forked from chreissel/hepaccelerate
-
Notifications
You must be signed in to change notification settings - Fork 0
/
lib_analysis.py
358 lines (266 loc) · 13.5 KB
/
lib_analysis.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
import os, glob
import argparse
import json
import numpy as np
import uproot
import hepaccelerate
from hepaccelerate.utils import Results, NanoAODDataset, Histogram, choose_backend
NUMPY_LIB = None
ha = None
############################################## OBJECT SELECTION ################################################
### Primary vertex selection
def vertex_selection(scalars, mask_events):
PV_isfake = (scalars["PV_score"] == 0) & (scalars["PV_chi2"] == 0)
PV_rho = NUMPY_LIB.sqrt(scalars["PV_x"]**2 + scalars["PV_y"]**2)
mask_events = mask_events & (~PV_isfake) & (scalars["PV_ndof"] > 4) & (scalars["PV_z"]<24) & (PV_rho < 2)
return mask_events
### Lepton selection
def lepton_selection(leps, cuts):
passes_eta = (NUMPY_LIB.abs(leps.eta) < cuts["eta"])
passes_subleading_pt = (leps.pt > cuts["subleading_pt"])
passes_leading_pt = (leps.pt > cuts["leading_pt"])
if cuts["type"] == "el":
sca = NUMPY_LIB.abs(leps.deltaEtaSC + leps.eta)
passes_id = (leps.cutBased >= 4)
passes_SC = NUMPY_LIB.invert((sca >= 1.4442) & (sca <= 1.5660))
# cuts taken from: https://twiki.cern.ch/twiki/bin/view/CMS/CutBasedElectronIdentificationRun2#Working_points_for_92X_and_later
passes_impact = ((leps.dz < 0.10) & (sca <= 1.479)) | ((leps.dz < 0.20) & (sca > 1.479)) | ((leps.dxy < 0.05) & (sca <= 1.479)) | ((leps.dxy < 0.1) & (sca > 1.479))
#select electrons
good_leps = passes_eta & passes_leading_pt & passes_id & passes_SC & passes_impact
veto_leps = passes_eta & passes_subleading_pt & NUMPY_LIB.invert(good_leps) & passes_id & passes_SC & passes_impact
elif cuts["type"] == "mu":
passes_leading_iso = (leps.pfRelIso04_all < cuts["leading_iso"])
passes_subleading_iso = (leps.pfRelIso04_all < cuts["subleading_iso"])
passes_id = (leps.tightId == 1)
#select muons
good_leps = passes_eta & passes_leading_pt & passes_leading_iso & passes_id
veto_leps = passes_eta & passes_subleading_pt & passes_subleading_iso & passes_id & NUMPY_LIB.invert(good_leps)
return good_leps, veto_leps
### Jet selection
def jet_selection(jets, leps, mask_leps, cuts, jets_met_corrected):
jets_pass_dr = ha.mask_deltar_first(jets, jets.masks["all"], leps, mask_leps, cuts["dr"])
jets.masks["pass_dr"] = jets_pass_dr
if jets_met_corrected:
good_jets = (jets.pt_nom > cuts["pt"]) & (NUMPY_LIB.abs(jets.eta) < cuts["eta"]) & (jets.jetId >= cuts["jetId"]) & jets_pass_dr
if cuts["type"] == "jet":
good_jets &= ((jets.puId>=cuts["puId"]) | (jets.pt_nom > 50.))
#good_jets &= (jets.puId>=cuts["puId"])
else:
good_jets = (jets.pt > cuts["pt"]) & (NUMPY_LIB.abs(jets.eta) < cuts["eta"]) & (jets.jetId >= cuts["jetId"]) & jets_pass_dr
if cuts["type"] == "jet":
good_jets &= ((jets.puId>=cuts["puId"]) | (jets.pt > 50.))
#good_jets &= (jets.puId>=cuts["puId"])
return good_jets
###################################################### WEIGHT / SF CALCULATION ##########################################################
### PileUp weight
def compute_pu_weights(pu_corrections_target, weights, mc_nvtx, reco_nvtx):
pu_edges, (values_nom, values_up, values_down) = pu_corrections_target
src_pu_hist = get_histogram(mc_nvtx, weights, pu_edges)
norm = sum(src_pu_hist.contents)
src_pu_hist.contents = src_pu_hist.contents/norm
src_pu_hist.contents_w2 = src_pu_hist.contents_w2/norm
ratio = values_nom / src_pu_hist.contents
remove_inf_nan(ratio)
pu_weights = NUMPY_LIB.zeros_like(weights)
ha.get_bin_contents(reco_nvtx, NUMPY_LIB.array(pu_edges), NUMPY_LIB.array(ratio), pu_weights)
#fix_large_weights(pu_weights)
return pu_weights
def load_puhist_target(filename):
fi = uproot.open(filename)
h = fi["pileup"]
edges = np.array(h.edges)
values_nominal = np.array(h.values)
values_nominal = values_nominal / np.sum(values_nominal)
h = fi["pileup_plus"]
values_up = np.array(h.values)
values_up = values_up / np.sum(values_up)
h = fi["pileup_minus"]
values_down = np.array(h.values)
values_down = values_down / np.sum(values_down)
return edges, (values_nominal, values_up, values_down)
# lepton scale factors
def compute_lepton_weights(leps, lepton_x, lepton_y, mask_rows, mask_content, evaluator, SF_list):
weights = NUMPY_LIB.ones(len(lepton_x))
for SF in SF_list:
if SF == "el_triggerSF":
weights *= evaluator[SF](lepton_y, lepton_x)
else:
weights *= evaluator[SF](lepton_x, lepton_y)
per_event_weights = ha.multiply_in_offsets(leps, weights, mask_rows, mask_content)
return per_event_weights
# btagging scale factor
def compute_btag_weights(jets, mask_rows, mask_content, sf, jets_met_corrected, btagalgorithm, sys_type):
pJet_weight = NUMPY_LIB.ones(len(mask_content))
for tag in [0, 4, 5]:
if jets_met_corrected:
SF_btag = sf.eval(sys_type, tag, abs(jets.eta), jets.pt_nom, getattr(jets, btagalgorithm), ignore_missing=True)
else:
SF_btag = sf.eval(sys_type, tag, abs(jets.eta), jets.pt, getattr(jets, btagalgorithm), ignore_missing=True)
if tag == 5:
SF_btag[jets.hadronFlavour != 5] = 1.
if tag == 4:
SF_btag[jets.hadronFlavour != 4] = 1.
#SF_btag[jets.hadronFlavour == 4] = 1. #DIRTY FIX TO REMOVE WEIGHT CONTRIBUTIONS FROM C JETS! TO BE FIXED! ALSO WOULD BE WRONG FOR UNCERTAINTIES AS THEY ARE CALCULATED FOR C
if tag == 0:
SF_btag[jets.hadronFlavour != 0] = 1.
pJet_weight *= SF_btag
per_event_weights = ha.multiply_in_offsets(jets, pJet_weight, mask_rows, mask_content)
return per_event_weights
############################################# HIGH LEVEL VARIABLES (DNN evaluation, ...) ############################################
def evaluate_DNN(jets, good_jets, electrons, good_electrons, muons, good_muons, scalars, mask_events, nEvents, DNN, DNN_model, jets_met_corrected, outdir="./", btag_DNN = 'deepCSV'):
# choose btag
if btag_DNN == 'deepCSV':
b_choice = 'btagDeepB'
elif btag_DNN == 'CSVV2':
b_choice = 'btagCSVV2'
elif btag_DNN == 'deepFlav':
b_choice = 'btagDeepFlavB'
# make inputs (defined in backend (not extremely nice))
if jets_met_corrected:
jets_feats = ha.make_jets_inputs(jets, jets.offsets, 10, ["pt_nom","eta","phi","en","px","py","pz", b_choice], mask_events, good_jets)
met_feats = ha.make_met_inputs(scalars, nEvents, ["phi_nom","pt_nom","sumEt","px","py"], mask_events)
else:
jets_feats = ha.make_jets_inputs(jets, jets.offsets, 10, ["pt","eta","phi","en","px","py","pz", b_choice], mask_events, good_jets)
met_feats = ha.make_met_inputs(scalars, nEvents, ["phi","pt","sumEt","px","py"], mask_events)
leps_feats = ha.make_leps_inputs(electrons, muons, nEvents, ["pt","eta","phi","en","px","py","pz"], mask_events, good_electrons, good_muons)
if DNN == "save-arrays":
return jets_feats[mask_events==1], leps_feats[mask_events==1], met_feats[mask_events==1]
inputs = [jets_feats, leps_feats, met_feats]
if "fcn" in DNN:
batch_size = 100
else:
batch_size = 2000
if DNN.startswith("ffwd"):
inputs = [NUMPY_LIB.reshape(x, (x.shape[0], -1)) for x in inputs]
inputs = NUMPY_LIB.hstack(inputs)
# numpy transfer needed for keras
inputs = NUMPY_LIB.asnumpy(inputs)
if DNN.startswith("cmb") or DNN.startswith("mass"):
# numpy transfer needed for keras
if "prtrn" in DNN:
inputs = [inputs[0], inputs[1], inputs[2], inputs[0], inputs[1], inputs[2], inputs[0], inputs[1], inputs[2]]
if not isinstance(jets_feats, np.ndarray):
inputs = [NUMPY_LIB.asnumpy(x) for x in inputs]
if DNN.startswith("gnet"):
# implement function which converts jets leps and met into nodes, edges and mask
if not isinstance(jets_feats, np.ndarray):
inputs = [NUMPY_LIB.asnumpy(x) for x in inputs]
edges, nodes, mask = make_graph_input(jets_feats, leps_feats, met_feats)
inputs = [edges, nodes, mask]
if not isinstance(edges, np.ndarray):
inputs = [NUMPY_LIB.asnumpy(x) for x in inputs]
if "fcn" in DNN:
inputs = [NUMPY_LIB.reshape(x, (x.shape[0], -1)) for x in inputs]
inputs = NUMPY_LIB.hstack(inputs)
# numpy transfer needed for keras
inputs = NUMPY_LIB.asnumpy(inputs)
# fix in case inputs are empty
if jets_feats.shape[0] == 0 or DNN=='save-arrays':
DNN_pred = NUMPY_LIB.zeros(nEvents, dtype=NUMPY_LIB.float32)
else:
# run prediction (done on GPU)
# in case of NUMPY_LIB is cupy: transfer numpy output back to cupy array for further computation
DNN_pred = NUMPY_LIB.array(DNN_model.predict(inputs, batch_size = batch_size))
if 'categorical' in DNN:
DNN_pred = DNN_pred[:,1]
#if DNN.endswith("binary"):
# DNN_pred = NUMPY_LIB.reshape(DNN_pred, DNN_pred.shape[0])
print("DNN inference finished.")
if DNN == "mass_fit":
dijet_masses = ha.dijet_masses(jets_feats, mask_events, DNN_pred)
return dijet_masses
return DNN_pred
# calculate simple object variables
def calculate_variable_features(z, mask_events, indices, var):
name, coll, mask_content, inds, feats = z
idx = indices[inds]
for f in feats:
var[inds+"_"+name+"_"+f] = ha.get_in_offsets(getattr(coll, f), getattr(coll, "offsets"), idx, mask_events, mask_content)
####################################################### Simple helpers #############################################################
def make_graph_input(jets_ft, leps_ft, met_ft):
graph_jets = NUMPY_LIB.zeros((jets_ft.shape[0], jets_ft.shape[1], jets_ft.shape[2]+3))
graph_jets[:, :, :8] = jets_ft
graph_jets[:, :, 8] = 1
graph_leps = NUMPY_LIB.zeros((leps_ft.shape[0], leps_ft.shape[1], leps_ft.shape[2]+4))
graph_leps[:, :, :7] = leps_ft
graph_leps[:, :, 9] = 1
met_ft = NUMPY_LIB.expand_dims(met_ft, 1)
# swap phi and pt for met
pt_met = NUMPY_LIB.copy(met_ft[:, :, 1])
pt_phi = NUMPY_LIB.copy(met_ft[:, :, 0])
met_ft[:, :, 0] = pt_met
met_ft[:, :, 1] = pt_phi
graph_met = NUMPY_LIB.zeros((met_ft.shape[0], met_ft.shape[1], met_ft.shape[2]+6))
graph_met[:, :, 0] = met_ft[:, :, 0]
graph_met[:, :, 2:6] = met_ft[:, :, 1:]
graph_met[:, :, 10] = 1
full_nodes = NUMPY_LIB.concatenate((graph_jets, graph_leps, graph_met), axis=1)
full_edges = NUMPY_LIB.copy(full_nodes[:, :, 1:3])
full_mask = NUMPY_LIB.copy(full_nodes[:, :, 0])
full_mask = NUMPY_LIB.expand_dims(full_mask, 2)
return full_edges, full_nodes, full_mask
def get_histogram(data, weights, bins):
return Histogram(*ha.histogram_from_vector(data, weights, bins))
def remove_inf_nan(arr):
arr[np.isinf(arr)] = 0
arr[np.isnan(arr)] = 0
arr[arr < 0] = 0
def chunks(l, n):
"""Yield successive n-sized chunks from l."""
for i in range(0, len(l), n):
yield l[i:i + n]
import keras.backend as K
import keras.losses
import keras.utils.generic_utils
from Disco_tf import distance_corr
def mse0(y_true,y_pred):
return K.mean( K.square(y_true[:,0] - y_pred[:,0]) )
def mae0(y_true,y_pred):
return K.mean( K.abs(y_true[:,0] - y_pred[:,0]) )
def r2_score0(y_true,y_pred):
return 1. - K.sum( K.square(y_true[:,0] - y_pred[:,0]) ) / K.sum( K.square(y_true[:,0] - K.mean(y_true[:,0]) ) )
def decorr(var_1, var_2, weights, kappa):
def loss(y_true, y_pred):
return keras.losses.categorical_crossentropy(y_true, y_pred) + kappa * distance_corr(var_1, var_2, weights)
return loss
def dijet_feats(x):
# position depends on input array
en = x[:,:,3] + x[:,:,11]
px = x[:,:,4] + x[:,:,12]
py = x[:,:,5] + x[:,:,13]
pz = x[:,:,6] + x[:,:,14]
m = K.sqrt(en*en - px*px - py*py - pz*pz)
pt = K.sqrt(px*px + py*py)
phi = tf.math.acos(py/px)
theta = tf.math.acos(pz/(K.sqrt(px*px + py*py + pz*pz)))
eta = -K.log(tf.math.tan(theta/2))
pt = tf.reshape(pt, [-1,45,1])
eta = tf.reshape(eta, [-1,45,1])
phi = tf.reshape(phi, [-1,45,1])
m = tf.reshape(m, [-1,45,1])
return Concatenate(axis=2)([pt, eta, phi, m])
def trijet_feats(x):
# position depends on input array
en = x[:,:,3] + x[:,:,11] + x[:,:,19]
px = x[:,:,4] + x[:,:,12] + x[:,:,20]
py = x[:,:,5] + x[:,:,13] + x[:,:,21]
pz = x[:,:,6] + x[:,:,14] + x[:,:,22]
m = K.sqrt(en*en - px*px - py*py - pz*pz)
pt = K.sqrt(px*px + py*py)
phi = tf.math.acos(py/px)
theta = tf.math.acos(pz/(K.sqrt(px*px + py*py + pz*pz)))
eta = -K.log(tf.math.tan(theta/2))
pt = tf.reshape(pt, [-1,120,1])
eta = tf.reshape(eta, [-1,120,1])
phi = tf.reshape(phi, [-1,120,1])
m = tf.reshape(m, [-1,120,1])
return Concatenate(axis=2)([pt, eta, phi, m])
def lep_feats(x):
# position depends on input array
px = x[:,:,4] + x[:,:,10] + x[:,:,16]
py = x[:,:,5] + x[:,:,11] + x[:,:,17]
pt = K.sqrt(px*px + py*py)
phi = tf.math.acos(py/px)
pt = tf.reshape(pt, [-1,10,1])
phi = tf.reshape(phi, [-1,10,1])
return Concatenate(axis=2)([pt, phi])