-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathwide_deep_emb_conv.py
765 lines (655 loc) · 46.8 KB
/
wide_deep_emb_conv.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
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import shutil
import os, sys, json, itertools
import pickle
import numpy as np
import tensorflow as tf
from sparse_array_categorical_column import categorical_column_with_array_input
#os.environ["TF_CPP_MIN_LOG_LEVEL"]='1' # 这是默认的显示等级，显示所有信息
#os.environ["TF_CPP_MIN_LOG_LEVEL"]='2' # 只显示 warning 和 Error
os.environ["TF_CPP_MIN_LOG_LEVEL"]='3' # 只显示 Error
#ages,age,gender,platform,phone,location,network,bidtype,psid,style,link,show,position,zerocate,fircate,seccat,hierarchy_smooth_ctr,history_ctr,gender_feed_smooth_ctr,gender_cust_smooth_ctr,platform_feed_smooth_ctr,platform_cust_smooth_ctr,cust60_smooth_ctr,custid,adid,feedid
_CSV_COLUMNS = ['ages','age','gender','platform','phone','location','network','bidtype','psid','style','link','show','position','zerocate','fircate','seccate','hierarchy_smooth_ctr','history_ctr','gender_feed_smooth_ctr','gender_cust_smooth_ctr','platform_feed_smooth_ctr','platform_cust_smooth_ctr','cust60_smooth_ctr','custid','adid','feedid','user_class','cust_tag','feed_word','user_bhvtag','user_bhvword','ig']
_CSV_COLUMN_DEFAULTS = [[''], [1008], [''],[''],[''],[''], [''], [''], [''], [''], [''], [''], [''], [''], [''], [''], [0], [0], [0], [0], [0], [0], [0], ['0'], ['0'], ['0'], ['0'],['0'], ['0'], ['0'],['0'], [0]]
# for 5 vector
#_CSV_COLUMNS = ['user_class', 'cust_tag', 'feed_word', 'user_bhvword', 'user_bhvtag', 'label']
#_CSV_COLUMN_DEFAULTS = [['NULL'], ['NULL'], ['NULL'], ['NULL'], ['NULL'], [0]]
_GLOBAL_FEATURES = []
def define_flags():
flags = tf.app.flags
tf.app.flags.DEFINE_string("task" , "train", "train/dump/pred")
# Flags Sina ML required
tf.app.flags.DEFINE_string("data_dir" , "", "Set local data path of train set. Coorpate with 'input-strategy DOWNLOAD'.")
tf.app.flags.DEFINE_string("validate_dir", "", "Set local data path of validate set. Coorpate with 'input-strategy DOWNLOAD'.")
tf.app.flags.DEFINE_string("train_dir" , "", "Set model save path. Not input path")
tf.app.flags.DEFINE_string("log_dir" , "", "Set tensorboard even log path.")
# Flags Sina ML required: for tf.train.ClusterSpec
tf.app.flags.DEFINE_string("ps_hosts" , "", "Comma-separated list of hostname:port pairs, you can also specify pattern like ps[1-5].example.com")
tf.app.flags.DEFINE_string("worker_hosts", "", "Comma-separated list of hostname:port pairs, you can also specify worker[1-5].example.co")
# Flags Sina ML required:Flags for defining the tf.train.Server
tf.app.flags.DEFINE_string("job_name" , "", "One of 'ps', 'worker'")
tf.app.flags.DEFINE_integer("task_index" , 0 , "Index of task within the job.Sina ML required arg.")
flags.DEFINE_string("checkpoints_dir" , "./checkpoints_dir" , "Set checkpoints path.")
#flags.DEFINE_string("model_dir" , "./model_dir" , "Set checkpoints path.")
flags.DEFINE_string("output_model" , "./model_output", "Path to the training data.")
flags.DEFINE_string("hidden_units" , "512,256,128" , "Comma-separated list of number of units in each hidden layer of the NN")
flags.DEFINE_integer("num_epochs" , 1000000 , "Number of (global) training steps to perform, default 1000000")
flags.DEFINE_integer("batch_size" , 10000 , "Training batch size, default 512")
flags.DEFINE_integer("shuffle_buffer_size" , 10000 , "dataset shuffle buffer size, default 10000")
flags.DEFINE_float("learning_rate" , 0.0001 , "Learning rate, default 0.01")
flags.DEFINE_float("dropout_rate" , 0.10 , "Drop out rate, default 0.25")
flags.DEFINE_integer("num_parallel_readers" , 5 , "number of parallel readers for training data, default 5")
flags.DEFINE_integer("save_checkpoints_steps", 100 , "Save checkpoints every this many steps, default 5000")
flags.DEFINE_boolean("run_on_cluster" , False , "Whether the cluster info need to be passed in as input, default False")
flags.DEFINE_string("input_strategy" , "local" , "Coorpate with \"input-strategy\", input data strategy")
flags.DEFINE_string("model_type" , "wide" , "wide/deep/conv")
flags.DEFINE_string("pretrain" , "no" , "yes/no")
flags.DEFINE_string("embedding_model" , "" , "./model_zoo/wide_deep_emb_conv_dropout/model.ckpt-47521")
flags.DEFINE_integer("embedding_dim" , 100 , "100/200")
flags.DEFINE_boolean("fineturn" , True , "fine-turn the embedding")
flags.DEFINE_boolean("only_usm_input" , False , "")
FLAGS = flags.FLAGS
return FLAGS
FLAGS = define_flags()
def parse_argument():
if FLAGS.job_name is None or FLAGS.job_name == "":
raise ValueError("Must specify an explicit `job_name`")
if FLAGS.task_index is None or FLAGS.task_index == "":
raise ValueError("Must specify an explicit `task_index`")
print("job name = %s" % FLAGS.job_name)
print("task index = %d" % FLAGS.task_index)
os.environ["TF_ROLE"] = FLAGS.job_name
os.environ["TF_INDEX"] = str(FLAGS.task_index)
# Construct the cluster and start the server
ps_spec = FLAGS.ps_hosts.split(",")
worker_spec = FLAGS.worker_hosts.split(",")
cluster = {"worker": worker_spec, "ps": ps_spec}
os.environ["TF_CLUSTER_DEF"] = json.dumps(cluster)
def set_tfconfig_environ():
if "TF_CLUSTER_DEF" in os.environ:
cluster = json.loads(os.environ["TF_CLUSTER_DEF"])
task_index = int(os.environ["TF_INDEX"])
task_type = os.environ["TF_ROLE"]
tf_config = dict()
worker_num = len(cluster["worker"])
if FLAGS.job_name == "ps":
tf_config["task"] = {"index": task_index, "type": task_type}
FLAGS.job_name = "ps"
FLAGS.task_index = task_index
else:
if task_index == 0:
tf_config["task"] = {"index": 0, "type": "chief"}
else:
tf_config["task"] = {"index": task_index - 1, "type": task_type}
FLAGS.job_name = "worker"
FLAGS.task_index = task_index
if worker_num == 1:
cluster["chief"] = cluster["worker"]
del cluster["worker"]
else:
cluster["chief"] = [cluster["worker"][0]]
del cluster["worker"][0]
tf_config["cluster"] = cluster
os.environ["TF_CONFIG"] = json.dumps(tf_config)
print("TF_CONFIG", json.loads(os.environ["TF_CONFIG"]))
def build_model_columns():
"""Builds a set of wide and deep feature columns."""
# Continuous columns
age = tf.feature_column.numeric_column('age')
ages = tf.feature_column.categorical_column_with_vocabulary_list('ages', [
'1101', '1102', '1103', '1104', '1105', '1106'])
gender = tf.feature_column.categorical_column_with_vocabulary_list('gender', ['400','401','402'])
platform = tf.feature_column.categorical_column_with_vocabulary_list('platform', ['', '90110100','90110200','90110201','90110202'])
phone = tf.feature_column.categorical_column_with_vocabulary_list('phone', ['','90101000','90102000','90103000','90104000','90105000','90106000','90107000','90108000','90109000','90110000','90199000'])
location = tf.feature_column.categorical_column_with_vocabulary_list('location', ['','30000','30101','30102','30103','30104','30105','30106','30107','30108','30109','30110','30111','30112','30113','30114','30115','30116','30117','30118','30119','30120','30121','30201','30301','30401','30402','30403','30404','30405','30406','30407','30408','30409','30410','30411','30501','30502','30503','30504','30505','30506','30507','30508','30509','30510','30511','30512','30513','30601','30602','30603','30604','30605','30606','30607','30608','30609','30610','30611','30612','30613','30614','30615','30616','30617','30701','30702','30703','30704','30705','30706','30707','30708','30709','30801','30802','30803','30804','30805','30806','30807','30808','30809','30810','30811','30812','30813','30814','30815','30816','30817','30818','30819','30820','30821','30900','30901','30902','30903','30904','30905','30906','30907','30908','30909','30910','30911','30912','30913','30914','31000','31001','31002','31003','31004','31005','31006','31007','31008','31009','31010','31011','31012','31013','31014','31015','31016','31017','31101','31102','31103','31104','31105','31106','31107','31108','31109','31110','31111','31112','31113','31114','31201','31202','31203','31204','31205','31206','31207','31208','31209','31210','31211','31301','31302','31303','31304','31305','31306','31307','31308','31309','31310','31311','31312','31313','31314','31401','31402','31403','31404','31405','31406','31407','31408','31409','31410','31501','31502','31503','31504','31505','31506','31507','31508','31509','31510','31511','31512','31513','31514','31601','31701','31702','31703','31704','31705','31706','31707','31708','31709','31710','31711','31712','31713','31714','31715','31716','31717','31801','31901','31902','31903','31904','31905','31906','31907','31908','31909','31910','31911','31912','31913','32001','32002','32003','32004','32005','32006','32007','32008','32009','32010','32011','32100','32101','32102','32103','32104','32105','32106','32107','32108','32109','32200','32201','32202','32203','32204','32205','32206','32207','32208','32209','32210','32211','32212','32213','32214','32215','32216','32301','32302','32303','32304','32305','32306','32307','32308','32309','32310','32311','32401','32402','32403','32404','32405','32406','32407','32408','32409','32501','32600','32601','32602','32603','32604','32605','32606','32607','32608','32609','32610','32611','32612','32701','32800','32801','32802','32900','32901','32902','32903','32904','32905','32906','32907','32908','32909','32910','32911','32912','32913','32914','32915','32916','32917','33001','33002','33003','33004','33005','33006','33007','33008','33009','33010','33011','33012','33013','33014','33101','33102','33103','33104','33105','33201','33300','33301','33302','33303','33304','33305','33306','33307','33308','33401','33402','33403','33404','33405','33406','33407','33501','33601','33701','33801','33901','34001','34101','34201'])
network = tf.feature_column.categorical_column_with_vocabulary_list('network', ['','60110100','60110200','60110302','60110303','60110304'])
bidtype = tf.feature_column.categorical_column_with_vocabulary_list('bidtype', ['','60211','60214','60218','602112'])
psid = tf.feature_column.categorical_column_with_vocabulary_list('psid', ['','70101011','70101012','70101031','70102023','70104011','70106011','70110011'])
style = tf.feature_column.categorical_column_with_vocabulary_list('style', ['','702100','702101','702102','702104','702105','702106','702110','702112','702113','702114','702115','702119','702120','702122','702124','702130','702140','702150','702152','702170','702197','702198'])
link = tf.feature_column.categorical_column_with_vocabulary_list('link', ['','80210','80211','80212','80213','80214','80215','80216','80217','80218'])
show = tf.feature_column.categorical_column_with_vocabulary_list('show', ['','90210','90211'])
position = tf.feature_column.categorical_column_with_vocabulary_list('position', ['','7100212','7100213','7100214','7100215','7100216','7100217','7100218','7100219','71002110','71002111','71002112','71002113','71002114','71002115','71002116','71002117','71002118','71002119','71002120','71002121','71002122','71002123','71002124','71002125','71002126','71002127','71002128','71002129','71002130','71002131','71002132','71002133','71002134','71002135','71002136','71002137','71002138','71002139','71002140','71002141','71002142','71002143','71002144','71002145','71002146','71002147','71002148','71002149','71002150','71002151','71002152','71002153','71002155','71002157','71002158','71002159','71002163','71002166','71002167','71002168','71002173','71002175','71002181','71002183','71002191','71002193'])
zerocate = tf.feature_column.categorical_column_with_vocabulary_list('zerocate', ['','null','20000000','29000000','31000000','32000000','33000000','34000000','35000000'])
fircate = tf.feature_column.categorical_column_with_vocabulary_list('fircate', ['','0','200100','200200','200300','200500','200600','200700','200800','200900','201000','201400','201600','201700','201800','202200','299900','31001000','31002000','32001000','33001000','33002000','33003000','33004000','33005000','33006000','33007000','33008000','33009000','34001000','34002000','34003000','34004000','34005000','34006000','34007000','34008000','34010000','34011000','34012000','35001000','35003000','35004000','35005000','35006000'])
seccate = tf.feature_column.categorical_column_with_vocabulary_list('seccate', ['','0','200101','200103','200199','200201','200202','200203','200204','200205','200206','200207','200208','200209','200210','200299','200399','200599','200601','200602','200604','200605','200607','200608','200609','200701','200802','200803','200804','200805','200899','200906','200907','201006','201401','201403','201404','201405','201499','201601','201604','201605','201699','201701','201702','201705','201799','201808','201899','202299','299999','31001001','31001002','31001003','31001004','31001006','31001007','31001008','31001010','31001011','31001012','31001014','31001015','31001016','31001017','31001018','31001019','31001020','31001021','31001023','31001024','31001026','31002001','31002002','31002004','31002008','31002009','31002010','31002011','31002014','31002015','32001001','32001002','32001003','32001004','32001005','32001006','32001009','32001010','32001011','32001014','32001015','33001001','33001002','33001003','33001004','33001005','33001006','33001009','33001013','33001014','33001015','33002001','33002002','33002003','33002004','33002005','33002006','33002008','33002009','33002010','33002011','33002012','33002013','33003001','33003002','33003003','33003004','33003005','33004001','33004002','33004003','33004004','33004006','33004007','33004008','33004009','33005001','33005002','33005003','33005004','33005005','33006002','33006003','33006004','33006005','33006007','33006008','33006009','33007001','33007005','33007007','33008001','33008002','33008003','33008004','33009001','33009002','34001001','34001003','34001004','34002001','34002002','34002003','34002004','34002005','34002006','34002007','34002008','34002009','34002010','34002012','34003001','34004001','34004002','34004003','34004004','34004005','34004006','34005001','34005002','34005003','34006001','34006002','34006003','34006004','34006005','34007001','34008001','34008003','34008004','34010001','34010004','34010005','34010006','34010007','34010008','34010009','34010010','34010012','34010014','34010015','34010016','34011001','34012001','34012002','34012003','34012004','34012006','34012007','34012008','35001001','35003001','35003005','35003007','35004001','35005004','35005005','35005006','35005007','35005008','35005009','35005010','35006001','35006002','35006003','35006004','35006005','35006006','35006007','35006008','35006009','35006010','35006011','35006012','35006013'])
hierarchy_smooth_ctr = tf.feature_column.numeric_column('hierarchy_smooth_ctr')
history_ctr = tf.feature_column.numeric_column('history_ctr')
gender_feed_smooth_ctr = tf.feature_column.numeric_column('gender_feed_smooth_ctr')
gender_cust_smooth_ctr = tf.feature_column.numeric_column('gender_cust_smooth_ctr')
platform_feed_smooth_ctr = tf.feature_column.numeric_column('platform_feed_smooth_ctr')
platform_cust_smooth_ctr = tf.feature_column.numeric_column('platform_cust_smooth_ctr')
cust60_smooth_ctr = tf.feature_column.numeric_column('cust60_smooth_ctr')
custid = tf.feature_column.categorical_column_with_hash_bucket('custid', hash_bucket_size=100000)
adid = tf.feature_column.categorical_column_with_hash_bucket('adid', hash_bucket_size=1000000)
feedid = tf.feature_column.categorical_column_with_hash_bucket('feedid', hash_bucket_size=1000000)
user_class = categorical_column_with_array_input('user_class', 20000, '|')
cust_tag = categorical_column_with_array_input('cust_tag', 20000, '|')
feed_word = categorical_column_with_array_input('feed_word', 20000, '|')
user_bhvword = categorical_column_with_array_input('user_bhvword', 20000, '|')
user_bhvtag = categorical_column_with_array_input('user_bhvtag', 20000, '|')
if FLAGS.embedding_model != '' and FLAGS.model_type.find('conv') >= 0:
trainable_flag = FLAGS.fineturn
user_class_emb = tf.feature_column.embedding_column(user_class, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/user_class_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=trainable_flag)
cust_tag_emb = tf.feature_column.embedding_column(cust_tag, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer_1/cust_tag_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=trainable_flag)
feed_word_emb = tf.feature_column.embedding_column(feed_word, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer_2/feed_word_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=trainable_flag)
user_bhvword_emb = tf.feature_column.embedding_column(user_bhvword, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer_3/user_bhvword_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=trainable_flag)
user_bhvtag_emb = tf.feature_column.embedding_column(user_bhvtag, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer_4/user_bhvtag_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=trainable_flag)
else:
he_init = tf.initializers.truncated_normal #tf.keras.initializers.he_normal
user_class_emb = tf.feature_column.embedding_column(user_class, dimension=FLAGS.embedding_dim, initializer=he_init)
cust_tag_emb = tf.feature_column.embedding_column(cust_tag, dimension=FLAGS.embedding_dim, initializer=he_init)
feed_word_emb = tf.feature_column.embedding_column(feed_word, dimension=FLAGS.embedding_dim, initializer=he_init)
user_bhvword_emb = tf.feature_column.embedding_column(user_bhvword, dimension=FLAGS.embedding_dim, initializer=he_init)
user_bhvtag_emb = tf.feature_column.embedding_column(user_bhvtag, dimension=FLAGS.embedding_dim, initializer=he_init)
he_init = tf.initializers.truncated_normal #tf.keras.initializers.he_normal
user_class_emb2 = tf.feature_column.embedding_column(user_class, dimension=FLAGS.embedding_dim, initializer=he_init)
cust_tag_emb2 = tf.feature_column.embedding_column(cust_tag, dimension=FLAGS.embedding_dim, initializer=he_init)
feed_word_emb2 = tf.feature_column.embedding_column(feed_word, dimension=FLAGS.embedding_dim, initializer=he_init)
user_bhvword_emb2 = tf.feature_column.embedding_column(user_bhvword, dimension=FLAGS.embedding_dim, initializer=he_init)
user_bhvtag_emb2 = tf.feature_column.embedding_column(user_bhvtag, dimension=FLAGS.embedding_dim, initializer=he_init)
'''
# Transformations.
age_buckets = tf.feature_column.bucketized_column(
age, boundaries=[18, 25, 30, 35, 40, 45, 50, 55, 60, 65])
'''
# Wide columns and deep columns.
base_columns = [
age, ages, gender, platform, phone, location, network, bidtype, psid, style, link, show, position, zerocate, fircate, seccate, custid, adid, feedid,
]
'''
crossed_columns = [
tf.feature_column.crossed_column(
['network', 'psid'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['style', 'zerocate'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['style', 'fircate'], hash_bucket_size=10000),
tf.feature_column.crossed_column(
['style', 'ages'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['style', 'gender'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['style', 'psid'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['ages', 'platform', 'psid'], hash_bucket_size=1000),
]
'''
wide_columns = base_columns # + crossed_columns
# age, ages, gender, platform, phone, location, network, bidtype, psid, style, link, show, position, zerocate, fircate, seccate, custid, adid, feedid,
'''deep_columns = [
age,
hierarchy_smooth_ctr,
history_ctr,
gender_feed_smooth_ctr,
gender_cust_smooth_ctr,
platform_feed_smooth_ctr,
platform_cust_smooth_ctr,
cust60_smooth_ctr,
tf.feature_column.indicator_column(gender),
tf.feature_column.indicator_column(phone),
tf.feature_column.indicator_column(bidtype),
tf.feature_column.indicator_column(position),
tf.feature_column.indicator_column(style),
tf.feature_column.indicator_column(link),
tf.feature_column.indicator_column(show),
tf.feature_column.indicator_column(ages),
# To show an example of embedding
tf.feature_column.embedding_column(seccate, dimension=8),
user_class_emb,
cust_tag_emb,
feed_word_emb,
user_bhvword_emb,
user_bhvtag_emb
]
'''
deep_columns = [user_class_emb,
cust_tag_emb,
feed_word_emb,
user_bhvword_emb,
user_bhvtag_emb]
usm_deep_columns = [
age,
hierarchy_smooth_ctr,
history_ctr,
gender_feed_smooth_ctr,
gender_cust_smooth_ctr,
platform_feed_smooth_ctr,
platform_cust_smooth_ctr,
cust60_smooth_ctr,
tf.feature_column.indicator_column(gender),
tf.feature_column.indicator_column(phone),
tf.feature_column.indicator_column(bidtype),
tf.feature_column.indicator_column(position),
tf.feature_column.indicator_column(style),
tf.feature_column.indicator_column(link),
tf.feature_column.indicator_column(show),
tf.feature_column.indicator_column(ages),
# To show an example of embedding
tf.feature_column.embedding_column(seccate, dimension=8)
]
embedding_columns = [user_class_emb, cust_tag_emb, feed_word_emb, user_bhvword_emb, user_bhvtag_emb]
#embedding_columns = [[]]
return wide_columns, deep_columns, embedding_columns, usm_deep_columns
'''
def build_model_columns_sparse():
#'user_class', 'cust_tag', 'feed_word', 'user_bhvword', 'user_bhvtag
user_class = categorical_column_with_array_input('user_class', 50000, '|')
cust_tag = categorical_column_with_array_input('cust_tag', 50000, '|')
feed_word = categorical_column_with_array_input('feed_word', 50000, '|')
user_bhvword = categorical_column_with_array_input('user_bhvword', 50000, '|')
user_bhvtag = categorical_column_with_array_input('user_bhvtag', 50000, '|')
if FLAGS.embedding_model != '': # ./model_zoo/wide_deep_emb_conv_dropout/model.ckpt-47521
user_class_emb = tf.feature_column.embedding_column(user_class, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/user_class_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=False)
cust_tag_emb = tf.feature_column.embedding_column(cust_tag, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/cust_tag_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=False)
feed_word_emb = tf.feature_column.embedding_column(feed_word, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/feed_word_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=False)
user_bhvword_emb = tf.feature_column.embedding_column(user_bhvword, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/user_bhvword_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=False)
user_bhvtag_emb = tf.feature_column.embedding_column(user_bhvtag, dimension=FLAGS.embedding_dim, tensor_name_in_ckpt='emb_conv_model/input_layer/user_bhvtag_embedding/embedding_weights', ckpt_to_load_from=FLAGS.embedding_model, trainable=False)
else:
user_class_emb = tf.feature_column.embedding_column(user_class, dimension=FLAGS.embedding_dim)
cust_tag_emb = tf.feature_column.embedding_column(cust_tag, dimension=FLAGS.embedding_dim)
feed_word_emb = tf.feature_column.embedding_column(feed_word, dimension=FLAGS.embedding_dim)
user_bhvword_emb = tf.feature_column.embedding_column(user_bhvword, dimension=FLAGS.embedding_dim)
user_bhvtag_emb = tf.feature_column.embedding_column(user_bhvtag, dimension=FLAGS.embedding_dim)
#return 0, 0, [[user_class_emb], [cust_tag_emb], [feed_word_emb], [user_bhvword_emb], [user_bhvtag_emb]]
return 0, 0, [[user_class_emb], [cust_tag_emb], [feed_word_emb], [user_bhvword_emb], [user_bhvtag_emb]]
'''
def emb_conv_nn(features, labels, mode, params):
embs = [] # embedding list of user and ad
for idx, feature_v in enumerate(params['embedding_feature']):
tensor = tf.feature_column.input_layer(features, feature_v)
embs.append(tensor)
# kernel = tf.Variable(tf.random_uniform([1, 2, 2], -0.5, 0.5), trainable=True, name='kernel') # [1, 2, 2] for [row, col, kernel_num] cause the shape of t2d(v_user,v_ad) is 100*2, so conv on row-dim make no means
# kernel-2: [1,1] / [1,-1] / [dot_product]
# kernel-3: [-1,1,1] / [1,-1,1] / [1,1,-1] / [dot_product]
kernel_2 = tf.constant([[[1.0, 1.0], [1.0, -1.0]]], name='kernel_2')
kernel_3 = tf.constant([[[-1.0,1.0,1.0], [1.0,-1.0,1.0], [1.0,1.0,-1.0]]], name='kernel_3')
kernel_4 = tf.constant([[[-1.0,1.0,1.0,1.0], [1.0,-1.0,1.0,1.0], [1.0,1.0,-1.0,1.0], [1.0,1.0,1.0,-1.0]]], name='kernel_4')
print('---------------------\ninput_vector:\t'+str(len(embs))+'\t'+str(embs))
multi_convs = []
for v1,v2 in itertools.combinations(embs, 2):
t2d = tf.stack([v1, v2], axis=2)
t2d_conv = tf.nn.conv1d(t2d, kernel_2, 1, 'VALID')
multi_convs.append(t2d_conv)
# v1.*v2
t2d_dot = tf.expand_dims(tf.multiply(v1, v2), 2)
multi_convs.append(t2d_dot)
print('2d:\t\t'+str(t2d)+'\t\t'+str(t2d_conv)+'\t\t'+str(t2d_dot))
if len(embs) > 2:
for v1,v2,v3 in itertools.combinations(embs, 3):
t3d = tf.stack([v1, v2, v3], axis=2)
t3d_conv = tf.nn.conv1d(t3d, kernel_3, 1, 'VALID')
multi_convs.append(t3d_conv)
# v1.*v2.*v3
t3d_dot = tf.expand_dims(tf.multiply(tf.multiply(v1, v2), v3), 2)
multi_convs.append(t3d_dot)
print('3d:\t\t'+str(t3d)+'\t\t'+str(t3d_conv)+'\t\t'+str(t3d_dot))
if len(embs) > 3:
for v1,v2,v3,v4 in itertools.combinations(embs, 4):
t4d = tf.stack([v1, v2, v3, v4], axis=2)
t4d_conv = tf.nn.conv1d(t4d, kernel_4, 1, 'VALID')
multi_convs.append(t4d_conv)
# v1.*v2.*v3
t4d_dot = tf.expand_dims(tf.multiply(tf.multiply(tf.multiply(v1, v2), v3), v4), 2)
multi_convs.append(t4d_dot)
print('4d:\t\t'+str(t4d)+'\t\t'+str(t4d_conv)+'\t\t'+str(t4d_dot))
conv_layer = tf.concat(multi_convs, axis=2)
print('---------------------\nconv_layer:\t'+str(conv_layer))
pool_1 = None #pool_1 = tf.nn.pool(conv_layer, [3], 'AVG', 'VALID', strides=[1])
pool_2 = None #pool_2 = tf.nn.pool(conv_layer, [7], 'AVG', 'VALID', strides=[3])
pool_3 = None #tf.nn.pool(conv_layer, [13], 'AVG','VALID', strides=[6])
pool_4 = tf.nn.pool(conv_layer, [25], 'AVG','VALID', strides=[13])
pool_5 = tf.nn.pool(conv_layer, [50], 'AVG','VALID', strides=[25])
pool_6 = tf.nn.pool(conv_layer, [100], 'AVG','VALID', strides=[50])
print('---------------------\npool:\t'+str(pool_1)+'\t'+str(pool_2)+'\t'+str(pool_3)+'\t'+str(pool_4)+'\t'+str(pool_5)+'\t'+str(pool_6))
#flatten_layer = tf.concat([tf.layers.flatten(pool_1), tf.layers.flatten(pool_2), tf.layers.flatten(pool_3), tf.layers.flatten(pool_4), tf.layers.flatten(pool_5)], axis=1, name='flatten_layer')
flatten_layer = tf.concat([tf.layers.flatten(pool_5), tf.layers.flatten(pool_6)], axis=1, name='flatten_layer')
print('---------------------\nflatten: '+str(flatten_layer)+'\n')
he_init = tf.contrib.layers.variance_scaling_initializer(mode="FAN_AVG")
logits = tf.layers.dense(flatten_layer, 1, activation=tf.nn.relu, kernel_regularizer=tf.contrib.layers.l2_regularizer(0.03), bias_regularizer=tf.contrib.layers.l2_regularizer(0.01), kernel_initializer=he_init, bias_initializer=tf.zeros_initializer(), name='logits')
# dropout is not suitable for pretrain
#if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
# logits = tf.layers.dropout(logits, params['dropout_rate'], training=(mode == tf.estimator.ModeKeys.TRAIN))
return (logits, flatten_layer)
def my_model_pretrain(features, labels, mode, params):
with tf.variable_scope('emb_conv_model'):
logits, flatten_layer = emb_conv_nn(features, labels, mode, params)
with tf.variable_scope('logits', reuse=True):
w = tf.get_variable('kernel')
bias = tf.get_variable('bias')
pred = tf.sigmoid(logits, name="pred")
if mode == tf.estimator.ModeKeys.PREDICT:
predictions = {
'pred': pred,
}
export_outputs = {
'prediction': tf.estimator.export.PredictOutput(predictions)
}
return tf.estimator.EstimatorSpec(mode, predictions=predictions, export_outputs=export_outputs)
# define loss
y = tf.reshape(labels['label'], [-1,1])
loss = tf.reduce_sum(tf.keras.backend.binary_crossentropy(y, pred), name="loss")
auc = tf.metrics.auc(y, pred)
avg = tf.metrics.mean(y)
pred_avg = tf.metrics.mean(pred)
w_avg = tf.metrics.mean(w)
bias_avg = tf.metrics.mean(bias)
metrics = {'auc': auc, 'ctr': avg, 'pred': pred_avg, 'bias': bias_avg, 'w': w_avg }
tf.summary.scalar('auc', auc[1])
tf.summary.scalar('ctr_avg', avg[1])
tf.summary.scalar('pred_avg', pred_avg[1])
tf.summary.scalar('w', w_avg[1])
tf.summary.scalar('bias', bias_avg[1])
logging_hook = tf.train.LoggingTensorHook({"loss" : loss, "auc" : auc[1], "ctr" : avg[1], "pred" : pred_avg[1], "w" : w_avg[1], "bias":bias_avg[1]}, every_n_iter=20)
if mode == tf.estimator.ModeKeys.EVAL:
return tf.estimator.EstimatorSpec(mode, loss=loss, eval_metric_ops=metrics, training_hooks = [logging_hook])
def init_fn(scaffold, sess):
pass #init_op = tf.initializers.global_variables()
#sess.run(init_op)
scaffold = tf.train.Scaffold(init_fn=init_fn)
# Create training op.
assert mode == tf.estimator.ModeKeys.TRAIN
#init_op = tf.initializers.global_variables()
#optimizer = tf.train.AdagradOptimizer(learning_rate=params['learning_rate'])
optimizer = tf.train.AdamOptimizer(learning_rate=params['learning_rate'])
#optimizer = tf.train.FtrlOptimizer(learning_rate=params['learning_rate'])
train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step())
return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op, eval_metric_ops=metrics, training_hooks = [logging_hook]) #, scaffold=scaffold)
def my_model(features, labels, mode, params):
model_type = params['model_type']
print('build model: '+str(params['model_type']))
logits_arr = []
if model_type.find('conv') >= 0:
with tf.variable_scope('emb_conv_model'):
logits_embconv, flatten_layer = emb_conv_nn(features, labels, mode, params)
if model_type.find('wide') >= 0:
with tf.variable_scope('linear_model'):
# use linear_model for fast-implement of LR
# example: https://www.jianshu.com/p/fceb64c790f3
logits_linear = tf.feature_column.linear_model(features, params['wide_feature'])
lnr_mean, lnr_vari = tf.nn.moments(logits_linear, [0])
logits_arr.append(logits_linear)
if model_type.find('deep') >= 0:
with tf.variable_scope('dnn_model'):
input_layer = tf.feature_column.input_layer(features, params['deep_feature'])
# concat emb_conv flatten tensor to input_layer
if model_type.find('conv') >= 0:
he_init = tf.contrib.layers.variance_scaling_initializer(mode="FAN_AVG")
delay_conv_layer = tf.layers.dense(tf.nn.dropout(flatten_layer, 0.3), params['usm_units'], activation=tf.nn.relu, name='delay_conv_layer', kernel_initializer=he_init, bias_initializer=tf.zeros_initializer()) #, kernel_regularizer=tf.contrib.layers.l1_regularizer(1.0))
delay_conv_layer = tf.nn.dropout(delay_conv_layer, 0.3)
print('---------------------\nwith_usm:\t'+str(flatten_layer)+' --> '+str(delay_conv_layer))
if FLAGS.only_usm_input:
input_layer = delay_conv_layer
else:
input_layer = tf.concat([input_layer, delay_conv_layer], axis=1)
print('---------------------\ninput deep layer:\t'+str(input_layer))
units = params['hidden_units']
layers = [input_layer]
for l_id, l_size in enumerate(units):
he_init = tf.contrib.layers.variance_scaling_initializer(mode="FAN_AVG")
layer = tf.layers.dense(layers[-1], l_size, activation=tf.nn.relu, name='hidden_layer_'+str(l_id), kernel_regularizer=tf.contrib.layers.l1_regularizer(0.03), bias_regularizer=tf.contrib.layers.l2_regularizer(0.01), kernel_initializer=he_init, bias_initializer=tf.zeros_initializer())
layers.append(layer)
print('---------------------\nhidden_layer:\t'+str(layer))
last_layer = layers[-1]
#if usm_layer is not None:
# last_layer = tf.concat([layers[-1], usm_layer], axis=1)
print('---------------------\nlast_layer:\t'+str(last_layer))
logits_dnn = tf.layers.dense(last_layer, 1, activation=None, name='output_layer', kernel_regularizer=tf.contrib.layers.l1_regularizer(0.03), bias_regularizer=tf.contrib.layers.l2_regularizer(0.01), kernel_initializer=he_init, bias_initializer=tf.zeros_initializer())
logits_arr.append(logits_dnn)
dnn_mean, dnn_vari = tf.nn.moments(logits_dnn, [0])
logits = tf.add_n(logits_arr, name='pred_logits')
pred = tf.sigmoid(logits, name="pred") #pred = tf.Print(ctr_pred, [ctr_pred, 'pred'], message='Debug message:',summarize=100)
if mode == tf.estimator.ModeKeys.PREDICT:
predictions = {
'pred': pred,
}
export_outputs = {
'prediction': tf.estimator.export.PredictOutput(predictions)
}
return tf.estimator.EstimatorSpec(mode, predictions=predictions, export_outputs=export_outputs)
# define loss
y = tf.reshape(labels['label'], [-1,1])
loss = tf.reduce_sum(tf.nn.sigmoid_cross_entropy_with_logits(labels=y, logits=logits), name='loss')
auc = tf.metrics.auc(y, pred)
avg = tf.metrics.mean(y)
pred_avg = tf.metrics.mean(pred)
metrics = {'auc':auc, 'ctr':avg, 'pred':pred_avg}
tf.summary.scalar('auc', auc[1])
tf.summary.scalar('ctr_avg', avg[1])
tf.summary.scalar('pred_avg', pred_avg[1])
hook_dict = {"loss":loss, "auc":auc[1], "ctr":avg[1], "pred":pred_avg[1]}
if model_type.find('wide') >= 0:
lnr_mean_avg = tf.metrics.mean(lnr_mean)
lnr_vari_avg = tf.metrics.mean(lnr_vari)
metrics.update({'lnr_mean':lnr_mean_avg, 'lnr_vari':lnr_vari_avg})
tf.summary.scalar('lnr_mean', lnr_mean_avg[1])
tf.summary.scalar('lnr_vari', lnr_vari_avg[1])
hook_dict.update({'lnr_mean':lnr_mean_avg[1], 'lnr_vari':lnr_vari_avg[1]})
if model_type.find('deep') >= 0:
dnn_mean_avg = tf.metrics.mean(dnn_mean)
dnn_vari_avg = tf.metrics.mean(dnn_vari)
metrics.update({'dnn_mean':dnn_mean_avg, 'dnn_vari':dnn_vari_avg})
tf.summary.scalar('dnn_mean', dnn_mean_avg[1])
tf.summary.scalar('dnn_vari', dnn_vari_avg[1])
hook_dict.update({'dnn_mean':dnn_mean_avg[1], 'dnn_vari':dnn_vari_avg[1]})
logging_hook = tf.train.LoggingTensorHook(hook_dict, every_n_iter=20)
if mode == tf.estimator.ModeKeys.EVAL:
return tf.estimator.EstimatorSpec(mode, loss=loss, eval_metric_ops=metrics, training_hooks = [logging_hook])
# Create training op.
assert mode == tf.estimator.ModeKeys.TRAIN
# AdagradOptimizer/AdamOptimizer
# train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step())
train_ops = []
if model_type.find('wide') >= 0:
# FtrlOptimizer
opt_linear = tf.train.AdamOptimizer(learning_rate=params['learning_rate'])
grads_linear = opt_linear.compute_gradients(loss, tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='linear_model'))
trainer_linear = opt_linear.apply_gradients(grads_linear, global_step=tf.train.get_global_step())
train_ops.append(trainer_linear)
if model_type.find('deep') >= 0:
flag = False
if model_type.find('wide') >= 0:
flag = True
opt_dnn = tf.train.AdamOptimizer(learning_rate=params['learning_rate'])
grads_dnn = opt_dnn.compute_gradients(loss, tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='dnn_model'))
if flag:
trainer_dnn = opt_dnn.apply_gradients(grads_dnn)
else:
trainer_dnn = opt_dnn.apply_gradients(grads_dnn, global_step=tf.train.get_global_step())
train_ops.append(trainer_dnn)
'''
# List variables
print('linear_model variables:')
for k in tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='linear_model'):
print(k)
print('dnn_model variables:')
for k in tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='dnn_model'):
print(k)
'''
train_op = tf.group(train_ops)
return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op, eval_metric_ops=metrics, training_hooks = [logging_hook])
def build_estimator(model_dir, model_type):
"""Build an estimator appropriate for the given model type."""
set_tfconfig_environ()
wide_columns, deep_columns, embedding_columns, usm_deep_columns = build_model_columns() #build_model_columns()
global _GLOBAL_FEATURES
_GLOBAL_FEATURES = wide_columns + deep_columns + embedding_columns
hidden_units = [300, 100, 80] #[100, 75, 50, 25] #[400, 200, 100] #[300, 100, 80, 80]
# Create a tf.estimator.RunConfig to ensure the model is run on CPU, which
# trains faster than GPU for this model.
run_config = tf.estimator.RunConfig().replace(
session_config=tf.ConfigProto(device_count={'GPU': 0}),
save_checkpoints_secs = FLAGS.save_checkpoints_steps, #300
keep_checkpoint_max = 3,
model_dir=model_dir)
if FLAGS.pretrain == 'no':
model = tf.estimator.Estimator(
model_fn=my_model,
params={
'wide_feature': wide_columns,
'deep_feature': deep_columns, # if model_type.find('conv') < 0 else usm_deep_columns,
'embedding_feature': embedding_columns,
'hidden_units': hidden_units, #FLAGS.hidden_units.split(','),
'usm_units': 100,
'learning_rate': FLAGS.learning_rate,
'dropout_rate': FLAGS.dropout_rate,
'model_type': model_type
},
config=tf.estimator.RunConfig(model_dir=FLAGS.checkpoints_dir, save_checkpoints_steps=FLAGS.save_checkpoints_steps)
)
else:
model = tf.estimator.Estimator(
model_fn=my_model_pretrain,
params={
'embedding_feature': embedding_columns,
'learning_rate': FLAGS.learning_rate,
'dropout_rate': FLAGS.dropout_rate
},
config=tf.estimator.RunConfig(model_dir=FLAGS.checkpoints_dir, save_checkpoints_steps=FLAGS.save_checkpoints_steps)
)
return model
def parse_csv(value):
columns = tf.decode_csv(value, record_defaults=_CSV_COLUMN_DEFAULTS)
features = dict(zip(_CSV_COLUMNS, columns))
labels = features.pop('ig')
return features, { 'label': tf.to_float(labels) } #tf.equal(labels, 1)
def input_fn(data_file, num_epochs, shuffle_buffer_size, batch_size):
#"""Generate an input function for the Estimator."""
#assert tf.gfile.Exists(data_file), (
# '%s not found. Please make sure you have either run data_download.py or '
# 'set both arguments --train_data and --test_data.' % data_file)
files = tf.data.Dataset.list_files(data_file)
# Extract lines from input files using the Dataset API.
dataset = files.apply(
tf.contrib.data.parallel_interleave(tf.data.TextLineDataset, cycle_length=5, sloppy=True)
)
#dataset = tf.data.TextLineDataset(files)
dataset = dataset.shuffle(buffer_size=shuffle_buffer_size)
dataset = dataset.map(parse_csv, num_parallel_calls=5)
# We call repeat after shuffling, rather than before, to prevent separate
# epochs from blending together.
dataset = dataset.repeat(num_epochs)
dataset = dataset.batch(batch_size)
dataset = dataset.prefetch(batch_size)
'''
# save the input tensor name of the graph
# use for inference
input_tensor_map = dict()
dataset_iter = dataset.make_initializable_iterator()
features, labels = dataset_iter.get_next()
for input_name, tensor in features.items():
input_tensor_map[input_name] = tensor.name
with open(os.path.join(FLAGS.checkpoints_dir, 'input_tensor_map.pickle'), 'wb') as f:
pickle.dump(input_tensor_map, f, protocol=pickle.HIGHEST_PROTOCOL)
tf.add_to_collection(tf.GraphKeys.TABLE_INITIALIZERS, dataset_iter.initializer)
'''
return dataset
def eval_input_fn(data_file, batch_size):
# Extract lines from input files using the Dataset API.
files = tf.data.Dataset.list_files(data_file)
dataset = files.apply(tf.contrib.data.parallel_interleave(tf.data.TextLineDataset, cycle_length=1))
#dataset = tf.data.TextLineDataset(files)
dataset = dataset.map(parse_csv, num_parallel_calls=5)
dataset = dataset.batch(batch_size)
return dataset
def main(unused_argv):
# Clean up the model directory if present
#shutil.rmtree(FLAGS.checkpoints_dir, ignore_errors=True)
model = build_estimator(FLAGS.checkpoints_dir, FLAGS.model_type)
if isinstance(FLAGS.data_dir, str) and os.path.isdir(FLAGS.data_dir):
train_files = [FLAGS.data_dir + '/' + x for x in os.listdir(FLAGS.data_dir)] if os.path.isdir(
FLAGS.data_dir) else FLAGS.data_dir
else:
train_files = FLAGS.data_dir
if isinstance(FLAGS.validate_dir, str) and os.path.isdir(FLAGS.validate_dir):
eval_files = [FLAGS.validate_dir + '/' + x for x in os.listdir(FLAGS.validate_dir)] if os.path.isdir(
FLAGS.validate_dir) else FLAGS.validate_dir
else:
eval_files = FLAGS.validate_dir
print('train files: '+str(train_files))
print('eval files: '+str(eval_files))
if FLAGS.pretrain == 'no':
premodels = FLAGS.embedding_model+' '+str(os.path.exists(FLAGS.embedding_model+'.meta'))
print('pre models: '+str(premodels))
# train process
train_spec = tf.estimator.TrainSpec(
input_fn=lambda: input_fn(train_files, FLAGS.num_epochs, FLAGS.shuffle_buffer_size, FLAGS.batch_size),
max_steps=FLAGS.num_epochs
)
input_fn_for_eval = lambda: eval_input_fn(eval_files, FLAGS.batch_size)
eval_spec = tf.estimator.EvalSpec(input_fn=input_fn_for_eval, throttle_secs=300)
print("before train and evaluate")
tf.estimator.train_and_evaluate(model, train_spec, eval_spec)
print("after train and evaluate")
# Evaluate accuracy.
results = model.evaluate(input_fn=input_fn_for_eval)
for key in sorted(results): print('%s: %s' % (key, results[key]))
print("after evaluate")
if FLAGS.job_name == "worker" and FLAGS.task_index == 0:
print("exporting model ...")
feature_spec = tf.feature_column.make_parse_example_spec(_GLOBAL_FEATURES)
print("feature spec: "+str(feature_spec))
serving_input_receiver_fn = tf.estimator.export.build_parsing_serving_input_receiver_fn(feature_spec)
model.export_savedmodel(FLAGS.output_model, serving_input_receiver_fn)
# tf.contrib.predictor.from_saved_model(export_dir=)
print("quit main")
'''
# Train and evaluate the model every `FLAGS.epochs_per_eval` epochs.
for n in range(FLAGS.train_epochs // FLAGS.epochs_per_eval):
model.train(input_fn=lambda: input_fn(
FLAGS.train_data, FLAGS.epochs_per_eval, True, FLAGS.batch_size))
results = model.evaluate(input_fn=lambda: input_fn(
FLAGS.test_data, 1, False, FLAGS.batch_size))
# Display evaluation metrics
print('Results at epoch', (n + 1) * FLAGS.epochs_per_eval)
print('-' * 60)
for key in sorted(results):
print('%s: %s' % (key, results[key]))
'''
def dump():
#files = tf.data.Dataset.list_files('model2')
#dataset = files.apply(tf.contrib.data.parallel_interleave(tf.data.TextLineDataset, cycle_length=1))
with tf.Session() as sess:
data_file = './model_zoo/wide_deep_emb_conv/model.ckpt-108.meta'
saver = tf.train.import_meta_graph(data_file, clear_devices=True)
print('saver ok')
saver.restore(sess, './model_zoo/wide_deep_emb_conv/model.ckpt-108')
graph = tf.get_default_graph()
w = []
bias = 0
# list all trainable variables
dim0 = 0
#for var in tf.get_default_graph().get_collection(tf.GraphKeys.MODEL_VARIABLES): #"trainable_variables"):
print('\nglobal variables:')
for var in tf.get_default_graph().get_collection(tf.GraphKeys.GLOBAL_VARIABLES):
vv = sess.run(var)
print(str(var.name) + '\t\t' + str(vv.shape)) #+str(vv))
#print(vv)
#idx = int(re.split('_|:', var.name)[1])
#if var.name.split('/')[0] == 'weights': #'weights/part_0:0':
# w.extend(list(vv[:,0]))
# dim0 += vv.shape[0]
#if var.name == 'params/bias:0':
# bias = vv
print('\nlocal variables:')
for var in tf.get_default_graph().get_collection(tf.GraphKeys.LOCAL_VARIABLES):
print(str(var.name) + '\t\t' + str(vv.shape))
print('\ntrainable variables:')
for var in tf.get_default_graph().get_collection(tf.GraphKeys.TRAINABLE_VARIABLES):
vv = sess.run(var)
print(str(var.name) + '\t\t' + str(np.shape(vv)))
#print(vv)
print('\ndim: '+str(dim0)+', len(w):'+str(len(w)))
def inference():
pass
if __name__ == '__main__':
if "CUDA_VISIBLE_DEVICES" in os.environ:
print("CUDA_VISIBLE_DEVICES:", os.environ["CUDA_VISIBLE_DEVICES"])
if FLAGS.run_on_cluster: parse_argument()
tf.logging.set_verbosity(tf.logging.INFO)
if FLAGS.task == 'dump':
dump()
else:
tf.app.run(main=main)