forked from Guanghan/ROLO
-
Notifications
You must be signed in to change notification settings - Fork 0
/
rnn.py
1325 lines (1110 loc) · 55.4 KB
/
rnn.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 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""RNN helpers for TensorFlow models."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import rnn_cell
from tensorflow.python.ops import tensor_array_ops
from tensorflow.python.ops import variable_scope as vs
from tensorflow.python.util import nest
# pylint: disable=protected-access
_state_size_with_prefix = rnn_cell._state_size_with_prefix
# pylint: enable=protected-access
def _infer_state_dtype(explicit_dtype, state):
"""Infer the dtype of an RNN state.
Args:
explicit_dtype: explicitly declared dtype or None.
state: RNN's hidden state. Must be a Tensor or a nested iterable containing
Tensors.
Returns:
dtype: inferred dtype of hidden state.
Raises:
ValueError: if `state` has heterogeneous dtypes or is empty.
"""
if explicit_dtype is not None:
return explicit_dtype
elif nest.is_sequence(state):
inferred_dtypes = [element.dtype for element in nest.flatten(state)]
if not inferred_dtypes:
raise ValueError("Unable to infer dtype from empty state.")
all_same = all([x == inferred_dtypes[0] for x in inferred_dtypes])
if not all_same:
raise ValueError(
"State has tensors of different inferred_dtypes. Unable to infer a "
"single representative dtype.")
return inferred_dtypes[0]
else:
return state.dtype
def rnn(cell, inputs, initial_state=None, dtype=None,
sequence_length=None, scope=None):
"""Creates a recurrent neural network specified by RNNCell `cell`.
The simplest form of RNN network generated is:
```python
state = cell.zero_state(...)
outputs = []
for input_ in inputs:
output, state = cell(input_, state)
outputs.append(output)
return (outputs, state)
```
However, a few other options are available:
An initial state can be provided.
If the sequence_length vector is provided, dynamic calculation is performed.
This method of calculation does not compute the RNN steps past the maximum
sequence length of the minibatch (thus saving computational time),
and properly propagates the state at an example's sequence length
to the final state output.
The dynamic calculation performed is, at time `t` for batch row `b`,
```python
(output, state)(b, t) =
(t >= sequence_length(b))
? (zeros(cell.output_size), states(b, sequence_length(b) - 1))
: cell(input(b, t), state(b, t - 1))
```
Args:
cell: An instance of RNNCell.
inputs: A length T list of inputs, each a `Tensor` of shape
`[batch_size, input_size]`, or a nested tuple of such elements.
initial_state: (optional) An initial state for the RNN.
If `cell.state_size` is an integer, this must be
a `Tensor` of appropriate type and shape `[batch_size, cell.state_size]`.
If `cell.state_size` is a tuple, this should be a tuple of
tensors having shapes `[batch_size, s] for s in cell.state_size`.
dtype: (optional) The data type for the initial state and expected output.
Required if initial_state is not provided or RNN state has a heterogeneous
dtype.
sequence_length: Specifies the length of each sequence in inputs.
An int32 or int64 vector (tensor) size `[batch_size]`, values in `[0, T)`.
scope: VariableScope for the created subgraph; defaults to "RNN".
Returns:
A pair (outputs, state) where:
- outputs is a length T list of outputs (one for each input), or a nested
tuple of such elements.
- state is the final state
Raises:
TypeError: If `cell` is not an instance of RNNCell.
ValueError: If `inputs` is `None` or an empty list, or if the input depth
(column size) cannot be inferred from inputs via shape inference.
"""
if not isinstance(cell, rnn_cell.RNNCell):
raise TypeError("cell must be an instance of RNNCell")
if not nest.is_sequence(inputs):
raise TypeError("inputs must be a sequence")
if not inputs:
raise ValueError("inputs must not be empty")
outputs = []
# Create a new scope in which the caching device is either
# determined by the parent scope, or is set to place the cached
# Variable using the same placement as for the rest of the RNN.
with vs.variable_scope(scope or "RNN") as varscope:
if varscope.caching_device is None:
varscope.set_caching_device(lambda op: op.device)
# Obtain the first sequence of the input
first_input = inputs
while nest.is_sequence(first_input):
first_input = first_input[0]
# Temporarily avoid EmbeddingWrapper and seq2seq badness
# TODO(lukaszkaiser): remove EmbeddingWrapper
if first_input.get_shape().ndims != 1:
input_shape = first_input.get_shape().with_rank_at_least(2)
fixed_batch_size = input_shape[0]
flat_inputs = nest.flatten(inputs)
for flat_input in flat_inputs:
input_shape = flat_input.get_shape().with_rank_at_least(2)
batch_size, input_size = input_shape[0], input_shape[1:]
fixed_batch_size.merge_with(batch_size)
for i, size in enumerate(input_size):
if size.value is None:
raise ValueError(
"Input size (dimension %d of inputs) must be accessible via "
"shape inference, but saw value None." % i)
else:
fixed_batch_size = first_input.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
batch_size = fixed_batch_size.value
else:
batch_size = array_ops.shape(first_input)[0]
if initial_state is not None:
state = initial_state
else:
if not dtype:
raise ValueError("If no initial_state is provided, "
"dtype must be specified")
state = cell.zero_state(batch_size, dtype)
if sequence_length is not None: # Prepare variables
def _create_zero_output(output_size):
# convert int to TensorShape if necessary
size = _state_size_with_prefix(output_size, prefix=[batch_size])
output = array_ops.zeros(
array_ops.pack(size), _infer_state_dtype(dtype, state))
shape = _state_size_with_prefix(
output_size, prefix=[fixed_batch_size.value])
output.set_shape(tensor_shape.TensorShape(shape))
return output
output_size = cell.output_size
flat_output_size = nest.flatten(output_size)
flat_zero_output = tuple(
_create_zero_output(size) for size in flat_output_size)
zero_output = nest.pack_sequence_as(structure=output_size,
flat_sequence=flat_zero_output)
sequence_length = math_ops.to_int32(sequence_length)
min_sequence_length = math_ops.reduce_min(sequence_length)
max_sequence_length = math_ops.reduce_max(sequence_length)
for time, input_ in enumerate(inputs):
if time > 0: varscope.reuse_variables()
# pylint: disable=cell-var-from-loop
call_cell = lambda: cell(input_, state)
# pylint: enable=cell-var-from-loop
if sequence_length is not None:
(output, state) = _rnn_step(
time=time,
sequence_length=sequence_length,
min_sequence_length=min_sequence_length,
max_sequence_length=max_sequence_length,
zero_output=zero_output,
state=state,
call_cell=call_cell,
state_size=cell.state_size)
else:
(output, state) = call_cell()
outputs.append(output)
return (outputs, state)
def state_saving_rnn(cell, inputs, state_saver, state_name,
sequence_length=None, scope=None):
"""RNN that accepts a state saver for time-truncated RNN calculation.
Args:
cell: An instance of `RNNCell`.
inputs: A length T list of inputs, each a `Tensor` of shape
`[batch_size, input_size]`.
state_saver: A state saver object with methods `state` and `save_state`.
state_name: Python string or tuple of strings. The name to use with the
state_saver. If the cell returns tuples of states (i.e.,
`cell.state_size` is a tuple) then `state_name` should be a tuple of
strings having the same length as `cell.state_size`. Otherwise it should
be a single string.
sequence_length: (optional) An int32/int64 vector size [batch_size].
See the documentation for rnn() for more details about sequence_length.
scope: VariableScope for the created subgraph; defaults to "RNN".
Returns:
A pair (outputs, state) where:
outputs is a length T list of outputs (one for each input)
states is the final state
Raises:
TypeError: If `cell` is not an instance of RNNCell.
ValueError: If `inputs` is `None` or an empty list, or if the arity and
type of `state_name` does not match that of `cell.state_size`.
"""
state_size = cell.state_size
state_is_tuple = nest.is_sequence(state_size)
state_name_tuple = nest.is_sequence(state_name)
if state_is_tuple != state_name_tuple:
raise ValueError(
"state_name should be the same type as cell.state_size. "
"state_name: %s, cell.state_size: %s"
% (str(state_name), str(state_size)))
if state_is_tuple:
state_name_flat = nest.flatten(state_name)
state_size_flat = nest.flatten(state_size)
if len(state_name_flat) != len(state_size_flat):
raise ValueError("#elems(state_name) != #elems(state_size): %d vs. %d"
% (len(state_name_flat), len(state_size_flat)))
initial_state = nest.pack_sequence_as(
structure=state_size,
flat_sequence=[state_saver.state(s) for s in state_name_flat])
else:
initial_state = state_saver.state(state_name)
(outputs, state) = rnn(cell, inputs, initial_state=initial_state,
sequence_length=sequence_length, scope=scope)
if state_is_tuple:
flat_state = nest.flatten(state)
state_name = nest.flatten(state_name)
save_state = [state_saver.save_state(name, substate)
for name, substate in zip(state_name, flat_state)]
else:
save_state = [state_saver.save_state(state_name, state)]
with ops.control_dependencies(save_state):
last_output = outputs[-1]
flat_last_output = nest.flatten(last_output)
flat_last_output = [
array_ops.identity(output) for output in flat_last_output]
outputs[-1] = nest.pack_sequence_as(structure=last_output,
flat_sequence=flat_last_output)
return (outputs, state)
# pylint: disable=unused-argument
def _rnn_step(
time, sequence_length, min_sequence_length, max_sequence_length,
zero_output, state, call_cell, state_size, skip_conditionals=False):
"""Calculate one step of a dynamic RNN minibatch.
Returns an (output, state) pair conditioned on the sequence_lengths.
When skip_conditionals=False, the pseudocode is something like:
if t >= max_sequence_length:
return (zero_output, state)
if t < min_sequence_length:
return call_cell()
# Selectively output zeros or output, old state or new state depending
# on if we've finished calculating each row.
new_output, new_state = call_cell()
final_output = np.vstack([
zero_output if time >= sequence_lengths[r] else new_output_r
for r, new_output_r in enumerate(new_output)
])
final_state = np.vstack([
state[r] if time >= sequence_lengths[r] else new_state_r
for r, new_state_r in enumerate(new_state)
])
return (final_output, final_state)
Args:
time: Python int, the current time step
sequence_length: int32 `Tensor` vector of size [batch_size]
min_sequence_length: int32 `Tensor` scalar, min of sequence_length
max_sequence_length: int32 `Tensor` scalar, max of sequence_length
zero_output: `Tensor` vector of shape [output_size]
state: Either a single `Tensor` matrix of shape `[batch_size, state_size]`,
or a list/tuple of such tensors.
call_cell: lambda returning tuple of (new_output, new_state) where
new_output is a `Tensor` matrix of shape `[batch_size, output_size]`.
new_state is a `Tensor` matrix of shape `[batch_size, state_size]`.
state_size: The `cell.state_size` associated with the state.
skip_conditionals: Python bool, whether to skip using the conditional
calculations. This is useful for `dynamic_rnn`, where the input tensor
matches `max_sequence_length`, and using conditionals just slows
everything down.
Returns:
A tuple of (`final_output`, `final_state`) as given by the pseudocode above:
final_output is a `Tensor` matrix of shape [batch_size, output_size]
final_state is either a single `Tensor` matrix, or a tuple of such
matrices (matching length and shapes of input `state`).
Raises:
ValueError: If the cell returns a state tuple whose length does not match
that returned by `state_size`.
"""
# Convert state to a list for ease of use
flat_state = nest.flatten(state)
flat_zero_output = nest.flatten(zero_output)
def _copy_one_through(output, new_output):
copy_cond = (time >= sequence_length)
return math_ops.select(copy_cond, output, new_output)
def _copy_some_through(flat_new_output, flat_new_state):
# Use broadcasting select to determine which values should get
# the previous state & zero output, and which values should get
# a calculated state & output.
flat_new_output = [
_copy_one_through(zero_output, new_output)
for zero_output, new_output in zip(flat_zero_output, flat_new_output)]
flat_new_state = [
_copy_one_through(state, new_state)
for state, new_state in zip(flat_state, flat_new_state)]
return flat_new_output + flat_new_state
def _maybe_copy_some_through():
"""Run RNN step. Pass through either no or some past state."""
new_output, new_state = call_cell()
nest.assert_same_structure(state, new_state)
flat_new_state = nest.flatten(new_state)
flat_new_output = nest.flatten(new_output)
return control_flow_ops.cond(
# if t < min_seq_len: calculate and return everything
time < min_sequence_length, lambda: flat_new_output + flat_new_state,
# else copy some of it through
lambda: _copy_some_through(flat_new_output, flat_new_state))
# TODO(ebrevdo): skipping these conditionals may cause a slowdown,
# but benefits from removing cond() and its gradient. We should
# profile with and without this switch here.
if skip_conditionals:
# Instead of using conditionals, perform the selective copy at all time
# steps. This is faster when max_seq_len is equal to the number of unrolls
# (which is typical for dynamic_rnn).
new_output, new_state = call_cell()
nest.assert_same_structure(state, new_state)
new_state = nest.flatten(new_state)
new_output = nest.flatten(new_output)
final_output_and_state = _copy_some_through(new_output, new_state)
else:
empty_update = lambda: flat_zero_output + flat_state
final_output_and_state = control_flow_ops.cond(
# if t >= max_seq_len: copy all state through, output zeros
time >= max_sequence_length, empty_update,
# otherwise calculation is required: copy some or all of it through
_maybe_copy_some_through)
if len(final_output_and_state) != len(flat_zero_output) + len(flat_state):
raise ValueError("Internal error: state and output were not concatenated "
"correctly.")
final_output = final_output_and_state[:len(flat_zero_output)]
final_state = final_output_and_state[len(flat_zero_output):]
for output, flat_output in zip(final_output, flat_zero_output):
output.set_shape(flat_output.get_shape())
for substate, flat_substate in zip(final_state, flat_state):
substate.set_shape(flat_substate.get_shape())
final_output = nest.pack_sequence_as(
structure=zero_output, flat_sequence=final_output)
final_state = nest.pack_sequence_as(
structure=state, flat_sequence=final_state)
return final_output, final_state
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, n_features)
or nested tuples of tensors.
lengths: A `Tensor` of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
flat_input_seq = tuple(nest.flatten(input_) for input_ in input_seq)
flat_results = [[] for _ in range(len(input_seq))]
for sequence in zip(*flat_input_seq):
input_shape = tensor_shape.unknown_shape(
ndims=sequence[0].get_shape().ndims)
for input_ in sequence:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(sequence)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r, flat_result in zip(result, flat_results):
r.set_shape(input_shape)
flat_result.append(r)
results = [nest.pack_sequence_as(structure=input_, flat_sequence=flat_result)
for input_, flat_result in zip(input_seq, flat_results)]
return results
def bidirectional_rnn(cell_fw, cell_bw, inputs,
initial_state_fw=None, initial_state_bw=None,
dtype=None, sequence_length=None, scope=None):
"""Creates a bidirectional recurrent neural network.
Similar to the unidirectional case above (rnn) but takes input and builds
independent forward and backward RNNs with the final forward and backward
outputs depth-concatenated, such that the output will have the format
[time][batch][cell_fw.output_size + cell_bw.output_size]. The input_size of
forward and backward cell must match. The initial state for both directions
is zero by default (but can be set optionally) and no intermediate states are
ever returned -- the network is fully unrolled for the given (passed in)
length(s) of the sequence(s) or completely unrolled if length(s) is not given.
Args:
cell_fw: An instance of RNNCell, to be used for forward direction.
cell_bw: An instance of RNNCell, to be used for backward direction.
inputs: A length T list of inputs, each a tensor of shape
[batch_size, input_size], or a nested tuple of such elements.
initial_state_fw: (optional) An initial state for the forward RNN.
This must be a tensor of appropriate type and shape
`[batch_size, cell_fw.state_size]`.
If `cell_fw.state_size` is a tuple, this should be a tuple of
tensors having shapes `[batch_size, s] for s in cell_fw.state_size`.
initial_state_bw: (optional) Same as for `initial_state_fw`, but using
the corresponding properties of `cell_bw`.
dtype: (optional) The data type for the initial state. Required if
either of the initial states are not provided.
sequence_length: (optional) An int32/int64 vector, size `[batch_size]`,
containing the actual lengths for each of the sequences.
scope: VariableScope for the created subgraph; defaults to "BiRNN"
Returns:
A tuple (outputs, output_state_fw, output_state_bw) where:
outputs is a length `T` list of outputs (one for each input), which
are depth-concatenated forward and backward outputs.
output_state_fw is the final state of the forward rnn.
output_state_bw is the final state of the backward rnn.
Raises:
TypeError: If `cell_fw` or `cell_bw` is not an instance of `RNNCell`.
ValueError: If inputs is None or an empty list.
"""
if not isinstance(cell_fw, rnn_cell.RNNCell):
raise TypeError("cell_fw must be an instance of RNNCell")
if not isinstance(cell_bw, rnn_cell.RNNCell):
raise TypeError("cell_bw must be an instance of RNNCell")
if not nest.is_sequence(inputs):
raise TypeError("inputs must be a sequence")
if not inputs:
raise ValueError("inputs must not be empty")
with vs.variable_scope(scope or "BiRNN"):
# Forward direction
with vs.variable_scope("FW") as fw_scope:
output_fw, output_state_fw = rnn(cell_fw, inputs, initial_state_fw, dtype,
sequence_length, scope=fw_scope)
# Backward direction
with vs.variable_scope("BW") as bw_scope:
reversed_inputs = _reverse_seq(inputs, sequence_length)
tmp, output_state_bw = rnn(cell_bw, reversed_inputs, initial_state_bw,
dtype, sequence_length, scope=bw_scope)
output_bw = _reverse_seq(tmp, sequence_length)
# Concat each of the forward/backward outputs
flat_output_fw = nest.flatten(output_fw)
flat_output_bw = nest.flatten(output_bw)
flat_outputs = tuple(array_ops.concat(1, [fw, bw])
for fw, bw in zip(flat_output_fw, flat_output_bw))
outputs = nest.pack_sequence_as(structure=output_fw,
flat_sequence=flat_outputs)
return (outputs, output_state_fw, output_state_bw)
def bidirectional_dynamic_rnn(cell_fw, cell_bw, inputs, sequence_length=None,
initial_state_fw=None, initial_state_bw=None,
dtype=None, parallel_iterations=None,
swap_memory=False, time_major=False, scope=None):
"""Creates a dynamic version of bidirectional recurrent neural network.
Similar to the unidirectional case above (rnn) but takes input and builds
independent forward and backward RNNs. The input_size of forward and
backward cell must match. The initial state for both directions is zero by
default (but can be set optionally) and no intermediate states are ever
returned -- the network is fully unrolled for the given (passed in)
length(s) of the sequence(s) or completely unrolled if length(s) is not
given.
Args:
cell_fw: An instance of RNNCell, to be used for forward direction.
cell_bw: An instance of RNNCell, to be used for backward direction.
inputs: The RNN inputs.
If time_major == False (default), this must be a tensor of shape:
`[batch_size, max_time, input_size]`.
If time_major == True, this must be a tensor of shape:
`[max_time, batch_size, input_size]`.
[batch_size, input_size].
sequence_length: An int32/int64 vector, size `[batch_size]`,
containing the actual lengths for each of the sequences.
initial_state_fw: (optional) An initial state for the forward RNN.
This must be a tensor of appropriate type and shape
`[batch_size, cell_fw.state_size]`.
If `cell_fw.state_size` is a tuple, this should be a tuple of
tensors having shapes `[batch_size, s] for s in cell_fw.state_size`.
initial_state_bw: (optional) Same as for `initial_state_fw`, but using
the corresponding properties of `cell_bw`.
dtype: (optional) The data type for the initial states and expected output.
Required if initial_states are not provided or RNN states have a
heterogeneous dtype.
parallel_iterations: (Default: 32). The number of iterations to run in
parallel. Those operations which do not have any temporal dependency
and can be run in parallel, will be. This parameter trades off
time for space. Values >> 1 use more memory but take less time,
while smaller values use less memory but computations take longer.
swap_memory: Transparently swap the tensors produced in forward inference
but needed for back prop from GPU to CPU. This allows training RNNs
which would typically not fit on a single GPU, with very minimal (or no)
performance penalty.
time_major: The shape format of the `inputs` and `outputs` Tensors.
If true, these `Tensors` must be shaped `[max_time, batch_size, depth]`.
If false, these `Tensors` must be shaped `[batch_size, max_time, depth]`.
Using `time_major = True` is a bit more efficient because it avoids
transposes at the beginning and end of the RNN calculation. However,
most TensorFlow data is batch-major, so by default this function
accepts input and emits output in batch-major form.
dtype: (optional) The data type for the initial state. Required if
initial_state is not provided.
sequence_length: An int32/int64 vector, size `[batch_size]`,
containing the actual lengths for each of the sequences.
either of the initial states are not provided.
scope: VariableScope for the created subgraph; defaults to "BiRNN"
Returns:
A tuple (outputs, output_states) where:
outputs: A tuple (output_fw, output_bw) containing the forward and
the backward rnn output `Tensor`.
If time_major == False (default),
output_fw will be a `Tensor` shaped:
`[batch_size, max_time, cell_fw.output_size]`
and output_bw will be a `Tensor` shaped:
`[batch_size, max_time, cell_bw.output_size]`.
If time_major == True,
output_fw will be a `Tensor` shaped:
`[max_time, batch_size, cell_fw.output_size]`
and output_bw will be a `Tensor` shaped:
`[max_time, batch_size, cell_bw.output_size]`.
It returns a tuple instead of a single concatenated `Tensor`, unlike
in the `bidirectional_rnn`. If the concatenated one is preferred,
the forward and backward outputs can be concatenated as
`tf.concat(2, outputs)`.
output_states: A tuple (output_state_fw, output_state_bw) containing
the forward and the backward final states of bidirectional rnn.
Raises:
TypeError: If `cell_fw` or `cell_bw` is not an instance of `RNNCell`.
"""
if not isinstance(cell_fw, rnn_cell.RNNCell):
raise TypeError("cell_fw must be an instance of RNNCell")
if not isinstance(cell_bw, rnn_cell.RNNCell):
raise TypeError("cell_bw must be an instance of RNNCell")
with vs.variable_scope(scope or "BiRNN"):
# Forward direction
with vs.variable_scope("FW") as fw_scope:
output_fw, output_state_fw = dynamic_rnn(
cell=cell_fw, inputs=inputs, sequence_length=sequence_length,
initial_state=initial_state_fw, dtype=dtype,
parallel_iterations=parallel_iterations, swap_memory=swap_memory,
time_major=time_major, scope=fw_scope)
# Backward direction
if not time_major:
time_dim = 1
batch_dim = 0
else:
time_dim = 0
batch_dim = 1
with vs.variable_scope("BW") as bw_scope:
inputs_reverse = array_ops.reverse_sequence(
input=inputs, seq_lengths=sequence_length,
seq_dim=time_dim, batch_dim=batch_dim)
tmp, output_state_bw = dynamic_rnn(
cell=cell_bw, inputs=inputs_reverse, sequence_length=sequence_length,
initial_state=initial_state_bw, dtype=dtype,
parallel_iterations=parallel_iterations, swap_memory=swap_memory,
time_major=time_major, scope=bw_scope)
output_bw = array_ops.reverse_sequence(
input=tmp, seq_lengths=sequence_length,
seq_dim=time_dim, batch_dim=batch_dim)
outputs = (output_fw, output_bw)
output_states = (output_state_fw, output_state_bw)
return (outputs, output_states)
def dynamic_rnn(cell, inputs, sequence_length=None, initial_state=None,
dtype=None, parallel_iterations=None, swap_memory=False,
time_major=False, scope=None):
"""Creates a recurrent neural network specified by RNNCell `cell`.
This function is functionally identical to the function `rnn` above, but
performs fully dynamic unrolling of `inputs`.
Unlike `rnn`, the input `inputs` is not a Python list of `Tensors`, one for
each frame. Instead, `inputs` may be a single `Tensor` where
the maximum time is either the first or second dimension (see the parameter
`time_major`). Alternatively, it may be a (possibly nested) tuple of
Tensors, each of them having matching batch and time dimensions.
The corresponding output is either a single `Tensor` having the same number
of time steps and batch size, or a (possibly nested) tuple of such tensors,
matching the nested structure of `cell.output_size`.
The parameter `sequence_length` is optional and is used to copy-through state
and zero-out outputs when past a batch element's sequence length. So it's more
for correctness than performance, unlike in rnn().
Args:
cell: An instance of RNNCell.
inputs: The RNN inputs.
If `time_major == False` (default), this must be a `Tensor` of shape:
`[batch_size, max_time, ...]`, or a nested tuple of such
elements.
If `time_major == True`, this must be a `Tensor` of shape:
`[max_time, batch_size, ...]`, or a nested tuple of such
elements.
This may also be a (possibly nested) tuple of Tensors satisfying
this property. The first two dimensions must match across all the inputs,
but otherwise the ranks and other shape components may differ.
In this case, input to `cell` at each time-step will replicate the
structure of these tuples, except for the time dimension (from which the
time is taken).
The input to `cell` at each time step will be a `Tensor` or (possibly
nested) tuple of Tensors each with dimensions `[batch_size, ...]`.
sequence_length: (optional) An int32/int64 vector sized `[batch_size]`.
initial_state: (optional) An initial state for the RNN.
If `cell.state_size` is an integer, this must be
a `Tensor` of appropriate type and shape `[batch_size, cell.state_size]`.
If `cell.state_size` is a tuple, this should be a tuple of
tensors having shapes `[batch_size, s] for s in cell.state_size`.
dtype: (optional) The data type for the initial state and expected output.
Required if initial_state is not provided or RNN state has a heterogeneous
dtype.
parallel_iterations: (Default: 32). The number of iterations to run in
parallel. Those operations which do not have any temporal dependency
and can be run in parallel, will be. This parameter trades off
time for space. Values >> 1 use more memory but take less time,
while smaller values use less memory but computations take longer.
swap_memory: Transparently swap the tensors produced in forward inference
but needed for back prop from GPU to CPU. This allows training RNNs
which would typically not fit on a single GPU, with very minimal (or no)
performance penalty.
time_major: The shape format of the `inputs` and `outputs` Tensors.
If true, these `Tensors` must be shaped `[max_time, batch_size, depth]`.
If false, these `Tensors` must be shaped `[batch_size, max_time, depth]`.
Using `time_major = True` is a bit more efficient because it avoids
transposes at the beginning and end of the RNN calculation. However,
most TensorFlow data is batch-major, so by default this function
accepts input and emits output in batch-major form.
scope: VariableScope for the created subgraph; defaults to "RNN".
Returns:
A pair (outputs, state) where:
outputs: The RNN output `Tensor`.
If time_major == False (default), this will be a `Tensor` shaped:
`[batch_size, max_time, cell.output_size]`.
If time_major == True, this will be a `Tensor` shaped:
`[max_time, batch_size, cell.output_size]`.
Note, if `cell.output_size` is a (possibly nested) tuple of integers
or `TensorShape` objects, then `outputs` will be a tuple having the
same structure as `cell.output_size`, containing Tensors having shapes
corresponding to the shape data in `cell.output_size`.
state: The final state. If `cell.state_size` is an int, this
will be shaped `[batch_size, cell.state_size]`. If it is a
`TensorShape`, this will be shaped `[batch_size] + cell.state_size`.
If it is a (possibly nested) tuple of ints or `TensorShape`, this will
be a tuple having the corresponding shapes.
Raises:
TypeError: If `cell` is not an instance of RNNCell.
ValueError: If inputs is None or an empty list.
"""
if not isinstance(cell, rnn_cell.RNNCell):
raise TypeError("cell must be an instance of RNNCell")
# By default, time_major==False and inputs are batch-major: shaped
# [batch, time, depth]
# For internal calculations, we transpose to [time, batch, depth]
flat_input = nest.flatten(inputs)
if not time_major:
# (B,T,D) => (T,B,D)
flat_input = tuple(array_ops.transpose(input_, [1, 0, 2])
for input_ in flat_input)
parallel_iterations = parallel_iterations or 32
if sequence_length is not None:
sequence_length = math_ops.to_int32(sequence_length)
sequence_length = array_ops.identity( # Just to find it in the graph.
sequence_length, name="sequence_length")
# Create a new scope in which the caching device is either
# determined by the parent scope, or is set to place the cached
# Variable using the same placement as for the rest of the RNN.
with vs.variable_scope(scope or "RNN") as varscope:
if varscope.caching_device is None:
varscope.set_caching_device(lambda op: op.device)
input_shape = tuple(array_ops.shape(input_) for input_ in flat_input)
batch_size = input_shape[0][1]
for input_ in input_shape:
if input_[1].get_shape() != batch_size.get_shape():
raise ValueError("All inputs should have the same batch size")
if initial_state is not None:
state = initial_state
else:
if not dtype:
raise ValueError("If no initial_state is provided, dtype must be.")
state = cell.zero_state(batch_size, dtype)
def _assert_has_shape(x, shape):
x_shape = array_ops.shape(x)
packed_shape = array_ops.pack(shape)
return control_flow_ops.Assert(
math_ops.reduce_all(math_ops.equal(x_shape, packed_shape)),
["Expected shape for Tensor %s is " % x.name,
packed_shape, " but saw shape: ", x_shape])
if sequence_length is not None:
# Perform some shape validation
with ops.control_dependencies(
[_assert_has_shape(sequence_length, [batch_size])]):
sequence_length = array_ops.identity(
sequence_length, name="CheckSeqLen")
inputs = nest.pack_sequence_as(structure=inputs, flat_sequence=flat_input)
(outputs, final_state) = _dynamic_rnn_loop(
cell,
inputs,
state,
parallel_iterations=parallel_iterations,
swap_memory=swap_memory,
sequence_length=sequence_length,
dtype=dtype)
# Outputs of _dynamic_rnn_loop are always shaped [time, batch, depth].
# If we are performing batch-major calculations, transpose output back
# to shape [batch, time, depth]
if not time_major:
# (T,B,D) => (B,T,D)
flat_output = nest.flatten(outputs)
flat_output = [array_ops.transpose(output, [1, 0, 2])
for output in flat_output]
outputs = nest.pack_sequence_as(
structure=outputs, flat_sequence=flat_output)
return (outputs, final_state)
def _dynamic_rnn_loop(cell,
inputs,
initial_state,
parallel_iterations,
swap_memory,
sequence_length=None,
dtype=None):
"""Internal implementation of Dynamic RNN.
Args:
cell: An instance of RNNCell.
inputs: A `Tensor` of shape [time, batch_size, input_size], or a nested
tuple of such elements.
initial_state: A `Tensor` of shape `[batch_size, state_size]`, or if
`cell.state_size` is a tuple, then this should be a tuple of
tensors having shapes `[batch_size, s] for s in cell.state_size`.
parallel_iterations: Positive Python int.
swap_memory: A Python boolean
sequence_length: (optional) An `int32` `Tensor` of shape [batch_size].
dtype: (optional) Expected dtype of output. If not specified, inferred from
initial_state.
Returns:
Tuple `(final_outputs, final_state)`.
final_outputs:
A `Tensor` of shape `[time, batch_size, cell.output_size]`. If
`cell.output_size` is a (possibly nested) tuple of ints or `TensorShape`
objects, then this returns a (possibly nsted) tuple of Tensors matching
the corresponding shapes.
final_state:
A `Tensor`, or possibly nested tuple of Tensors, matching in length
and shapes to `initial_state`.
Raises:
ValueError: If the input depth cannot be inferred via shape inference
from the inputs.
"""
state = initial_state
assert isinstance(parallel_iterations, int), "parallel_iterations must be int"
state_size = cell.state_size
flat_input = nest.flatten(inputs)
flat_output_size = nest.flatten(cell.output_size)
# Construct an initial output
input_shape = array_ops.shape(flat_input[0])
time_steps = input_shape[0]
batch_size = input_shape[1]
inputs_got_shape = tuple(input_.get_shape().with_rank_at_least(3)
for input_ in flat_input)
const_time_steps, const_batch_size = inputs_got_shape[0].as_list()[:2]
for shape in inputs_got_shape:
if not shape[2:].is_fully_defined():
raise ValueError(
"Input size (depth of inputs) must be accessible via shape inference,"
" but saw value None.")
got_time_steps = shape[0]
got_batch_size = shape[1]
if const_time_steps != got_time_steps:
raise ValueError(
"Time steps is not the same for all the elements in the input in a "
"batch.")
if const_batch_size != got_batch_size:
raise ValueError(
"Batch_size is not the same for all the elements in the input.")
# Prepare dynamic conditional copying of state & output
def _create_zero_arrays(size):
size = _state_size_with_prefix(size, prefix=[batch_size])
return array_ops.zeros(
array_ops.pack(size), _infer_state_dtype(dtype, state))
flat_zero_output = tuple(_create_zero_arrays(output)
for output in flat_output_size)
zero_output = nest.pack_sequence_as(structure=cell.output_size,
flat_sequence=flat_zero_output)
if sequence_length is not None:
min_sequence_length = math_ops.reduce_min(sequence_length)
max_sequence_length = math_ops.reduce_max(sequence_length)
time = array_ops.constant(0, dtype=dtypes.int32, name="time")
with ops.name_scope("dynamic_rnn") as scope:
base_name = scope
def _create_ta(name, dtype):
return tensor_array_ops.TensorArray(dtype=dtype,
size=time_steps,
tensor_array_name=base_name + name)
output_ta = tuple(_create_ta("output_%d" % i,
_infer_state_dtype(dtype, state))
for i in range(len(flat_output_size)))
input_ta = tuple(_create_ta("input_%d" % i, flat_input[0].dtype)
for i in range(len(flat_input)))
input_ta = tuple(ta.unpack(input_)
for ta, input_ in zip(input_ta, flat_input))
def _time_step(time, output_ta_t, state):
"""Take a time step of the dynamic RNN.
Args:
time: int32 scalar Tensor.
output_ta_t: List of `TensorArray`s that represent the output.
state: nested tuple of vector tensors that represent the state.
Returns:
The tuple (time + 1, output_ta_t with updated flow, new_state).
"""
input_t = tuple(ta.read(time) for ta in input_ta)
# Restore some shape information
for input_, shape in zip(input_t, inputs_got_shape):
input_.set_shape(shape[1:])
input_t = nest.pack_sequence_as(structure=inputs, flat_sequence=input_t)
call_cell = lambda: cell(input_t, state)
if sequence_length is not None:
(output, new_state) = _rnn_step(
time=time,
sequence_length=sequence_length,
min_sequence_length=min_sequence_length,
max_sequence_length=max_sequence_length,
zero_output=zero_output,
state=state,
call_cell=call_cell,
state_size=state_size,
skip_conditionals=True)
else:
(output, new_state) = call_cell()
# Pack state if using state tuples
output = nest.flatten(output)
output_ta_t = tuple(
ta.write(time, out) for ta, out in zip(output_ta_t, output))
return (time + 1, output_ta_t, new_state)
_, output_final_ta, final_state = control_flow_ops.while_loop(
cond=lambda time, *_: time < time_steps,
body=_time_step,