-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathexperiments.py
516 lines (463 loc) · 22.5 KB
/
experiments.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
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
#!/usr/bin/env python
# coding: utf-8
'''
Experiments for ECML17
Author: Huang Xiao
Group: Cognitive Security Technologies
Institute: Fraunhofer AISEC
Mail: [email protected]
Copyright@2017
'''
import argparse
import json
from datetime import datetime
from itertools import product
from multiprocessing import cpu_count
from joblib import Parallel, delayed, dump, load
from evaluator import *
from gcdag import H3GCDiGraph
from helper import *
from drawing import *
from bokeh.plotting import show, gridplot
from Plottings.DataViz import DataViz
from sklearn.model_selection import KFold
# globals
datapath = 'Datasets/'
jsonpath = 'jsons/'
figpath = 'Figures/'
output_path = 'Results/'
# run joblib
def run_job_(inputs, outputs, output_idx, clf, Xtt=None, dag_true=None):
start = datetime.now()
clf.fit(inputs)
clf.fit_structure(clf.precision_)
delta = datetime.now() - start
if dag_true is not None:
errors, missings, extras, inverses = shd(dag_true, clf.conditional_independences_)
outputs[output_idx[0], output_idx[1], output_idx[2], 0] = errors
outputs[output_idx[0], output_idx[1], output_idx[2], 1] = delta.total_seconds()
outputs[output_idx[0], output_idx[1], output_idx[2], 2] = missings
outputs[output_idx[0], output_idx[1], output_idx[2], 3] = extras
outputs[output_idx[0], output_idx[1], output_idx[2], 4] = inverses
elif Xtt is not None:
errors = clf.score_function(Xtt) # log-likelihood per instance
outputs[output_idx[0], output_idx[1], output_idx[2], 0] = errors
outputs[output_idx[0], output_idx[1], output_idx[2], 1] = delta.total_seconds()
# EVALUATE REAL SAMPLES
def eval_realsamples(net=None,
csv_file='',
sizes = [100, 200, 500, 1000, 2000, 4000],
n_sizes=3,
folds=5,
penalty=0.01,
alpha=0.1,
methods=['mle', 'glasso', 'ledoit_wolf'],
save_path=output_path,
logger=logging.getLogger('gcdag.experiments.realsamples.log'),
overwrite=False,
n_jobs=-1):
input_data, node_names = read_csv(csv_file)
N, d = input_data.shape
if sizes:
sample_sizes = sizes
else:
sample_sizes = np.linspace(100, N, n_sizes, endpoint=True, dtype=int).tolist()
if n_jobs < 0:
results = np.zeros(shape=(folds, len(methods), len(sample_sizes), 5))
logger.debug('[Experiment sample sizes vs. per-instance log-likelihood]')
# No parallel
logger.debug(
'{:10s} {:10s} {:10s} {:10s} {:10s}'.format('Method', 'Fold', 'Size', '(perInst) log-ll', 'Time elapsed'))
for i, m in enumerate(methods):
clf = H3GCDiGraph(penalty=penalty, method=m, verbose=False, alpha=alpha, vnames=node_names)
for j, size in enumerate(sample_sizes):
bootstrap = input_data[np.random.choice(N, size)]
kfold = KFold(shuffle=True, n_splits=folds)
splits = kfold.split(bootstrap)
for fid, split in enumerate(splits):
train_idx, test_idx = split[0], split[1]
run_job_(bootstrap[train_idx], results, (fid, i, j), clf, dag_true=None, Xtt=bootstrap[test_idx])
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format(m, str(fid), str(size),
str(results[fid, i, j, 0]),
str(results[fid, i, j, 1])))
else:
results = np.memmap('tmp/joblib_temp.mmap',
dtype=float,
mode='w+',
shape=(folds, len(methods), len(sample_sizes), 5))
dump(input_data, 'tmp/input_data')
input_data = load('tmp/input_data', mmap_mode='r')
if n_jobs == 0:
n_jobs = cpu_count()
splits_set = []
bootstraps = []
for s in sample_sizes:
bootstraps.append(np.random.choice(N, s))
kfold = KFold(shuffle=True, n_splits=folds)
splits = kfold.split(input_data[bootstraps[-1]])
split_folds = []
for tr_id, tt_id in splits:
split_folds.append((tr_id, tt_id))
splits_set.append(split_folds)
Parallel(n_jobs=n_jobs)(delayed(run_job_)(inputs=input_data[bootstraps[s]][splits_set[s][f][0]],
outputs=results,
output_idx=(f, m, s),
clf=H3GCDiGraph(penalty=penalty,
method=methods[m],
verbose=False,
alpha=alpha,
vnames=node_names),
dag_true=None,
Xtt=input_data[bootstraps[s]][splits_set[s][f][1]])
for f, m, s in product(range(folds),
range(len(methods)),
range(len(sample_sizes))))
# save to pickle file
output = dict()
output['data'] = results
output['folds'] = folds
output['samplesizes'] = sample_sizes
output['methods'] = methods
output['nodes'] = node_names
output['parameters'] = {'penalty': penalty,
'alpha': alpha}
if overwrite:
with open(''.join([save_path, csv_file.split('/')[-1].split('_')[0], '_samplesizes', '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
else:
time_str = str(datetime.now())
with open(''.join([save_path, csv_file.split('/')[-1].split('_')[0], '_samplesizes', '_', time_str, '.p']),
'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
return fd.name
# EVALUATE SAMPLESIZES
def eval_samplesizes(net=None,
csv_file='',
sizes=range(100, 2100, 100),
folds=5,
penalty=0.01,
alpha=0.1,
methods=['mle', 'glasso', 'ledoit_wolf', 'ic'],
save_path=output_path,
logger=logging.getLogger('gcdag.experiments.samplesizes.log'),
overwrite=False,
n_jobs=-1):
# initialization
network_name = net
network = network_name + '.json'
dataset_name = network_name + '_' + str(10000)
dataset = dataset_name + '.csv'
input_data, node_names = read_csv(datapath + dataset)
dag_true = json2adj(json.load(open(jsonpath + network)), node_names=node_names)
samples_sizes = sizes
cv = folds
methods = methods
# init results array
# :,:,0,: for shd erro, :,:,1,: for runtime
if n_jobs < 0:
results = np.zeros(shape=(cv, len(methods), len(samples_sizes), 5))
logger.debug('[Experiment sample sizes vs. SHD error]')
# No parallel
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format('Method', 'Fold', 'Size', 'SHD', 'Time elapsed'))
for i, m in enumerate(methods):
clf = H3GCDiGraph(penalty=penalty, method=m, verbose=False, alpha=alpha, vnames=node_names)
for j, size in enumerate(samples_sizes):
bootstrap = input_data[np.random.choice(10000, size)]
for fold in range(cv):
run_job_(bootstrap, results, (fold, i, j), clf, dag_true=dag_true)
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format(m, str(fold), str(size),
str(results[fold, i, j, 0]),
str(results[fold, i, j, 1])))
else:
results = np.memmap('tmp/joblib_temp.mmap',
dtype=float,
mode='w+',
shape=(cv, len(methods), len(samples_sizes), 5))
dump(input_data, 'tmp/input_data')
input_data = load('tmp/input_data', mmap_mode='r')
if n_jobs == 0:
n_jobs = cpu_count()
Parallel(n_jobs=n_jobs)(delayed(run_job_)(inputs=input_data[np.random.choice(10000, samples_sizes[s])],
outputs=results,
output_idx=(f, m, s),
clf=H3GCDiGraph(penalty=penalty,
method=methods[m],
verbose=False,
alpha=alpha,
vnames=node_names),
dag_true=dag_true)
for f, m, s in product(range(cv),
range(len(methods)),
range(len(samples_sizes))))
# save to pickle file
output = dict()
output['data'] = results
output['folds'] = cv
output['samplesizes'] = samples_sizes
output['methods'] = methods
output['nodes'] = node_names
output['parameters'] = {'penalty': penalty,
'alpha': alpha}
if overwrite:
with open(''.join([save_path, net, '_samplesizes', '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
else:
time_str = str(datetime.now())
with open(''.join([save_path, net, '_samplesizes', '_', time_str, '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
return fd.name
# EVALUATE ALPHAS
def eval_alphas(net=None,
csv_file='',
size=1000,
folds=5,
penalty=0.01,
alphas=np.linspace(0, 0.5, 10, endpoint=True),
methods=['mle', 'glasso', 'ledoit_wolf'],
save_path=output_path,
logger=logging.getLogger('gcdag.experiments.alphas.log'),
overwrite=False,
n_jobs=-1):
# initialization
network_name = net
network = network_name + '.json'
dataset_name = network_name + '_' + str(10000)
dataset = dataset_name + '.csv'
input_data, node_names = read_csv(datapath + dataset)
dag_true = json2adj(json.load(open(jsonpath + network)), node_names=node_names)
cv = folds
methods = methods
# init results array
# :,:,0,: for shd erro, :,:,1,: for runtime
if n_jobs < 0:
results = np.zeros(shape=(cv, len(methods), len(alphas), 5))
logger.debug('[Experiment alphas vs. SHD error]')
# No parallel
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format('Method', 'Fold', 'alpha', 'SHD', 'Time elapsed'))
for i, m in enumerate(methods):
for j, alpha in enumerate(alphas):
clf = H3GCDiGraph(penalty=penalty, method=m, verbose=False, alpha=alpha, vnames=node_names)
for fold in range(cv):
bootstrap = input_data[np.random.choice(10000, size)]
run_job_(bootstrap, results, (fold, i, j), clf, dag_true=dag_true)
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format(m, str(fold), str(alpha),
str(results[fold, i, j, 0]),
str(results[fold, i, j, 1])))
else:
results = np.memmap('tmp/joblib_temp.mmap',
dtype=float,
mode='w+',
shape=(cv, len(methods), len(alphas), 5))
dump(input_data, 'tmp/input_data')
input_data = load('tmp/input_data', mmap_mode='r')
if n_jobs == 0:
n_jobs = cpu_count()
Parallel(n_jobs=n_jobs)(delayed(run_job_)(inputs=input_data[np.random.choice(10000, size)],
outputs=results,
output_idx=(f, m, a),
clf=H3GCDiGraph(penalty=penalty,
method=methods[m],
verbose=False,
alpha=alphas[a],
vnames=node_names),
dag_true=dag_true)
for f, m, a in product(range(cv),
range(len(methods)),
range(len(alphas))))
# save to pickle file
output = dict()
output['data'] = results
output['folds'] = cv
output['alphas'] = alphas
output['methods'] = methods
output['nodes'] = node_names
output['parameters'] = {'penalty': penalty,
'size': size}
if overwrite:
with open(''.join([save_path, net, '_alphas', '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
else:
time_str = str(datetime.now())
with open(''.join([save_path, net, '_alphas', '_', time_str, '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
# return the filename
return fd.name
# EVALUATE PENALTY
def eval_penalty(net=None,
csv_file='',
size=1000,
folds=5,
penalty=np.linspace(0, 0.2, 10, endpoint=True),
alpha=0.1,
methods=['mle', 'glasso'],
save_path=output_path,
logger=logging.getLogger('gcdag.experiments.penalty.log'),
overwrite=False,
n_jobs=-1):
# initialization
network_name = net
network = network_name + '.json'
dataset_name = network_name + '_' + str(10000)
dataset = dataset_name + '.csv'
input_data, node_names = read_csv(datapath + dataset)
dag_true = json2adj(json.load(open(jsonpath + network)), node_names=node_names)
cv = folds
methods = methods
# init results array
# :,:,0,: for shd erro, :,:,1,: for runtime
if n_jobs < 0:
results = np.zeros(shape=(cv, len(methods), len(penalty), 5))
logger.debug('[Experiment penalty vs. SHD error]')
# No parallel
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format('Method', 'Fold', 'penalty', 'SHD', 'Time elapsed'))
for i, m in enumerate(methods):
for j, pen in enumerate(penalty):
clf = H3GCDiGraph(penalty=pen, method=m, verbose=False, alpha=alpha, vnames=node_names)
for fold in range(cv):
bootstrap = input_data[np.random.choice(10000, size)]
run_job_(bootstrap, results, (fold, i, j), clf, dag_true=dag_true)
logger.debug('{:10s} {:10s} {:10s} {:10s} {:10s}'.format(m, str(fold), str(pen),
str(results[fold, i, j, 0]),
str(results[fold, i, j, 1])))
else:
results = np.memmap('tmp/joblib_temp.mmap',
dtype=float,
mode='w+',
shape=(cv, len(methods), len(penalty), 5))
dump(input_data, 'tmp/input_data')
input_data = load('tmp/input_data', mmap_mode='r')
if n_jobs == 0:
n_jobs = cpu_count()
Parallel(n_jobs=n_jobs)(delayed(run_job_)(inputs=input_data[np.random.choice(10000, size)],
outputs=results,
output_idx=(f, m, p),
clf=H3GCDiGraph(penalty=penalty[p],
method=methods[m],
verbose=False,
alpha=alpha,
vnames=node_names),
dag_true=dag_true)
for f, m, p in product(range(cv),
range(len(methods)),
range(len(penalty))))
# save to pickle file
output = dict()
output['data'] = results
output['folds'] = cv
output['penalty'] = penalty
output['methods'] = methods
output['nodes'] = node_names
output['parameters'] = {'alpha': alpha,
'size': size}
if overwrite:
with open(''.join([save_path, net, '_penalty', '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
else:
time_str = str(datetime.now())
with open(''.join([save_path, net, '_penalty', '_', time_str, '.p']), 'w') as fd:
print colored('Done! writting to ' + fd.name, 'red')
pickle.dump(output, fd)
# return the filename
return fd.name
# main ---
argparser = argparse.ArgumentParser(prog='experiments',
usage="this program conducts experiments for ECML17 paper.",
description='''
experiments code for ECML'17
''',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
argparser.add_argument("--net", help="network JSON file name ,e.g., credit_loan", default='credit_loan')
argparser.add_argument("--csv", help="use csv dataset for real dataset evaluation", default='')
argparser.add_argument("--eval_func", help="Evaluation function name")
argparser.add_argument("--folds", help="How many folds to conduct.. ", default=3, type=int)
argparser.add_argument("--save_path", help="Where to save experiments results.", default=output_path)
argparser.add_argument("--ncpus", help="How many cpus we use. defaults use all available CPUs.", type=int, default=0)
argparser.add_argument("--view", help="show the results in plotting.", action='store_true')
argparser.add_argument("--view_only", help="No run, only view.", action='store_true')
args = argparser.parse_args()
net = args.net
csv_file = args.csv
func = args.eval_func
save_path = args.save_path
view = args.view
view_only = args.view_only
ncpus = args.ncpus
folds = args.folds
logger = logging.getLogger('gcdag.experiments.log')
logging.basicConfig(format='[%(asctime)s] - %(name)s - %(levelname)s - %(message)s', )
logger.setLevel(logging.DEBUG)
xticks_key = {'eval_samplesizes': 'samplesizes',
'eval_alphas': 'alphas',
'eval_penalty': 'penalty',
'eval_realsamples': 'samplesizes'}
if csv_file:
output_html = ''.join([csv_file.split('/')[-1].split('.')[0], xticks_key[func], '.html'])
ylabel = 'log-likelihood per Instance'
saved_file = ''.join([save_path, csv_file.split('/')[-1].split('_')[0], '_', xticks_key[func], '.p'])
else:
output_html = ''.join([net, xticks_key[func], '.html'])
ylabel = 'SHD'
saved_file = ''.join([save_path, net, '_', xticks_key[func], '.p'])
conf = {'colormap': 'Set2_',
'width': 600,
'height': 400,
'output_file': output_html}
if not view_only:
saved_file = locals()[func](net=net,
csv_file=csv_file,
# sizes=[100, 500, 1000, 2000],
folds=folds,
# penalty=0.1,
# alpha=0.15,
# methods=['mle', 'glasso', 'ledoit_wolf', 'pc'],
save_path=save_path,
logger=logger,
overwrite=True,
n_jobs=ncpus)
if view:
with open(saved_file, 'r') as fd:
plotting_data = pickle.load(fd)
plotting = DataViz(config=conf)
f1 = plotting.fill_between(xticks=plotting_data[xticks_key[func]],
mean=plotting_data['data'][:, :, :, 0].mean(axis=0),
std=plotting_data['data'][:, :, :, 0].std(axis=0), xlabel=xticks_key[func],
ylabel=ylabel,
legend=plotting_data['methods'],
legend_orientation='horizontal',
legend_loc='top_right')
f2 = plotting.fill_between(xticks=plotting_data[xticks_key[func]],
mean=plotting_data['data'][:, :, :, 1].mean(axis=0),
std=plotting_data['data'][:, :, :, 1].std(axis=0), xlabel=xticks_key[func],
ylabel='Runtime',
legend=plotting_data['methods'],
legend_orientation='horizontal',
legend_loc='top_right')
show(gridplot([f1, f2], ncols=2))
plotting.send_to_server()
else:
with open(saved_file, 'r') as fd:
plotting_data = pickle.load(fd)
plotting = DataViz(config=conf)
f1 = plotting.fill_between(xticks=plotting_data[xticks_key[func]],
mean=plotting_data['data'][:, :, :, 0].mean(axis=0),
std=plotting_data['data'][:, :, :, 0].std(axis=0), xlabel=xticks_key[func],
ylabel=ylabel,
legend=plotting_data['methods'],
legend_orientation='horizontal',
legend_loc='top_right')
f2 = plotting.fill_between(xticks=plotting_data[xticks_key[func]],
mean=plotting_data['data'][:, :, :, 1].mean(axis=0),
std=plotting_data['data'][:, :, :, 1].std(axis=0), xlabel=xticks_key[func],
ylabel='Runtime',
legend=plotting_data['methods'],
legend_orientation='horizontal',
legend_loc='top_right')
show(gridplot([f1, f2], ncols=2))
plotting.send_to_server()