angular')
jets = FormJets.SpectralFull(eventWise,
assign=False,
dict_jet_params=spectral_jet_params)
solution = eventWise.Solution
source = eventWise.JetInputs_SourceIdx
truth = np.fromiter((next(j for j, jet in enumerate(solution) if i in jet)
for i in source),
dtype=int)
affinities = [jets._affinity]
energies = [jets.Energy]
pts = [jets.PT]
rapidities = [jets.Rapidity]
phis = [jets.Phi]
showers = [truth]
old_phys_space = np.array(list(zip(rapidities[-1], phis[-1])))
num_showers = np.max(truth) + 1
input_idxs = [jets.InputIdx[:jets.currently_avalible]]
embedding_coordinates = [jets._eigenspace]
continuations = []
joins = []
embedding_continuations = []
continuation_showers = []
embedding_joins = []
join_types = []
while True:
jets._step_assign_parents()
if jets.currently_avalible == 0:
break
input_idxs.append(jets.InputIdx[:jets.currently_avalible])
carried_to = np.fromiter((i in input_idxs[-2] for i in input_idxs[-1]),
dtype=bool)
carried_from = np.fromiter(
(i in input_idxs[-1] for i in input_idxs[-2]), dtype=bool)
new_embedding = jets._eigenspace
new_showers = np.empty(new_embedding.shape[0], dtype=int)
new_showers[carried_to] = showers[-1][carried_from]
merged_showers = showers[-1][~carried_from]
heavier = np.argmax(energies[-1][~carried_from])
good_merge = len(set(merged_showers)) == 1
join_types.append(good_merge)
# in a good merge it won't matter
new_showers[~carried_to] = merged_showers[heavier]
showers.append(new_showers)
energies.append(jets.Energy)
pts.append(jets.PT)
rapidities.append(jets.Rapidity)
phis.append(jets.Phi)
phys_space = np.array(list(zip(rapidities[-1], phis[-1])))
affinities.append(jets._affinity)
orientations = FormJets.parity(input_idxs[-2],
embedding_coordinates[-1],
input_idxs[-1], new_embedding)
new_embedding[:, :len(orientations)] *= orientations[np.newaxis]
embedding_coordinates.append(new_embedding)
continuations += [[[
old_phys_space[~carried_from][0], phys_space[~carried_to][0]
]], [[old_phys_space[~carried_from][1], phys_space[~carried_to][0]]]]
embedding_continuations += [
list(
zip(embedding_coordinates[-2][carried_from],
embedding_coordinates[-1][carried_to])) + [[
embedding_coordinates[-2][~carried_from][heavier],
embedding_coordinates[-1][~carried_to][0]
]]
]
continuation_showers += [
list(showers[-2][carried_from]) +
[merged_showers[0], merged_showers[1]]
]
embedding_joins += [[
embedding_coordinates[-2][~carried_from][0],
embedding_coordinates[-2][~carried_from][1]
]]
joins += [[
old_phys_space[~carried_from][0], old_phys_space[~carried_from][1]
]]
old_phys_space = phys_space
data = {
'num_showers': num_showers,
'PT': pts,
'Rapidity': rapidities,
'Phi': phis,
'Energy': energies,
'affinity': affinities,
'embed_coords': embedding_coordinates,
'shower': showers,
'embed_cont': embedding_continuations,
'cont': continuations,
'cont_shower': continuation_showers,
'join': joins,
'embed_join': embedding_joins,
'join_type': join_types,
}
return data