-
Notifications
You must be signed in to change notification settings - Fork 0
/
mainframe.py
413 lines (320 loc) · 13.7 KB
/
mainframe.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
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
from os import path
import mods.ROOTmanager as manager
import mods.configuration as config
import ROOT.gSystem as ROOTgSystem
#ROOTgSystem.Load('/nfs/slac/g/ldmx/users/${USER}/ldmx-sw/install/lib/libFramework.so')
ROOTgSystem.Load('libFramework.so')
def main():
# Parse command line args and init most important ones
pdict = config.parse()
action_str = pdict['action']
configFile = pdict['config']
if action_str == 'trees': actor = BdtTreeMaker
elif action_str == 'train': actor = BdtTrainer
elif action_str == 'eval': actor = BdtEval
else: quit('\nProvide a valid action')
print('\nUsing {} action from conf file: {}'.format(action_str,configFile))
# If batch, just do the thing and exit
if pdict['batch'] and action_str != 'train':
actor( config.parse_batch_config(pdict) ).run()
quit('\nDONE!')
# Parse an print Config and (Overrides parse options if provided in both)
# Maybe give an overriding message in these cases
proc_confs = config.parse_bdt_config(
action_str,
configFile,
clargs = pdict
)
# Wait for confirmation
if not pdict['batch']:
input('\nPress Enter to continue... ( Also hi, have a good day :D )')
# Construct processes
# Mayhaps multiprocessing for this?
procs = []
for proc_config in proc_confs:
procs.append( actor( proc_config ) )
# Process processes
for proc in procs:
# RUN
proc.run(
strEvent = pdict['startEvent'],
maxEvents = pdict['maxEvents'],
pfreq = 1000
)
del proc # Done with process, remove from memory
print('\nU DONE GOOD UWU !\n')
##################################################
# "Actors"
##################################################
class BdtTreeMaker(manager.TreeProcess):
"""
Make flat trees to train on
Consider multiple inheritance later
"""
def __init__(self, proc_conf):
""" Set up TreeMaker, define event_proces, and init TreeProccess """
# Set up tfMaker to have branches_info
self.tfMaker = manager.TreeMaker(proc_conf.tConfig)
# Get set of branches
branches = set()
for fun_info in proc_conf.tConfig.funcs.values():
for br_tup in fun_info['brs']:
branches.add(br_tup)
# Init parent TreeProcess (sets attrs used in next funcs loop)
super().__init__(
proc_conf.pConfig,
branches = branches,
endfs=[ self.tfMaker.wq ]
)
# Get lists of dicts containing funcs, args, and priorities
# for use in appropriate event_process sections
func_groups = {
'init': ( 0, 2),
'closing': (40,50)
}
dets = ('tracker', 'ecal', 'hcal')
steps = ('_init', '_l1', '_med', '_l2', '_closing')
for det in dets:
for step in steps:
func_groups[ det + step ] =\
(10*(dets.index(det)+1) + steps.index(step),
10*(dets.index(det)+1) + steps.index(step) + 1)
# Begin with list containin dummy to simplify appending condition
for g in func_groups:
setattr(self, g, [{'func': None}])
# Check each function
for fun, info in proc_conf.tConfig.funcs.items():
# Against func_groups and determine which it belongs in
for lab, lims in func_groups.items():
if lims[0] <= info['priority'] < lims[1]:
g = lab
break
# If it's new, add its info to its group
if fun not in [ f for f in getattr(self,g) ]:
getattr(self,g).append( {
'func': fun,
'args': { tup[0]: getattr(self, tup[0]) \
for tup in info['brs'] },
'priority': info['priority']
}
)
# Sort function groups in case of an internal hierarchy, then pop prio
for g in func_groups:
getattr(self,g).pop(0)
if getattr(self,g) != []:
getattr(self,g).sort(key = lambda d: d['priority'])
for fund in getattr(self,g):
fund.pop('priority')
def doFeatFuncs(f_dict, funcs_list, store_dict={}, lq=None):
""" Short function for looping over others in funcs lists """
for fun in funcs_list:
fun['func']( f_dict, fun['args'], store_dict, lq)
def doDetector(f_dict, det, lb_prefix, d_store):
""" Do all doFeatFuncs steps for a given detector """
# Init loop items
doFeatFuncs(f_dict, getattr(self, det + '_init'), d_store)
if getattr(self, det + '_l1') != []:
# Get input branch name
for d in getattr(self, det + '_l1'):
for br in d['args']:
if br[:len(lb_prefix)] == lb_prefix: loop_branch = br
# Loop over said branch while doing lfs
for hit in getattr(self, loop_branch):
doFeatFuncs(
f_dict,
getattr(self, det + '_l1'),
d_store,
hit
)
# Do any intermediary functions
doFeatFuncs(f_dict, getattr(self, det + '_med'), d_store)
# Loop again if needed
if getattr(self, det + '_l2') != []:
for hit in getattr(self, loop_branch):
doFeatFuncs(
f_dict,
getattr(self, det + '_l2'),
d_store,
hit
)
# Any further det functions
doFeatFuncs(f_dict, getattr(self, det + '_closing'), d_store)
# Main event algorithm
def event_process():
""" Algorithm for computing and storing all features """
# Initialize BDT input variables w/ defaults
feats = self.tfMaker.resetFeats()
# Copy from input and other basiic assignments
g_store = {'ebeam': proc_conf.pConfig.ebeam}
doFeatFuncs(feats, self.init, g_store)
# Tell detectors about info so far
tracker_store = {'globals': g_store}
ecal_store = tracker_store.copy()
hcal_store = tracker_store.copy()
# Tracker
doDetector(feats, 'tracker', 'TrackerRecHits', tracker_store) # ?
# Ecal
doDetector(feats, 'ecal', 'EcalRecHits', ecal_store)
# Hcal
doDetector(feats, 'hcal', 'HcalRecHits', hcal_store)
# Any final closing functions
doFeatFuncs(feats, self.closing)
self.tfMaker.fillEvent(feats)
# Tell self about event_process
setattr(self, 'event_process', event_process)
# Light-hearted attempt to save on memory
del branches
del dets
del steps
del func_groups
class BdtTrainer():
""" Train a BDT """
def __init__(self, conf_dict):
""" Set up BDT parameters """
# Print config
config.print_dict(conf_dict, prefix='\nBDT configuration:')
# Set config items as attrs
for k,v in conf_dict.items():
setattr(self, k, v)
# Warning if things haven't been hadded
# Could use normal TChain method (as suggested by use of 'dir') but nah
# Forcing hadding is good incase things crash anyway
self.sets = ('bkg', 'sig')
for st in self.sets:
thingy = getattr(self, f'{st}_indir')
if not path.exists(thingy):
quit(
f'{getattr(self, f"{st}_indir")} does NOT exist'
+ 'Lemme help, try:\n'
+ f'ldmx hadd {thingy} '
+ f'{ "/".join(thingy.split("/")[:-1]) }/{{{st}}}/*'
+ '\nand then try me again'
)
# Yet another conf dictionary
self.params_dict = {
"objective": "binary:logistic",
"eta": self.eta,
"max_depth": self.tree_depth,
"min_child_weight": 20,
"silent": 1,
"subsample": .9,
"colsample_bytree": .85,
"eval_metric": 'error',
"seed": 1,
"nthread": 1,
"verbosity": 1,
"early_stopping_rounds" : 10
}
def run(self, strEvent=None, maxEvents=1.25e6 , pfreq=None):
""" Run the traning - startEvent and pfreq are placeholders """
# import some stuff
import os
import logging
import numpy as np
import pickle as pkl
import xgboost as xgb
import matplotlib as plt
# Seed and logging
np.random.seed( int(self.seed) )
ml_logger = logging.getLogger('matplotlib')
ml_logger.setLevel(logging.WARNING)
plt.use('Agg')
# Organize data for training
for st in self.sets:
# Load tree
tree = manager.load(
[getattr(self, '{}_indir'.format(st))],
self.tree_name
)
events = []
for event in tree:
if len(events) == maxEvents: break
events.append(
[ getattr(event, feat) for feat in self.branches ]
)
events = np.array(events)
new_idx = np.random.permutation(
np.arange( np.shape(events)[0] )
)
np.take(events, new_idx, axis = 0, out=events)
setattr(self, '{}_train_x'.format(st), events)
setattr(self,
'{}_train_y'.format(st),
np.zeros(
len(
getattr( self, '{}_train_x'.format(st) )
)
) + (st == 'sig')
)
# Combine data
train_x = np.vstack(( self.sig_train_x, self.bkg_train_x ))
train_y = np.append( self.sig_train_y, self.bkg_train_y )
train_x[ np.isnan( train_x ) ] = 0.000
train_y[ np.isnan( train_y ) ] = 0.000
training_matrix = xgb.DMatrix(train_x, train_y)
# Actual training
gbm = xgb.train(
self.params_dict,
training_matrix,
int(self.tree_number)
)
# Store BDT
outname = self.outdir.split('/')[-1]
if not os.path.exists(self.outdir):
print( 'Creating %s' % (self.outdir) )
os.makedirs(self.outdir)
output = open('{}/{}_weights.pkl'.format(self.outdir, outname), 'wb')
pkl.dump(gbm, output)
# Plot feature importances
xgb.plot_importance(gbm)
plt.pyplot.savefig(
'{}/{}_fimportance.png'.format(self.outdir, outname), # png name
dpi=500, bbox_inches='tight', pad_inches=0.5 # png parameters
)
# Anounce save location
print('Files saved in: {}'.format(self.outdir))
class BdtEval(manager.TreeProcess):
""" Evaluate BDT on reserved flat trees """
def __init__(self, proc_conf):
# Set up tfMaker to have branches_info
self.tfMaker = manager.TreeMaker(proc_conf.tConfig)
# Init parent TreeProcess (sets attrs used in next funcs loop)
super().__init__(
proc_conf.pConfig,
endfs=[ self.tfMaker.wq ]
)
# import some stuff
import numpy as np
import pickle as pkl
import xgboost as xgb
from mods.feats import trees_info_analysis
# Pretty much always want to pass this for bias plots
# Could be handled better, but skipping the config is nice
analysis_vars = tuple( trees_info_analysis.keys() )
# Set bdt
self.bdt = pkl.load( open( proc_conf.pConfig.bdt, 'rb' ) )
# Store discValue name
for k in self.tfMaker.branches_info.keys():
if k[:9] == 'discValue': discValue_name = k
# Don't try to include these in the list given to self.bdt.predict
no_eval = (*analysis_vars, discValue_name)
# Main event algorithm
def event_process():
# Collect features from this event
feats = []
for feat in self.tfMaker.branches_info:
if feat in no_eval: continue
feats.append( getattr( self.tree, feat ) )
# Copy existing variables to new tree
for f_name in self.tfMaker.branches_info.keys():
if f_name == discValue_name: continue
self.tfMaker.branches[f_name][0] = getattr( self.tree, f_name )
# Add prediction to new tree
evtarray = np.array([feats])
self.tfMaker.branches[discValue_name][0] = \
float( self.bdt.predict( xgb.DMatrix(evtarray) )[0] )
# Fill new tree with current event values
self.tfMaker.tree.Fill()
# Tell self about event_process
setattr(self, 'event_process', event_process)
if __name__ == '__main__': main()