-
Notifications
You must be signed in to change notification settings - Fork 16
/
tfutil.py
executable file
·1459 lines (1291 loc) · 67.6 KB
/
tfutil.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
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
#
#
#Attribution-NonCommercial 4.0 International
#
#=======================================================================
#
#Creative Commons Corporation ("Creative Commons") is not a law firm and
#does not provide legal services or legal advice. Distribution of
#Creative Commons public licenses does not create a lawyer-client or
#other relationship. Creative Commons makes its licenses and related
#information available on an "as-is" basis. Creative Commons gives no
#warranties regarding its licenses, any material licensed under their
#terms and conditions, or any related information. Creative Commons
#disclaims all liability for damages resulting from their use to the
#fullest extent possible.
#
#Using Creative Commons Public Licenses
#
#Creative Commons public licenses provide a standard set of terms and
#conditions that creators and other rights holders may use to share
#original works of authorship and other material subject to copyright
#and certain other rights specified in the public license below. The
#following considerations are for informational purposes only, are not
#exhaustive, and do not form part of our licenses.
#
# Considerations for licensors: Our public licenses are
# intended for use by those authorized to give the public
# permission to use material in ways otherwise restricted by
# copyright and certain other rights. Our licenses are
# irrevocable. Licensors should read and understand the terms
# and conditions of the license they choose before applying it.
# Licensors should also secure all rights necessary before
# applying our licenses so that the public can reuse the
# material as expected. Licensors should clearly mark any
# material not subject to the license. This includes other CC-
# licensed material, or material used under an exception or
# limitation to copyright. More considerations for licensors:
# wiki.creativecommons.org/Considerations_for_licensors
#
# Considerations for the public: By using one of our public
# licenses, a licensor grants the public permission to use the
# licensed material under specified terms and conditions. If
# the licensor's permission is not necessary for any reason--for
# example, because of any applicable exception or limitation to
# copyright--then that use is not regulated by the license. Our
# licenses grant only permissions under copyright and certain
# other rights that a licensor has authority to grant. Use of
# the licensed material may still be restricted for other
# reasons, including because others have copyright or other
# rights in the material. A licensor may make special requests,
# such as asking that all changes be marked or described.
# Although not required by our licenses, you are encouraged to
# respect those requests where reasonable. More_considerations
# for the public:
# wiki.creativecommons.org/Considerations_for_licensees
#
#=======================================================================
#
#Creative Commons Attribution-NonCommercial 4.0 International Public
#License
#
#By exercising the Licensed Rights (defined below), You accept and agree
#to be bound by the terms and conditions of this Creative Commons
#Attribution-NonCommercial 4.0 International Public License ("Public
#License"). To the extent this Public License may be interpreted as a
#contract, You are granted the Licensed Rights in consideration of Your
#acceptance of these terms and conditions, and the Licensor grants You
#such rights in consideration of benefits the Licensor receives from
#making the Licensed Material available under these terms and
#conditions.
#
#
#Section 1 -- Definitions.
#
# a. Adapted Material means material subject to Copyright and Similar
# Rights that is derived from or based upon the Licensed Material
# and in which the Licensed Material is translated, altered,
# arranged, transformed, or otherwise modified in a manner requiring
# permission under the Copyright and Similar Rights held by the
# Licensor. For purposes of this Public License, where the Licensed
# Material is a musical work, performance, or sound recording,
# Adapted Material is always produced where the Licensed Material is
# synched in timed relation with a moving image.
#
# b. Adapter's License means the license You apply to Your Copyright
# and Similar Rights in Your contributions to Adapted Material in
# accordance with the terms and conditions of this Public License.
#
# c. Copyright and Similar Rights means copyright and/or similar rights
# closely related to copyright including, without limitation,
# performance, broadcast, sound recording, and Sui Generis Database
# Rights, without regard to how the rights are labeled or
# categorized. For purposes of this Public License, the rights
# specified in Section 2(b)(1)-(2) are not Copyright and Similar
# Rights.
# d. Effective Technological Measures means those measures that, in the
# absence of proper authority, may not be circumvented under laws
# fulfilling obligations under Article 11 of the WIPO Copyright
# Treaty adopted on December 20, 1996, and/or similar international
# agreements.
#
# e. Exceptions and Limitations means fair use, fair dealing, and/or
# any other exception or limitation to Copyright and Similar Rights
# that applies to Your use of the Licensed Material.
#
# f. Licensed Material means the artistic or literary work, database,
# or other material to which the Licensor applied this Public
# License.
#
# g. Licensed Rights means the rights granted to You subject to the
# terms and conditions of this Public License, which are limited to
# all Copyright and Similar Rights that apply to Your use of the
# Licensed Material and that the Licensor has authority to license.
#
# h. Licensor means the individual(s) or entity(ies) granting rights
# under this Public License.
#
# i. NonCommercial means not primarily intended for or directed towards
# commercial advantage or monetary compensation. For purposes of
# this Public License, the exchange of the Licensed Material for
# other material subject to Copyright and Similar Rights by digital
# file-sharing or similar means is NonCommercial provided there is
# no payment of monetary compensation in connection with the
# exchange.
#
# j. Share means to provide material to the public by any means or
# process that requires permission under the Licensed Rights, such
# as reproduction, public display, public performance, distribution,
# dissemination, communication, or importation, and to make material
# available to the public including in ways that members of the
# public may access the material from a place and at a time
# individually chosen by them.
#
# k. Sui Generis Database Rights means rights other than copyright
# resulting from Directive 96/9/EC of the European Parliament and of
# the Council of 11 March 1996 on the legal protection of databases,
# as amended and/or succeeded, as well as other essentially
# equivalent rights anywhere in the world.
#
# l. You means the individual or entity exercising the Licensed Rights
# under this Public License. Your has a corresponding meaning.
#
#
#Section 2 -- Scope.
#
# a. License grant.
#
# 1. Subject to the terms and conditions of this Public License,
# the Licensor hereby grants You a worldwide, royalty-free,
# non-sublicensable, non-exclusive, irrevocable license to
# exercise the Licensed Rights in the Licensed Material to:
#
# a. reproduce and Share the Licensed Material, in whole or
# in part, for NonCommercial purposes only; and
#
# b. produce, reproduce, and Share Adapted Material for
# NonCommercial purposes only.
#
# 2. Exceptions and Limitations. For the avoidance of doubt, where
# Exceptions and Limitations apply to Your use, this Public
# License does not apply, and You do not need to comply with
# its terms and conditions.
#
# 3. Term. The term of this Public License is specified in Section
# 6(a).
#
# 4. Media and formats; technical modifications allowed. The
# Licensor authorizes You to exercise the Licensed Rights in
# all media and formats whether now known or hereafter created,
# and to make technical modifications necessary to do so. The
# Licensor waives and/or agrees not to assert any right or
# authority to forbid You from making technical modifications
# necessary to exercise the Licensed Rights, including
# technical modifications necessary to circumvent Effective
# Technological Measures. For purposes of this Public License,
# simply making modifications authorized by this Section 2(a)
# (4) never produces Adapted Material.
#
# 5. Downstream recipients.
#
# a. Offer from the Licensor -- Licensed Material. Every
# recipient of the Licensed Material automatically
# receives an offer from the Licensor to exercise the
# Licensed Rights under the terms and conditions of this
# Public License.
#
# b. No downstream restrictions. You may not offer or impose
# any additional or different terms or conditions on, or
# apply any Effective Technological Measures to, the
# Licensed Material if doing so restricts exercise of the
# Licensed Rights by any recipient of the Licensed
# Material.
#
# 6. No endorsement. Nothing in this Public License constitutes or
# may be construed as permission to assert or imply that You
# are, or that Your use of the Licensed Material is, connected
# with, or sponsored, endorsed, or granted official status by,
# the Licensor or others designated to receive attribution as
# provided in Section 3(a)(1)(A)(i).
#
# b. Other rights.
#
# 1. Moral rights, such as the right of integrity, are not
# licensed under this Public License, nor are publicity,
# privacy, and/or other similar personality rights; however, to
# the extent possible, the Licensor waives and/or agrees not to
# assert any such rights held by the Licensor to the limited
# extent necessary to allow You to exercise the Licensed
# Rights, but not otherwise.
#
# 2. Patent and trademark rights are not licensed under this
# Public License.
#
# 3. To the extent possible, the Licensor waives any right to
# collect royalties from You for the exercise of the Licensed
# Rights, whether directly or through a collecting society
# under any voluntary or waivable statutory or compulsory
# licensing scheme. In all other cases the Licensor expressly
# reserves any right to collect such royalties, including when
# the Licensed Material is used other than for NonCommercial
# purposes.
#
#
#Section 3 -- License Conditions.
#
#Your exercise of the Licensed Rights is expressly made subject to the
#following conditions.
#
# a. Attribution.
#
# 1. If You Share the Licensed Material (including in modified
# form), You must:
#
# a. retain the following if it is supplied by the Licensor
# with the Licensed Material:
#
# i. identification of the creator(s) of the Licensed
# Material and any others designated to receive
# attribution, in any reasonable manner requested by
# the Licensor (including by pseudonym if
# designated);
#
# ii. a copyright notice;
#
# iii. a notice that refers to this Public License;
#
# iv. a notice that refers to the disclaimer of
# warranties;
#
# v. a URI or hyperlink to the Licensed Material to the
# extent reasonably practicable;
#
# b. indicate if You modified the Licensed Material and
# retain an indication of any previous modifications; and
#
# c. indicate the Licensed Material is licensed under this
# Public License, and include the text of, or the URI or
# hyperlink to, this Public License.
#
# 2. You may satisfy the conditions in Section 3(a)(1) in any
# reasonable manner based on the medium, means, and context in
# which You Share the Licensed Material. For example, it may be
# reasonable to satisfy the conditions by providing a URI or
# hyperlink to a resource that includes the required
# information.
#
# 3. If requested by the Licensor, You must remove any of the
# information required by Section 3(a)(1)(A) to the extent
# reasonably practicable.
#
# 4. If You Share Adapted Material You produce, the Adapter's
# License You apply must not prevent recipients of the Adapted
# Material from complying with this Public License.
#
#
#Section 4 -- Sui Generis Database Rights.
#
#Where the Licensed Rights include Sui Generis Database Rights that
#apply to Your use of the Licensed Material:
#
# a. for the avoidance of doubt, Section 2(a)(1) grants You the right
# to extract, reuse, reproduce, and Share all or a substantial
# portion of the contents of the database for NonCommercial purposes
# only;
#
# b. if You include all or a substantial portion of the database
# contents in a database in which You have Sui Generis Database
# Rights, then the database in which You have Sui Generis Database
# Rights (but not its individual contents) is Adapted Material; and
#
# c. You must comply with the conditions in Section 3(a) if You Share
# all or a substantial portion of the contents of the database.
#
#For the avoidance of doubt, this Section 4 supplements and does not
#replace Your obligations under this Public License where the Licensed
#Rights include other Copyright and Similar Rights.
#
#
#Section 5 -- Disclaimer of Warranties and Limitation of Liability.
#
# a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
# EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
# AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
# ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
# IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
# WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
# PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
# ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
# KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
# ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
#
# b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
# TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
# NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
# INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
# COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
# USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
# ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
# DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
# IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
#
# c. The disclaimer of warranties and limitation of liability provided
# above shall be interpreted in a manner that, to the extent
# possible, most closely approximates an absolute disclaimer and
# waiver of all liability.
#
#
#Section 6 -- Term and Termination.
#
# a. This Public License applies for the term of the Copyright and
# Similar Rights licensed here. However, if You fail to comply with
# this Public License, then Your rights under this Public License
# terminate automatically.
#
# b. Where Your right to use the Licensed Material has terminated under
# Section 6(a), it reinstates:
#
# 1. automatically as of the date the violation is cured, provided
# it is cured within 30 days of Your discovery of the
# violation; or
#
# 2. upon express reinstatement by the Licensor.
#
# For the avoidance of doubt, this Section 6(b) does not affect any
# right the Licensor may have to seek remedies for Your violations
# of this Public License.
#
# c. For the avoidance of doubt, the Licensor may also offer the
# Licensed Material under separate terms or conditions or stop
# distributing the Licensed Material at any time; however, doing so
# will not terminate this Public License.
#
# d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
# License.
#
#
#Section 7 -- Other Terms and Conditions.
#
# a. The Licensor shall not be bound by any additional or different
# terms or conditions communicated by You unless expressly agreed.
#
# b. Any arrangements, understandings, or agreements regarding the
# Licensed Material not stated herein are separate from and
# independent of the terms and conditions of this Public License.
#
#
#Section 8 -- Interpretation.
#
# a. For the avoidance of doubt, this Public License does not, and
# shall not be interpreted to, reduce, limit, restrict, or impose
# conditions on any use of the Licensed Material that could lawfully
# be made without permission under this Public License.
#
# b. To the extent possible, if any provision of this Public License is
# deemed unenforceable, it shall be automatically reformed to the
# minimum extent necessary to make it enforceable. If the provision
# cannot be reformed, it shall be severed from this Public License
# without affecting the enforceability of the remaining terms and
# conditions.
#
# c. No term or condition of this Public License will be waived and no
# failure to comply consented to unless expressly agreed to by the
# Licensor.
#
# d. Nothing in this Public License constitutes or may be interpreted
# as a limitation upon, or waiver of, any privileges and immunities
# that apply to the Licensor or You, including from the legal
# processes of any jurisdiction or authority.
#
#=======================================================================
#
#Creative Commons is not a party to its public
#licenses. Notwithstanding, Creative Commons may elect to apply one of
#its public licenses to material it publishes and in those instances
#will be considered the "Licensor." The text of the Creative Commons
#public licenses is dedicated to the public domain under the CC0 Public
#Domain Dedication. Except for the limited purpose of indicating that
#material is shared under a Creative Commons public license or as
#otherwise permitted by the Creative Commons policies published at
#creativecommons.org/policies, Creative Commons does not authorize the
#use of the trademark "Creative Commons" or any other trademark or logo
#of Creative Commons without its prior written consent including,
#without limitation, in connection with any unauthorized modifications
#to any of its public licenses or any other arrangements,
#understandings, or agreements concerning use of licensed material. For
#the avoidance of doubt, this paragraph does not form part of the
#public licenses.
#
#Creative Commons may be contacted at creativecommons.org.
import os
import sys
import inspect
import importlib
import imp
import numpy as np
from collections import OrderedDict
import tensorflow as tf
from tensorflow.python import debug as tf_debug
import pdb
import os
import scipy
import misc
# ----------------------------------------------------------------------------
# Convenience.
def run(*args, **kwargs): # Run the specified ops in the default session.
# GUI tensorflow debugger using tensorboard
# session = tf.Session()
# with tf_debug.TensorBoardDebugWrapperSession(session, 'localhost:6064') as sess:
# sess.run(*args, **kwargs)
# return
# CLI tensorflow debugger
# session = tf.Session()
# with tf_debug.LocalCLIDebugWrapperSession(session) as sess:
# sess.run(*args, **kwargs)
# return
# with tf.Session() as sess:
# import pdb
# pdb.set_trace()
return tf.get_default_session().run(*args, **kwargs)
def is_tf_expression(x):
return isinstance(x, tf.Tensor) or isinstance(x, tf.Variable) or isinstance(x, tf.Operation)
def shape_to_list(shape):
return [dim.value for dim in shape]
def flatten(x):
with tf.name_scope('Flatten'):
return tf.reshape(x, [-1])
def log2(x):
with tf.name_scope('Log2'):
return tf.log(x) * np.float32(1.0 / np.log(2.0))
def exp2(x):
with tf.name_scope('Exp2'):
return tf.exp(x * np.float32(np.log(2.0)))
def lerp(a, b, t):
with tf.name_scope('Lerp'):
return a + (b - a) * t
def lerp_clip(a, b, t):
with tf.name_scope('LerpClip'):
return a + (b - a) * tf.clip_by_value(t, 0.0, 1.0)
def absolute_name_scope(scope): # Forcefully enter the specified name scope, ignoring any surrounding scopes.
return tf.name_scope(scope + '/')
# ----------------------------------------------------------------------------
# Initialize TensorFlow graph and session using good default settings.
def init_tf(config_dict=dict()):
if tf.get_default_session() is None:
tf.set_random_seed(np.random.randint(1 << 31))
create_session(config_dict, force_as_default=True)
# ----------------------------------------------------------------------------
# Create tf.Session based on config dict of the form
# {'gpu_options.allow_growth': True}
def create_session(config_dict=dict(), force_as_default=False):
config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=True)
for key, value in config_dict.items():
fields = key.split('.')
obj = config
for field in fields[:-1]:
obj = getattr(obj, field)
setattr(obj, fields[-1], value)
session = tf.Session(config=config)
if force_as_default:
session._default_session = session.as_default()
session._default_session.enforce_nesting = False
session._default_session.__enter__()
return session
# ----------------------------------------------------------------------------
# Initialize all tf.Variables that have not already been initialized.
# Equivalent to the following, but more efficient and does not bloat the tf graph:
# tf.variables_initializer(tf.report_unitialized_variables()).run()
def init_uninited_vars(vars=None):
if vars is None: vars = tf.global_variables()
test_vars = []
test_ops = []
with tf.control_dependencies(None): # ignore surrounding control_dependencies
for var in vars:
assert is_tf_expression(var)
try:
tf.get_default_graph().get_tensor_by_name(var.name.replace(':0', '/IsVariableInitialized:0'))
except KeyError:
# Op does not exist => variable may be uninitialized.
test_vars.append(var)
with absolute_name_scope(var.name.split(':')[0]):
test_ops.append(tf.is_variable_initialized(var))
init_vars = [var for var, inited in zip(test_vars, run(test_ops)) if not inited]
run([var.initializer for var in init_vars])
# ----------------------------------------------------------------------------
# Set the values of given tf.Variables.
# Equivalent to the following, but more efficient and does not bloat the tf graph:
# tfutil.run([tf.assign(var, value) for var, value in var_to_value_dict.items()]
def set_vars(var_to_value_dict):
ops = []
feed_dict = {}
for var, value in var_to_value_dict.items():
assert is_tf_expression(var)
try:
setter = tf.get_default_graph().get_tensor_by_name(
var.name.replace(':0', '/setter:0')) # look for existing op
except KeyError:
with absolute_name_scope(var.name.split(':')[0]):
with tf.control_dependencies(None): # ignore surrounding control_dependencies
setter = tf.assign(var, tf.placeholder(var.dtype, var.shape, 'new_value'),
name='setter') # create new setter
ops.append(setter)
feed_dict[setter.op.inputs[1]] = value
run(ops, feed_dict)
# ----------------------------------------------------------------------------
# Autosummary creates an identity op that internally keeps track of the input
# values and automatically shows up in TensorBoard. The reported value
# represents an average over input components. The average is accumulated
# constantly over time and flushed when save_summaries() is called.
#
# Notes:
# - The output tensor must be used as an input for something else in the
# graph. Otherwise, the autosummary op will not get executed, and the average
# value will not get accumulated.
# - It is perfectly fine to include autosummaries with the same name in
# several places throughout the graph, even if they are executed concurrently.
# - It is ok to also pass in a python scalar or numpy array. In this case, it
# is added to the average immediately.
_autosummary_vars = OrderedDict() # name => [var, ...]
_autosummary_immediate = OrderedDict() # name => update_op, update_value
_autosummary_finalized = False
def autosummary(name, value):
id = name.replace('/', '_')
if is_tf_expression(value):
with tf.name_scope('summary_' + id), tf.device(value.device):
update_op = _create_autosummary_var(name, value)
with tf.control_dependencies([update_op]):
return tf.identity(value)
else: # python scalar or numpy array
if name not in _autosummary_immediate:
with absolute_name_scope('Autosummary/' + id), tf.device(None), tf.control_dependencies(None):
update_value = tf.placeholder(tf.float32)
update_op = _create_autosummary_var(name, update_value)
_autosummary_immediate[name] = update_op, update_value
update_op, update_value = _autosummary_immediate[name]
run(update_op, {update_value: np.float32(value)})
return value
# Create the necessary ops to include autosummaries in TensorBoard report.
# Note: This should be done only once per graph.
def finalize_autosummaries():
global _autosummary_finalized
if _autosummary_finalized:
return
_autosummary_finalized = True
init_uninited_vars([var for vars in _autosummary_vars.values() for var in vars])
with tf.device(None), tf.control_dependencies(None):
for name, vars in _autosummary_vars.items():
id = name.replace('/', '_')
with absolute_name_scope('Autosummary/' + id):
sum = tf.add_n(vars)
avg = sum[0] / sum[1]
with tf.control_dependencies([avg]): # read before resetting
reset_ops = [tf.assign(var, tf.zeros(2)) for var in vars]
with tf.name_scope(None), tf.control_dependencies(reset_ops): # reset before reporting
tf.summary.scalar(name, avg)
# Internal helper for creating autosummary accumulators.
def _create_autosummary_var(name, value_expr):
assert not _autosummary_finalized
v = tf.cast(value_expr, tf.float32)
if v.shape.ndims is 0:
v = [v, np.float32(1.0)]
elif v.shape.ndims is 1:
v = [tf.reduce_sum(v), tf.cast(tf.shape(v)[0], tf.float32)]
else:
v = [tf.reduce_sum(v), tf.reduce_prod(tf.cast(tf.shape(v), tf.float32))]
v = tf.cond(tf.is_finite(v[0]), lambda: tf.stack(v), lambda: tf.zeros(2))
with tf.control_dependencies(None):
var = tf.Variable(tf.zeros(2)) # [numerator, denominator]
update_op = tf.cond(tf.is_variable_initialized(var), lambda: tf.assign_add(var, v), lambda: tf.assign(var, v))
if name in _autosummary_vars:
_autosummary_vars[name].append(var)
else:
_autosummary_vars[name] = [var]
return update_op
# ----------------------------------------------------------------------------
# Call filewriter.add_summary() with all summaries in the default graph,
# automatically finalizing and merging them on the first call.
_summary_merge_op = None
def save_summaries(filewriter, global_step=None):
global _summary_merge_op
if _summary_merge_op is None:
finalize_autosummaries()
with tf.device(None), tf.control_dependencies(None):
_summary_merge_op = tf.summary.merge_all()
filewriter.add_summary(_summary_merge_op.eval(), global_step)
# ----------------------------------------------------------------------------
# Utilities for importing modules and objects by name.
def import_module(module_or_obj_name):
parts = module_or_obj_name.split('.')
parts[0] = {'np': 'numpy', 'tf': 'tensorflow'}.get(parts[0], parts[0])
for i in range(len(parts), 0, -1):
try:
module = importlib.import_module('.'.join(parts[:i]))
relative_obj_name = '.'.join(parts[i:])
return module, relative_obj_name
except ImportError:
pass
raise ImportError(module_or_obj_name)
def find_obj_in_module(module, relative_obj_name):
obj = module
for part in relative_obj_name.split('.'):
obj = getattr(obj, part)
return obj
def import_obj(obj_name):
module, relative_obj_name = import_module(obj_name)
return find_obj_in_module(module, relative_obj_name)
def call_func_by_name(*args, func=None, **kwargs):
assert func is not None
return import_obj(func)(*args, **kwargs)
# ----------------------------------------------------------------------------
# Wrapper for tf.train.Optimizer that automatically takes care of:
# - Gradient averaging for multi-GPU training.
# - Dynamic loss scaling and typecasts for FP16 training.
# - Ignoring corrupted gradients that contain NaNs/Infs.
# - Reporting statistics.
# - Well-chosen default settings.
class Optimizer:
def __init__(
self,
name='Train',
tf_optimizer='tf.train.AdamOptimizer',
learning_rate=0.001,
use_loss_scaling=False,
loss_scaling_init=64.0,
loss_scaling_inc=0.0005,
loss_scaling_dec=1.0,
**kwargs):
# Init fields.
self.name = name
self.learning_rate = tf.convert_to_tensor(learning_rate)
self.id = self.name.replace('/', '.')
self.scope = tf.get_default_graph().unique_name(self.id)
self.optimizer_class = import_obj(tf_optimizer)
self.optimizer_kwargs = dict(kwargs)
self.use_loss_scaling = use_loss_scaling
self.loss_scaling_init = loss_scaling_init
self.loss_scaling_inc = loss_scaling_inc
self.loss_scaling_dec = loss_scaling_dec
self._grad_shapes = None # [shape, ...]
self._dev_opt = OrderedDict() # device => optimizer
self._dev_grads = OrderedDict() # device => [[(grad, var), ...], ...]
self._dev_ls_var = OrderedDict() # device => variable (log2 of loss scaling factor)
self._updates_applied = False
# Register the gradients of the given loss function with respect to the given variables.
# Intended to be called once per GPU.
def register_gradients(self, loss, vars):
assert not self._updates_applied
# Validate arguments.
if isinstance(vars, dict):
vars = list(vars.values()) # allow passing in Network.trainables as vars
assert isinstance(vars, list) and len(vars) >= 1
assert all(is_tf_expression(expr) for expr in vars + [loss])
if self._grad_shapes is None:
self._grad_shapes = [shape_to_list(var.shape) for var in vars]
assert len(vars) == len(self._grad_shapes)
assert all(shape_to_list(var.shape) == var_shape for var, var_shape in zip(vars, self._grad_shapes))
dev = loss.device
assert all(var.device == dev for var in vars)
# Register device and compute gradients.
with tf.name_scope(self.id + '_grad'), tf.device(dev):
if dev not in self._dev_opt:
opt_name = self.scope.replace('/', '_') + '_opt%d' % len(self._dev_opt)
self._dev_opt[dev] = self.optimizer_class(name=opt_name, learning_rate=self.learning_rate,
**self.optimizer_kwargs)
self._dev_grads[dev] = []
loss = self.apply_loss_scaling(tf.cast(loss, tf.float32))
grads = self._dev_opt[dev].compute_gradients(loss, vars,
gate_gradients=tf.train.Optimizer.GATE_NONE) # disable gating to reduce memory usage
grads = [(g, v) if g is not None else (tf.zeros_like(v), v) for g, v in
grads] # replace disconnected gradients with zeros
self._dev_grads[dev].append(grads)
# Construct training op to update the registered variables based on their gradients.
def apply_updates(self):
assert not self._updates_applied
self._updates_applied = True
devices = list(self._dev_grads.keys())
total_grads = sum(len(grads) for grads in self._dev_grads.values())
assert len(devices) >= 1 and total_grads >= 1
ops = []
with absolute_name_scope(self.scope):
# Cast gradients to FP32 and calculate partial sum within each device.
dev_grads = OrderedDict() # device => [(grad, var), ...]
for dev_idx, dev in enumerate(devices):
with tf.name_scope('ProcessGrads%d' % dev_idx), tf.device(dev):
sums = []
for gv in zip(*self._dev_grads[dev]):
assert all(v is gv[0][1] for g, v in gv)
g = [tf.cast(g, tf.float32) for g, v in gv]
g = g[0] if len(g) == 1 else tf.add_n(g)
sums.append((g, gv[0][1]))
dev_grads[dev] = sums
# Sum gradients across devices.
if len(devices) > 1:
with tf.name_scope('SumAcrossGPUs'), tf.device(None):
for var_idx, grad_shape in enumerate(self._grad_shapes):
g = [dev_grads[dev][var_idx][0] for dev in devices]
if np.prod(grad_shape): # nccl does not support zero-sized tensors
g = tf.contrib.nccl.all_sum(g)
for dev, gg in zip(devices, g):
dev_grads[dev][var_idx] = (gg, dev_grads[dev][var_idx][1])
# Apply updates separately on each device.
for dev_idx, (dev, grads) in enumerate(dev_grads.items()):
with tf.name_scope('ApplyGrads%d' % dev_idx), tf.device(dev):
# Scale gradients as needed.
if self.use_loss_scaling or total_grads > 1:
with tf.name_scope('Scale'):
coef = tf.constant(np.float32(1.0 / total_grads), name='coef')
coef = self.undo_loss_scaling(coef)
grads = [(g * coef, v) for g, v in grads]
# Check for overflows.
with tf.name_scope('CheckOverflow'):
grad_ok = tf.reduce_all(tf.stack([tf.reduce_all(tf.is_finite(g)) for g, v in grads]))
# Update weights and adjust loss scaling.
with tf.name_scope('UpdateWeights'):
opt = self._dev_opt[dev]
ls_var = self.get_loss_scaling_var(dev)
if not self.use_loss_scaling:
ops.append(tf.cond(grad_ok, lambda: opt.apply_gradients(grads), tf.no_op))
else:
ops.append(tf.cond(grad_ok,
lambda: tf.group(tf.assign_add(ls_var, self.loss_scaling_inc),
opt.apply_gradients(grads)),
lambda: tf.group(tf.assign_sub(ls_var, self.loss_scaling_dec))))
# Report statistics on the last device.
if dev == devices[-1]:
with tf.name_scope('Statistics'):
ops.append(autosummary(self.id + '/learning_rate', self.learning_rate))
ops.append(autosummary(self.id + '/overflow_frequency', tf.where(grad_ok, 0, 1)))
if self.use_loss_scaling:
ops.append(autosummary(self.id + '/loss_scaling_log2', ls_var))
# Initialize variables and group everything into a single op.
self.reset_optimizer_state()
init_uninited_vars(list(self._dev_ls_var.values()))
return tf.group(*ops, name='TrainingOp')
# Reset internal state of the underlying optimizer.
def reset_optimizer_state(self):
run([var.initializer for opt in self._dev_opt.values() for var in opt.variables()])
# Get or create variable representing log2 of the current dynamic loss scaling factor.
def get_loss_scaling_var(self, device):
if not self.use_loss_scaling:
return None
if device not in self._dev_ls_var:
with absolute_name_scope(self.scope + '/LossScalingVars'), tf.control_dependencies(None):
self._dev_ls_var[device] = tf.Variable(np.float32(self.loss_scaling_init), name='loss_scaling_var')
return self._dev_ls_var[device]
# Apply dynamic loss scaling for the given expression.
def apply_loss_scaling(self, value):
assert is_tf_expression(value)
if not self.use_loss_scaling:
return value
return value * exp2(self.get_loss_scaling_var(value.device))
# Undo the effect of dynamic loss scaling for the given expression.
def undo_loss_scaling(self, value):
assert is_tf_expression(value)
if not self.use_loss_scaling:
return value
return value * exp2(-self.get_loss_scaling_var(value.device))
# ----------------------------------------------------------------------------
# Generic network abstraction.
#
# Acts as a convenience wrapper for a parameterized network construction
# function, providing several utility methods and convenient access to
# the inputs/outputs/weights.
#
# Network objects can be safely pickled and unpickled for long-term
# archival purposes. The pickling works reliably as long as the underlying
# network construction function is defined in a standalone Python module
# that has no side effects or application-specific imports.
network_import_handlers = [] # Custom import handlers for dealing with legacy data in pickle import.
_network_import_modules = [] # Temporary modules create during pickle import.
class Network:
def __init__(self,
name=None, # Network name. Used to select TensorFlow name and variable scopes.
func=None, # Fully qualified name of the underlying network construction function.
**static_kwargs): # Keyword arguments to be passed in to the network construction function.
self._init_fields()
self.name = name
self.static_kwargs = dict(static_kwargs)
# Init build func.
module, self._build_func_name = import_module(func)
self._build_module_src = inspect.getsource(module)
self._build_func = find_obj_in_module(module, self._build_func_name)
# Init graph.
self._init_graph()
self.reset_vars()
def _init_fields(self):
self.name = None # User-specified name, defaults to build func name if None.
self.scope = None # Unique TF graph scope, derived from the user-specified name.
self.static_kwargs = dict() # Arguments passed to the user-supplied build func.
self.num_inputs = 0 # Number of input tensors.
self.num_outputs = 0 # Number of output tensors.
self.input_shapes = [[]] # Input tensor shapes (NC or NCHW), including minibatch dimension.
self.output_shapes = [[]] # Output tensor shapes (NC or NCHW), including minibatch dimension.
self.input_shape = [] # Short-hand for input_shapes[0].
self.output_shape = [] # Short-hand for output_shapes[0].
self.input_templates = [] # Input placeholders in the template graph.
self.output_templates = [] # Output tensors in the template graph.
self.input_names = [] # Name string for each input.
self.output_names = [] # Name string for each output.
self.vars = OrderedDict() # All variables (localname => var).
self.trainables = OrderedDict() # Trainable variables (localname => var).
self._build_func = None # User-supplied build function that constructs the network.
self._build_func_name = None # Name of the build function.
self._build_module_src = None # Full source code of the module containing the build function.
self._run_cache = dict() # Cached graph data for Network.run().
def _init_graph(self):
# Collect inputs.
self.input_names = []
for param in inspect.signature(self._build_func).parameters.values():
if param.kind == param.POSITIONAL_OR_KEYWORD and param.default is param.empty:
self.input_names.append(param.name)
self.num_inputs = len(self.input_names)
assert self.num_inputs >= 1
# Choose name and scope.
if self.name is None:
self.name = self._build_func_name
self.scope = tf.get_default_graph().unique_name(self.name.replace('/', '_'), mark_as_used=False)
# Build template graph.
with tf.variable_scope(self.scope, reuse=tf.AUTO_REUSE):
assert tf.get_variable_scope().name == self.scope
with absolute_name_scope(self.scope): # ignore surrounding name_scope
with tf.control_dependencies(None): # ignore surrounding control_dependencies
self.input_templates = [tf.placeholder(tf.float32, name=name) for name in self.input_names]
out_expr = self._build_func(*self.input_templates, is_template_graph=True, **self.static_kwargs)
# Collect outputs.
assert is_tf_expression(out_expr) or isinstance(out_expr, tuple)
self.output_templates = [out_expr] if is_tf_expression(out_expr) else list(out_expr)
self.output_names = [t.name.split('/')[-1].split(':')[0] for t in self.output_templates]
self.num_outputs = len(self.output_templates)
assert self.num_outputs >= 1
# Populate remaining fields.
self.input_shapes = [shape_to_list(t.shape) for t in self.input_templates]
self.output_shapes = [shape_to_list(t.shape) for t in self.output_templates]
self.input_shape = self.input_shapes[0]
self.output_shape = self.output_shapes[0]
self.vars = OrderedDict([(self.get_var_localname(var), var) for var in tf.global_variables(self.scope + '/')])
self.trainables = OrderedDict(
[(self.get_var_localname(var), var) for var in tf.trainable_variables(self.scope + '/')])
# Run initializers for all variables defined by this network.
def reset_vars(self):
run([var.initializer for var in self.vars.values()])
# Run initializers for all trainable variables defined by this network.
def reset_trainables(self):
run([var.initializer for var in self.trainables.values()])
# Get TensorFlow expression(s) for the output(s) of this network, given the inputs.
def get_output_for(self, *in_expr, return_as_list=False, **dynamic_kwargs):
assert len(in_expr) == self.num_inputs
all_kwargs = dict(self.static_kwargs)
all_kwargs.update(dynamic_kwargs)
with tf.variable_scope(self.scope, reuse=True):
assert tf.get_variable_scope().name == self.scope
named_inputs = [tf.identity(expr, name=name) for expr, name in zip(in_expr, self.input_names)]
out_expr = self._build_func(*named_inputs, **all_kwargs)
assert is_tf_expression(out_expr) or isinstance(out_expr, tuple)
if return_as_list:
out_expr = [out_expr] if is_tf_expression(out_expr) else list(out_expr)
return out_expr
# Get the local name of a given variable, excluding any surrounding name scopes.
def get_var_localname(self, var_or_globalname):
assert is_tf_expression(var_or_globalname) or isinstance(var_or_globalname, str)
globalname = var_or_globalname if isinstance(var_or_globalname, str) else var_or_globalname.name
assert globalname.startswith(self.scope + '/')
localname = globalname[len(self.scope) + 1:]
localname = localname.split(':')[0]
return localname
# Find variable by local or global name.
def find_var(self, var_or_localname):
assert is_tf_expression(var_or_localname) or isinstance(var_or_localname, str)
return self.vars[var_or_localname] if isinstance(var_or_localname, str) else var_or_localname
# Get the value of a given variable as NumPy array.
# Note: This method is very inefficient -- prefer to use tfutil.run(list_of_vars) whenever possible.
def get_var(self, var_or_localname):
return self.find_var(var_or_localname).eval()
# Set the value of a given variable based on the given NumPy array.
# Note: This method is very inefficient -- prefer to use tfutil.set_vars() whenever possible.
def set_var(self, var_or_localname, new_value):
return set_vars({self.find_var(var_or_localname): new_value})