-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathrun_nbvae.py
335 lines (241 loc) · 11.5 KB
/
run_nbvae.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
import os
import numpy as np
import tensorflow as tf
from scipy import sparse
import sys
import getopt
import time
from load_cf_data import load_tr_te_data, load_train_data
from load_text_data import load_data
from evaluate import evaluate_all
from nb_vae import NegativeBinomialVAE
from nb_vae_b import NegativeBinomialVAEb
def main():
collection_step = 10
total_anneal_steps = 20000
data_dir = './datasets'
batch_size = 500
batch_size_vad = 2000
n_epochs = 800
try:
opts, args = getopt.getopt(sys.argv[1:], 'hd:b:m:s:t:a:n:',
['dataset=', 'is-data-binary=' 'model-id=', 'save-dir=',
'is-training=', 'arch-id=', 'num-training-epochs'])
except getopt.GetoptError as err:
print(err)
sys.exit(2)
for o, a in opts:
if o in ('-d', '--dataset'):
dataset = str(a)
elif o in ('-b', '--is-data-binary'):
is_data_binary = int(a)
elif o in ('-m', '--model-id'):
model_id = int(a)
elif o in ('-s', '--save-dir'):
base_dir = a
elif o in ('-t', '--training'):
is_training = int(a)
elif o in ('-a', '--arch-id'):
arch_id = int(a)
elif o in ('-n', '--num-training-epochs'):
n_epochs = int(a)
elif o == '-h':
print('-d or --dataset: dataset name (has to be the same to the folder that saves the dataset), e.g., '
'20NG, ML-10M ...')
print('-m or --model-id: 1. NBVAE, 2. NBVAE_dm, 3. NBVAE_b')
print('-s or --save-dir: the folder that saves the model and the log files')
print('-t or --is-training: 1. training phrase, 0. testing phrase')
print('-a or --arch-id: network architecture, 1. [128], 2. [64-128], 3. [32-64-128], 4. [128-256], '
'5. [256-512], 6. [200-600]')
print('-n or --num-training-epochs')
exit()
else:
assert False, "unhandled option %s %s" % (o, a)
data_dir = '%s/%s' % (data_dir, dataset)
if is_data_binary == 1: # cf data
pro_dir = os.path.join(data_dir, 'pro_sg')
unique_sid = list()
with open(os.path.join(pro_dir, 'unique_sid.txt'), 'r') as f:
for line in f:
unique_sid.append(line.strip())
V = len(unique_sid)
train_data = load_train_data(os.path.join(pro_dir, 'train.csv'), V)
vad_data_tr, vad_data_te = load_tr_te_data(os.path.join(pro_dir, 'validation_tr.csv'),
os.path.join(pro_dir, 'validation_te.csv'), V)
else: # text
train_data, vad_data_tr, vad_data_te = load_data('%s/data.mat' % data_dir)
vad_data_te_mask = vad_data_te.todense() > 0
V = train_data.shape[1]
if arch_id == 1:
arch = [128, V]
elif arch_id == 2:
arch = [64, 128, V]
elif arch_id == 3:
arch = [32, 64, 128, V]
elif arch_id == 4:
arch = [128, 256, V]
elif arch_id == 5:
arch = [256, 512, V]
else:
arch = [200, 600, V]
N = train_data.shape[0]
N_vad = vad_data_tr.shape[0]
idxlist_vad = range(N_vad)
if is_data_binary == 1:
anneal_cap = 0.2
else:
anneal_cap = 1.0
tf.reset_default_graph()
if model_id == 1:
vae = NegativeBinomialVAE(arch, lr=1e-3, random_seed=1)
else:
vae = NegativeBinomialVAEb(arch, lr=1e-3, random_seed=1)
saver, train_op, h_r, h_p = vae.build_graph()
arch_str = '%s' % ('-'.join([str(d) for d in arch[:-1]]))
if model_id == 1:
model_str = 'nbvae'
else:
model_str = 'nbvae_b'
model_save_dir = '{}/saved_models/{}/{}/{}'.format(base_dir, dataset, model_str, arch_str)
log_dir = '{}/logs/{}/{}/{}/'.format(base_dir, dataset, model_str, arch_str)
if is_training:
if not os.path.exists(log_dir):
os.makedirs(log_dir)
if not os.path.exists(model_save_dir):
os.makedirs(model_save_dir)
nms = []
nss = []
rms = []
rss = []
pps_test = []
with tf.Session() as sess:
init = tf.global_variables_initializer()
sess.run(init)
best_ndcg_vad = -np.inf
update_count = 0.0
for epoch in range(n_epochs):
idxlist = np.random.permutation(N)
start_time = time.time()
for bnum, st_idx in enumerate(range(0, N, batch_size)):
end_idx = min(st_idx + batch_size, N)
X = train_data[idxlist[st_idx:end_idx]]
if sparse.isspmatrix(X):
X = X.toarray()
X = X.astype('float32')
if anneal_cap > 0:
anneal = min(anneal_cap, 1. * update_count / total_anneal_steps)
else:
anneal = 1.0
feed_dict = {vae.input_ph: X,
vae.keep_prob_ph: 0.5,
vae.anneal_ph: anneal,
vae.is_training_ph: 1}
sess.run(train_op, feed_dict=feed_dict)
sys.stdout.write('\r')
sys.stdout.write('epoch: %d, %d' % (epoch, bnum))
sys.stdout.flush()
update_count += 1
end_time = time.time()
print('\n epoch: %d in %s secs' % (epoch, end_time - start_time))
if epoch == n_epochs - 1 and is_data_binary == 0:
saver.save(sess, '{}/model-last'.format(model_save_dir))
if epoch % collection_step == 0:
if is_data_binary == 1:
ndcgs = None
recalls = None
for bnum, st_idx in enumerate(range(0, N_vad, batch_size_vad)):
end_idx = min(st_idx + batch_size_vad, N_vad)
X_vad_seen = vad_data_tr[idxlist_vad[st_idx:end_idx]]
X_vad_unseen = vad_data_te[idxlist_vad[st_idx:end_idx]]
if sparse.isspmatrix(X_vad_seen):
X_vad_seen = X_vad_seen.todense()
X_vad_seen = X_vad_seen.astype('float32')
r, p = sess.run((h_r, h_p), feed_dict={vae.input_ph: X_vad_seen})
pred_l = vae.get_predictive_rate(r, p, X_vad_seen)
pred_l[X_vad_seen > 0] = -np.inf
ns, rs = evaluate_all(pred_l, X_vad_unseen)
if ndcgs is not None:
ndcgs = np.append(ndcgs, ns, axis=1)
recalls = np.append(recalls, rs, axis=1)
else:
ndcgs = ns
recalls = rs
nm = np.mean(ndcgs, axis=1)
ns = np.std(ndcgs, axis=1) / ndcgs.shape[1]
rm = np.mean(recalls, axis=1)
rs = np.std(recalls, axis=1) / ndcgs.shape[1]
nms.append(nm)
nss.append(ns)
rms.append(rm)
rss.append(rs)
np.save('%s/nms' % (log_dir), nms)
np.save('%s/nss' % (log_dir), nss)
np.save('%s/rms' % (log_dir), rms)
np.save('%s/rss' % (log_dir), rss)
print('vad-ndcg@50: %f' % nm[-1])
if nm[-1] > best_ndcg_vad:
best_ndcg_vad = nm[-1]
saver.save(sess, '{}/model'.format(model_save_dir))
else:
log_prob_test = 0
word_count_test = 0
for bnum, st_idx in enumerate(range(0, N_vad, batch_size_vad)):
end_idx = min(st_idx + batch_size_vad, N_vad)
X_vad_seen = vad_data_tr[idxlist_vad[st_idx:end_idx]]
X_vad_unseen = vad_data_te[idxlist_vad[st_idx:end_idx]]
X_vad_unseen_mask = vad_data_te_mask[idxlist_vad[st_idx:end_idx]]
if sparse.isspmatrix(X_vad_seen):
X_vad_seen = X_vad_seen.todense()
X_vad_seen = X_vad_seen.astype('float32')
r, p = sess.run((h_r, h_p), feed_dict={vae.input_ph: X_vad_seen})
pred_l = vae.get_predictive_rate(r, p, X_vad_seen)
pred_l = pred_l / np.sum(pred_l, axis=1).reshape(pred_l.shape[0], 1)
log_prob = np.multiply(X_vad_unseen[X_vad_unseen_mask],
np.log(pred_l[X_vad_unseen_mask]))
log_prob_test += np.sum(log_prob)
word_count_test += np.sum(X_vad_unseen)
pp_test = log_prob_test / word_count_test
pp_test = np.exp(-pp_test)
pps_test.append(pp_test)
print('test-pp: %f' % pp_test)
np.save('%s/pps-test' % (log_dir), pps_test)
else:
if is_data_binary == 1:
test_data_tr, test_data_te = load_tr_te_data(
os.path.join(pro_dir, 'test_tr.csv'),
os.path.join(pro_dir, 'test_te.csv'), V)
print("saved model directory: %s" % model_save_dir)
with tf.Session() as sess:
saver.restore(sess, '{}/model'.format(model_save_dir))
ndcgs = None
recalls = None
N_test = test_data_tr.shape[0]
for bnum, st_idx in enumerate(range(0, N_test, batch_size_vad)):
end_idx = min(st_idx + batch_size_vad, N_vad)
X_test_seen = test_data_tr[idxlist_vad[st_idx:end_idx]]
X_test_unseen = test_data_te[idxlist_vad[st_idx:end_idx]]
if sparse.isspmatrix(X_test_seen):
X_test_seen = X_test_seen.todense()
X_test_seen = X_test_seen.astype('float32')
r, p = sess.run((h_r, h_p), feed_dict={vae.input_ph: X_test_seen})
pred_l = vae.get_predictive_rate(r, p, X_test_seen)
pred_l[X_test_seen > 0] = -np.inf
ns, rs = evaluate_all(pred_l, X_test_unseen)
if ndcgs is not None:
ndcgs = np.append(ndcgs, ns, axis=1)
recalls = np.append(recalls, rs, axis=1)
else:
ndcgs = ns
recalls = rs
nm = np.mean(ndcgs, axis=1)
ns = np.std(ndcgs, axis=1) / ndcgs.shape[1]
rm = np.mean(recalls, axis=1)
rs = np.std(recalls, axis=1) / ndcgs.shape[1]
np.save('%s/nm-test' % (log_dir), nm)
np.save('%s/ns-test' % (log_dir), ns)
np.save('%s/rm-test' % (log_dir), rm)
np.save('%s/rs-test' % (log_dir), rs)
print('test NDCG@1: %f, @5: %f, @10: %f, @20: %f, @50: %f\n' % (nm[0], nm[1], nm[2], nm[3], nm[4]))
print('test Recall@1: %f, @5: %f, @10: %f, @20: %f, @50: %f\n' % (rm[0], rm[1], rm[2], rm[3], rm[4]))
if __name__ == "__main__":
main()