-
Notifications
You must be signed in to change notification settings - Fork 3
/
OCPDL_benchmark.py
524 lines (452 loc) · 21.8 KB
/
OCPDL_benchmark.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
517
518
519
520
521
522
523
524
from utils.BCD_DR import ALS_DR
from utils.ocpdl import Online_CPDL
import numpy as np
import matplotlib.pyplot as plt
import pickle
from itertools import cycle
from scipy.interpolate import interp1d
import os
from headlines_preprocessing import generate_tensor
plt.rcParams['font.family'] = 'serif'
plt.rcParams['font.serif'] = ['Times New Roman'] + plt.rcParams['font.serif']
def initialize_loading(data_shape=[10,10,10], n_components=5, scale=1):
### loading = python dict of [U1, U2, \cdots, Un], each Ui is I_i x R array
loading = {}
for i in np.arange(len(data_shape)): # n_modes = X.ndim -1 where -1 for the last `batch mode'
loading.update({'U' + str(i): scale*np.random.rand(data_shape[i], n_components)})
return loading
def out(loading, drop_last_mode=False):
### given loading, take outer product of respected columns to get CPdict
### Use drop_last_mode for ALS
CPdict = {}
n_components = loading.get("U0").shape[1]
for i in np.arange(n_components):
A = np.array([1])
if drop_last_mode:
n_modes_multiplied = len(loading.keys()) - 1
else:
n_modes_multiplied = len(loading.keys()) # also equals self.X_dim - 1
for j in np.arange(n_modes_multiplied):
loading_factor = loading.get('U' + str(j)) ### I_i X n_components matrix
# print('loading_factor', loading_factor)
A = np.multiply.outer(A, loading_factor[:, i])
A = A[0]
CPdict.update({'A' + str(i): A})
return CPdict
def Out_tensor(loading):
### given loading, take outer product of respected columns to get CPdict
CPdict = out(loading, drop_last_mode=False)
recons = np.zeros(CPdict.get('A0').shape)
for j in np.arange(len(CPdict.keys())):
recons += CPdict.get('A' + str(j))
return recons
def ALS_run(X,
n_components=10,
iter=100,
regularizer=None, # L1 regularizer for each factor matrix
ini_loading=None,
beta=None,
search_radius_const=1000,
if_compute_recons_error=True,
save_folder='Output_files',
subsample_ratio = None,
output_results=True):
ALSDR = ALS_DR(X=X,
n_components=n_components,
ini_loading=None,
ini_A=None,
ini_B=None,
alpha=regularizer)
result_dict = ALSDR.ALS(iter=iter,
ini_loading=ini_loading,
beta=beta,
search_radius_const=None,
if_compute_recons_error=if_compute_recons_error,
save_folder=save_folder,
subsample_ratio=subsample_ratio,
output_results=output_results)
return result_dict
def MU_run(X,
n_components=10,
iter=100,
regularizer=0,
ini_loading=None,
if_compute_recons_error=True,
save_folder='Output_files',
output_results=True):
ALSDR = ALS_DR(X=X,
n_components=n_components,
ini_loading=None,
ini_A=None,
ini_B=None,
alpha=regularizer)
result_dict = ALSDR.MU(iter=iter,
ini_loading=ini_loading,
if_compute_recons_error=if_compute_recons_error,
save_folder=save_folder,
output_results=output_results)
return result_dict
def OCPDL_run(X,
n_components=10,
iter=100,
regularizer=0,
ini_loading=None,
batch_size=100,
beta=None,
mode_2be_subsampled=-1,
search_radius_const = None,
if_compute_recons_error=True,
save_folder='Output_files',
output_results=True):
OCPDL = Online_CPDL(X=X,
batch_size=batch_size,
iterations=iter,
n_components=n_components,
ini_loading=ini_loading,
ini_A=None,
ini_B=None,
alpha=regularizer,
beta=beta,
subsample=True)
result_dict = OCPDL.train_dict(mode_2be_subsampled=mode_2be_subsampled,
if_compute_recons_error=if_compute_recons_error,
search_radius_const = search_radius_const,
save_folder=save_folder,
output_results=output_results)
return result_dict
def plot_benchmark_errors(full_result_list, save_folder):
time_records = []
errors = []
f_interpolated_list = []
search_radius_const_list = []
# max duration and time records
x_all_max = 0
for i in np.arange(len(full_result_list)):
errors0 = full_result_list[i].get('timed_errors_trials')
x_all_max = max(x_all_max, max(errors0[:, :, -1][:, 0]))
x_all = np.linspace(0, x_all_max, num=101, endpoint=True)
for i in np.arange(len(full_result_list)):
errors0 = full_result_list[i].get('timed_errors_trials')
time_records.append(x_all[x_all < min(errors0[:, :, -1][:, 0])])
for i in np.arange(len(full_result_list)):
c = full_result_list[i].get('search_radius_const')
if c is not None:
search_radius_const_list.append(c)
search_radius_const_list = list(set(search_radius_const_list))
print('search_radius_const_list', search_radius_const_list)
# interpolate data and have common carrier
for i in np.arange(len(full_result_list)):
errors0 = full_result_list[i].get('timed_errors_trials')
f0_interpolated = []
for j in np.arange(errors0.shape[0]): # trials for same setting
f0 = interp1d(errors0[j, 0, :], errors0[j, 1, :], fill_value="extrapolate")
x_all_0 = time_records[i]
f0_interpolated.append(f0(x_all_0))
f0_interpolated = np.asarray(f0_interpolated)
f_interpolated_list.append(f0_interpolated)
# make figure
for c in search_radius_const_list:
plot_idx = 0
color_list = ['g', 'k', 'r', 'c', 'b']
marker_list = ['*', '|', 'x', 'o', '+']
fig, axs = plt.subplots(nrows=1, ncols=1, figsize=(6, 5))
for i in np.arange(len(full_result_list)):
result_dict = full_result_list[i]
search_radius_const = result_dict.get('search_radius_const')
if not ((result_dict.get("method") == "OCPDL") and (c != search_radius_const)):
print('search_radius_const', search_radius_const)
f0_interpolated = f_interpolated_list[i]
f_avg0 = np.sum(f0_interpolated, axis=0) / f0_interpolated.shape[0] ### axis-0 : trials
f_std0 = np.std(f0_interpolated, axis=0)
x_all_0 = time_records[i]
color = color_list[plot_idx % len(color_list)]
marker = marker_list[plot_idx % len(marker_list)]
plot_idx += 1
beta = result_dict.get("beta")
if (beta is None) and (result_dict.get("method")=="OCPDL"):
label0 = result_dict.get("method") + " ($\\beta=$None)"
elif (beta is None):
label0 = result_dict.get("method")
else:
label0 = result_dict.get("method") + " ($\\beta=${}, $c'=${})".format(beta, search_radius_const)
markers, caps, bars = axs.errorbar(x_all_0, f_avg0, yerr=f_std0,
fmt=color+'-', marker=marker, label=label0, errorevery=5)
axs.fill_between(x_all_0, f_avg0 - f_std0, f_avg0 + f_std0, facecolor=color, alpha=0.1)
# min_max duration
x_all_min_max = []
for i in np.arange(len(time_records)):
x_all_ALS0 = time_records[i]
x_all_min_max.append(max(x_all_ALS0))
x_all_min_max = min(x_all_min_max)
axs.set_xlim(0, x_all_min_max)
[bar.set_alpha(0.5) for bar in bars]
# axs.set_ylim(0, np.maximum(np.max(f_OCPDL_avg + f_OCPDL_std), np.max(f_ALS_avg + f_ALS_std)) * 1.1)
axs.set_xlabel('Elapsed time (s)', fontsize=13)
axs.set_ylabel('Relative recons. error', fontsize=13)
data_name = full_result_list[0].get('data_name')
title = data_name
plt.suptitle(title, fontsize=13)
axs.legend(fontsize=13)
plt.tight_layout()
plt.subplots_adjust(0.15, 0.1, 0.9, 0.9, 0.00, 0.00)
# file_name
n_trials = full_result_list[0].get("num_trials")
n_components = full_result_list[0].get('n_components')
search_radius_const = c
data_name = full_result_list[0].get('data_name')
save_path = save_folder + "/full_result_error_plot" + '_ntrials_' + str(n_trials) + "_" + "_ncomps_" + str(
n_components) + "_src_" + str(search_radius_const) + "_" + str(data_name) + ".pdf"
directory = os.path.dirname(save_path)
if not os.path.exists(directory):
os.makedirs(directory)
plt.savefig(save_path)
def main(n_components = 5,
iter = 10,
num_repeat = 10,
save_folder = "Output_files/JMLR_OCPDL3",
data_name = "Synthetic", # "Synthetic" or "Twitter"
run_ALS = True,
run_MU = True,
run_OCPDL = True,
plot_errors = True,
search_radius_const_list = [1,10,100,1000]):
# Load data
loading = {}
if data_name == "Synthetic":
scale = 0.01
np.random.seed(1)
#U0 = np.random.rand(20, 1*n_components)
#U0 = np.random.randint(2,size=[20, 5*n_components])
U0 = np.random.rand(100, 10*n_components)
np.random.seed(2)
U1 = np.random.rand(100, 10*n_components)
np.random.seed(3)
U2 = np.random.rand(100, 10*n_components)
loading.update({'U0': U0})
loading.update({'U1': U1})
loading.update({'U2': U2})
X = Out_tensor(loading) * scale
elif data_name == "Twitter":
path = "Data/Twitter/top_1000_daily/data_tensor_top1000.pickle"
dict = pickle.load(open(path, "rb"))
X = dict[1]
file_name = "Twitter"
elif data_name == "20Newsgroups":
# Get 20news_tfidf_tensor.pickle from https://github.com/lara-kassab/dynamic-tensor-topic-modeling/tree/main/data
# and place it in Data folder before running this code.
path = "Data/20news_tfidf_tensor.pickle"
dict = pickle.load(open(path, "rb"))
X = dict
file_name = "20Newsgroups"
elif data_name == "Headlines":
# in order to generate the tensor pickle file (~8GB),
# first read [abcnews-date-text.csv (60.2MB)] from data folder
# and generate tf-idf tensor (~8GB)
#path = "Data/20news_tfidf_tensor.pickle"
#dict = pickle.load(open(path, "rb"))
#X = dict
full_tensor = generate_tensor("Data/abcnews-date-text.csv", if_save=False)
X = full_tensor[1] # shape [203, 7000, 700]
file_name = "20Newsgroups"
print('X.shape', X.shape)
# X = 10 * X/np.linalg.norm(X)
#search_radius_const = 100
#search_radius_const_list = [1,10,100]
# search_radius_const = int(np.linalg.norm(X.reshape(-1,1),1))
loading_list = []
scale_X = 1/np.linalg.norm(X.reshape(-1,1))
for i in np.arange(num_repeat):
loading_list.append(initialize_loading(data_shape=X.shape, n_components=n_components, scale=scale_X))
full_result_list = []
if run_ALS:
# beta_list = [1 / 2, 1, None]
beta_list = [None]
for beta in beta_list:
print('!!! ALS initialized with beta:', beta)
list_full_timed_errors = []
results_dict = {}
for i in np.arange(num_repeat):
result_dict_ALS = ALS_run(X,
n_components=n_components,
iter=iter,
regularizer=0,
# inverse regularizer on time mode (to promote long-lasting topics),
# no regularizer on on words and tweets
ini_loading=loading_list[i],
beta=beta,
search_radius_const=None,
subsample_ratio=None,
if_compute_recons_error=True,
save_folder=save_folder,
output_results=True)
time_error = result_dict_ALS.get('time_error')
list_full_timed_errors.append(time_error.copy())
# print('!!! list_full_timed_errors', len(list_full_timed_errors))
timed_errors_trials = np.asarray(
list_full_timed_errors) # shape (# trials) x (2 for time, error) x (iterations)
if beta is None:
results_dict.update({"method": "ALS"})
else:
results_dict.update({"method": "BCD-DR"})
results_dict.update({"data_name": data_name})
results_dict.update({"beta": beta})
#results_dict.update({"search_radius_const": search_radius_const})
results_dict.update({"n_components": n_components})
results_dict.update({"iterations": iter})
results_dict.update({"num_trials": num_repeat})
results_dict.update({'timed_errors_trials': timed_errors_trials})
full_result_list.append(results_dict.copy())
save_path = save_folder + "/full_result_list_" + str(data_name)
directory = os.path.dirname(save_path)
if not os.path.exists(directory):
os.makedirs(directory)
np.save(save_path, full_result_list)
# print('full_result_list:', full_result_list)
if run_MU:
results_dict = {}
print('!!! MU initialized')
list_full_timed_errors = []
iter1 = iter
for i in np.arange(num_repeat):
result_dict_MU = MU_run(X,
n_components=n_components,
iter=iter1,
regularizer=0,
ini_loading=loading_list[i],
if_compute_recons_error=True,
save_folder=save_folder,
output_results=True)
time_error = result_dict_MU.get('time_error')
list_full_timed_errors.append(time_error.copy())
# print('!!! list_full_timed_errors', len(list_full_timed_errors))
timed_errors_trials = np.asarray(
list_full_timed_errors) # shape (# trials) x (2 for time, error) x (iterations)
results_dict.update({"method": "MU"})
results_dict.update({"data_name": data_name})
results_dict.update({"n_components": n_components})
#results_dict.update({"search_radius_const": search_radius_const})
results_dict.update({"iterations": iter1})
results_dict.update({"num_trials": num_repeat})
results_dict.update({'timed_errors_trials': timed_errors_trials})
full_result_list.append(results_dict.copy())
save_path = save_folder + "/full_result_list_" + str(data_name)
directory = os.path.dirname(save_path)
if not os.path.exists(directory):
os.makedirs(directory)
np.save(save_path, full_result_list)
# print('full_result_list:', full_result_list)
if run_OCPDL:
# search_radius_const_list = [1, 10, 100, 1000]
#np.random.seed(2)
for c in search_radius_const_list:
print('search_radius_const', c)
beta_list = [3/4, 1, None]
#beta_list = [3/4]
for beta in beta_list:
print('!!! OCPDL initialized with beta:', beta)
list_full_timed_errors = []
results_dict = {}
for i in np.arange(num_repeat):
result_dict_OCPDL = OCPDL_run(X,
n_components=n_components,
iter=iter,
regularizer=0,
batch_size = int(X.shape[-1]/5),
#batch_size = 10,
# inverse regularizer on time mode (to promote long-lasting topics),
# no regularizer on on words and tweets
ini_loading=loading_list[i].copy(),
beta=beta,
search_radius_const=c,
if_compute_recons_error=True,
#mode_2be_subsampled=1,
save_folder=save_folder,
output_results=True)
time_error = result_dict_OCPDL.get('time_error')
print('!!! time_error', time_error)
list_full_timed_errors.append(time_error.copy())
# print('!!! list_full_timed_errors', len(list_full_timed_errors))
timed_errors_trials = np.asarray(
list_full_timed_errors) # shape (# trials) x (2 for time, error) x (iterations)
results_dict.update({"method": "OCPDL"})
results_dict.update({"data_name": data_name})
results_dict.update({"beta": beta})
results_dict.update({"search_radius_const": c})
results_dict.update({"n_components": n_components})
results_dict.update({"iterations": iter})
results_dict.update({"num_trials": num_repeat})
results_dict.update({'timed_errors_trials': timed_errors_trials})
full_result_list.append(results_dict.copy())
save_path = save_folder + "/full_result_list_" + str(data_name)
directory = os.path.dirname(save_path)
if not os.path.exists(directory):
os.makedirs(directory)
np.save(save_path, full_result_list)
# print('full_result_list:', full_result_list)
if plot_errors:
full_result_list = np.load(save_folder + "/full_result_list_" + str(data_name) + ".npy", allow_pickle=True)
full_result_list = full_result_list[::-1]
#n_trials = full_result_list[0].get("num_trials")
#n_components = full_result_list[0].get('n_components')
#search_radius_const = full_result_list[0].get('search_radius_const')
#data_name = full_result_list[0].get('data_name')
#save_path = save_folder + "/full_result_error_plot" + '_ntrials_' + str(n_trials) + "_" + "_ncomps_" + str(
# n_components) + "_src_" + str(search_radius_const) + "_" + str(data_name) + ".pdf"
plot_benchmark_errors(full_result_list, save_folder=save_folder)
print('!!! plot saved')
def generate_figures():
# synthetic
main(n_components = 5,
iter = 10,
num_repeat = 5,
save_folder = "Output_files/JMLR_OCPDL1",
data_name = "Synthetic", # "Synthetic" or "Twitter"
run_ALS = True,
run_MU = True,
run_OCPDL = True,
plot_errors = True,
search_radius_const_list=[1,10,100,1000])
# Headlines
main(n_components = 5,
iter = 30,
num_repeat = 10,
save_folder = "Output_files/JMLR_OCPDL1",
data_name = "Headlines", # "Synthetic" or "Twitter"
run_ALS = True,
run_MU = True,
run_OCPDL = True,
plot_errors = True,
search_radius_const_list=[10])
"""
# 20Newsgroups
# Get 20news_tfidf_tensor.pickle from https://github.com/lara-kassab/dynamic-tensor-topic-modeling/tree/main/data
# and place it in Data folder before running this code.
main(n_components = 5,
iter = 40,
num_repeat = 2,
save_folder = "Output_files/JMLR_OCPDL1",
data_name = "20Newsgroups", # "Synthetic" or "Twitter"
run_ALS = True,
run_MU = True,
run_OCPDL = True,
plot_errors = True,
search_radius_const_list=[1,10,100,1000])
"""
"""
# Twitter (This requires to get data_tensor_top1000.pickle (~3GB).)
# Refer to https://github.com/lara-kassab/dynamic-tensor-topic-modeling
# Also available upon request ([email protected])
main(n_components = 5,
iter = 30,
num_repeat = 10,
save_folder = "Output_files/JMLR_OCPDL1",
data_name = "Twitter", # "Synthetic" or "Twitter"
run_ALS = True,
run_MU = True,
run_OCPDL = True,
plot_errors = True,
search_radius_const_list=[10])
"""
if __name__ == '__main__':
# main()
generate_figures()