forked from pytorch/tutorials
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathseq2seq_translation_tutorial.html
1467 lines (1230 loc) · 148 KB
/
seq2seq_translation_tutorial.html
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
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Translation with a Sequence to Sequence Network and Attention — PyTorch Tutorials 0.1.12_2 documentation</title>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/pytorch_theme.css" type="text/css" />
<link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Lato" type="text/css" />
<link rel="index" title="Index"
href="../genindex.html"/>
<link rel="search" title="Search" href="../search.html"/>
<link rel="top" title="PyTorch Tutorials 0.1.12_2 documentation" href="../index.html"/>
<link rel="next" title="Reinforcement Learning (DQN) tutorial" href="reinforcement_q_learning.html"/>
<link rel="prev" title="Generating Names with a Character-Level RNN" href="char_rnn_generation_tutorial.html"/>
<script src="../_static/js/modernizr.min.js"></script>
</head>
<body class="wy-body-for-nav" role="document">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search">
<a href="../index.html" class="icon icon-home"> PyTorch Tutorials
<img src="../_static/pytorch-logo-dark.svg" class="logo" />
</a>
<div class="version">
0.1.12_2
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">Beginner Tutorials</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">Deep Learning with PyTorch: A 60 Minute Blitz</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html">What is PyTorch?</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#getting-started">Getting Started</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#tensors">Tensors</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#operations">Operations</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#numpy-bridge">Numpy Bridge</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#converting-torch-tensor-to-numpy-array">Converting torch Tensor to numpy Array</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#converting-numpy-array-to-torch-tensor">Converting numpy Array to torch Tensor</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/tensor_tutorial.html#cuda-tensors">CUDA Tensors</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/blitz/autograd_tutorial.html">Autograd: automatic differentiation</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/autograd_tutorial.html#variable">Variable</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/autograd_tutorial.html#gradients">Gradients</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/blitz/neural_networks_tutorial.html">Neural Networks</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/neural_networks_tutorial.html#define-the-network">Define the network</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/neural_networks_tutorial.html#loss-function">Loss Function</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/neural_networks_tutorial.html#backprop">Backprop</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/neural_networks_tutorial.html#update-the-weights">Update the weights</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html">Training a classifier</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#what-about-data">What about data?</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#training-an-image-classifier">Training an image classifier</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#loading-and-normalizing-cifar10">1. Loading and normalizing CIFAR10</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#define-a-convolution-neural-network">2. Define a Convolution Neural Network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#define-a-loss-function-and-optimizer">3. Define a Loss function and optimizer</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#train-the-network">4. Train the network</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#test-the-network-on-the-test-data">5. Test the network on the test data</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#training-on-gpu">Training on GPU</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/blitz/cifar10_tutorial.html#where-do-i-go-next">Where do I go next?</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/former_torchies_tutorial.html">PyTorch for former Torch users</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html">Tensors</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#inplace-out-of-place">Inplace / Out-of-place</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#zero-indexing">Zero Indexing</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#no-camel-casing">No camel casing</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#numpy-bridge">Numpy Bridge</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#converting-torch-tensor-to-numpy-array">Converting torch Tensor to numpy Array</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#converting-numpy-array-to-torch-tensor">Converting numpy Array to torch Tensor</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/tensor_tutorial.html#cuda-tensors">CUDA Tensors</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/former_torchies/autograd_tutorial.html">Autograd</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/autograd_tutorial.html#variable">Variable</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/autograd_tutorial.html#gradients">Gradients</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/former_torchies/nn_tutorial.html">nn package</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/nn_tutorial.html#example-1-convnet">Example 1: ConvNet</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/nn_tutorial.html#forward-and-backward-function-hooks">Forward and Backward Function Hooks</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/nn_tutorial.html#example-2-recurrent-net">Example 2: Recurrent Net</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/former_torchies/parallelism_tutorial.html">Multi-GPU examples</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/parallelism_tutorial.html#dataparallel">DataParallel</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/former_torchies/parallelism_tutorial.html#part-of-the-model-on-cpu-and-part-on-the-gpu">Part of the model on CPU and part on the GPU</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">Learning PyTorch with Examples</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/pytorch_with_examples.html#tensors">Tensors</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#warm-up-numpy">Warm-up: numpy</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-tensors">PyTorch: Tensors</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/pytorch_with_examples.html#autograd">Autograd</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-variables-and-autograd">PyTorch: Variables and autograd</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-defining-new-autograd-functions">PyTorch: Defining new autograd functions</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#tensorflow-static-graphs">TensorFlow: Static Graphs</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/pytorch_with_examples.html#nn-module"><cite>nn</cite> module</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-nn">PyTorch: nn</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-optim">PyTorch: optim</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-custom-nn-modules">PyTorch: Custom nn Modules</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#pytorch-control-flow-weight-sharing">PyTorch: Control Flow + Weight Sharing</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/pytorch_with_examples.html#examples">Examples</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#id1">Tensors</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_tensor/two_layer_net_numpy.html">Warm-up: numpy</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_tensor/two_layer_net_tensor.html">PyTorch: Tensors</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#id2">Autograd</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_autograd/two_layer_net_autograd.html">PyTorch: Variables and autograd</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_autograd/two_layer_net_custom_function.html">PyTorch: Defining new autograd functions</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_autograd/tf_two_layer_net.html">TensorFlow: Static Graphs</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/pytorch_with_examples.html#id3"><cite>nn</cite> module</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_nn/two_layer_net_nn.html">PyTorch: nn</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_nn/two_layer_net_optim.html">PyTorch: optim</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_nn/two_layer_net_module.html">PyTorch: Custom nn Modules</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/examples_nn/dynamic_net.html">PyTorch: Control Flow + Weight Sharing</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">Transfer Learning tutorial</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#load-data">Load Data</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#visualize-a-few-images">Visualize a few images</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#training-the-model">Training the model</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#learning-rate-scheduler">Learning rate scheduler</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#visualizing-the-model-predictions">Visualizing the model predictions</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#finetuning-the-convnet">Finetuning the convnet</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#train-and-evaluate">Train and evaluate</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#convnet-as-fixed-feature-extractor">ConvNet as fixed feature extractor</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html#id1">Train and evaluate</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/data_loading_tutorial.html">Data Loading and Processing Tutorial</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/data_loading_tutorial.html#dataset-class">Dataset class</a></li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/data_loading_tutorial.html#transforms">Transforms</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/data_loading_tutorial.html#compose-transforms">Compose transforms</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/data_loading_tutorial.html#iterating-through-the-dataset">Iterating through the dataset</a></li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/data_loading_tutorial.html#afterword-torchvision">Afterword: torchvision</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_nlp_tutorial.html">Deep Learning for NLP with Pytorch</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html">Introduction to PyTorch</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html#introduction-to-torch-s-tensor-library">Introduction to Torch’s tensor library</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html#creating-tensors">Creating Tensors</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html#operations-with-tensors">Operations with Tensors</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html#reshaping-tensors">Reshaping Tensors</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/pytorch_tutorial.html#computation-graphs-and-automatic-differentiation">Computation Graphs and Automatic Differentiation</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html">Deep Learning with PyTorch</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#deep-learning-building-blocks-affine-maps-non-linearities-and-objectives">Deep Learning Building Blocks: Affine maps, non-linearities and objectives</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#affine-maps">Affine Maps</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#non-linearities">Non-Linearities</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#softmax-and-probabilities">Softmax and Probabilities</a></li>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#objective-functions">Objective Functions</a></li>
</ul>
</li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#optimization-and-training">Optimization and Training</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#creating-network-components-in-pytorch">Creating Network Components in Pytorch</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../beginner/nlp/deep_learning_tutorial.html#example-logistic-regression-bag-of-words-classifier">Example: Logistic Regression Bag-of-Words classifier</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/nlp/word_embeddings_tutorial.html">Word Embeddings: Encoding Lexical Semantics</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/word_embeddings_tutorial.html#getting-dense-word-embeddings">Getting Dense Word Embeddings</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/word_embeddings_tutorial.html#word-embeddings-in-pytorch">Word Embeddings in Pytorch</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/word_embeddings_tutorial.html#an-example-n-gram-language-modeling">An Example: N-Gram Language Modeling</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/word_embeddings_tutorial.html#exercise-computing-word-embeddings-continuous-bag-of-words">Exercise: Computing Word Embeddings: Continuous Bag-of-Words</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/nlp/sequence_models_tutorial.html">Sequence Models and Long-Short Term Memory Networks</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/sequence_models_tutorial.html#lstm-s-in-pytorch">LSTM’s in Pytorch</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/sequence_models_tutorial.html#example-an-lstm-for-part-of-speech-tagging">Example: An LSTM for Part-of-Speech Tagging</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/sequence_models_tutorial.html#exercise-augmenting-the-lstm-part-of-speech-tagger-with-character-level-features">Exercise: Augmenting the LSTM part-of-speech tagger with character-level features</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../beginner/nlp/advanced_tutorial.html">Advanced: Making Dynamic Decisions and the Bi-LSTM CRF</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/advanced_tutorial.html#dynamic-versus-static-deep-learning-toolkits">Dynamic versus Static Deep Learning Toolkits</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/advanced_tutorial.html#bi-lstm-conditional-random-field-discussion">Bi-LSTM Conditional Random Field Discussion</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/advanced_tutorial.html#implementation-notes">Implementation Notes</a></li>
<li class="toctree-l3"><a class="reference internal" href="../beginner/nlp/advanced_tutorial.html#exercise-a-new-loss-function-for-discriminative-tagging">Exercise: A new loss function for discriminative tagging</a></li>
</ul>
</li>
</ul>
</li>
</ul>
<p class="caption"><span class="caption-text">Intermediate Tutorials</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="char_rnn_classification_tutorial.html">Classifying Names with a Character-Level RNN</a><ul>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_classification_tutorial.html#preparing-the-data">Preparing the Data</a><ul>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_classification_tutorial.html#turning-names-into-tensors">Turning Names into Tensors</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_classification_tutorial.html#creating-the-network">Creating the Network</a></li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_classification_tutorial.html#training">Training</a><ul>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_classification_tutorial.html#preparing-for-training">Preparing for Training</a></li>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_classification_tutorial.html#training-the-network">Training the Network</a></li>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_classification_tutorial.html#plotting-the-results">Plotting the Results</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_classification_tutorial.html#evaluating-the-results">Evaluating the Results</a><ul>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_classification_tutorial.html#running-on-user-input">Running on User Input</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_classification_tutorial.html#exercises">Exercises</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_generation_tutorial.html">Generating Names with a Character-Level RNN</a><ul>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_generation_tutorial.html#preparing-the-data">Preparing the Data</a></li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_generation_tutorial.html#creating-the-network">Creating the Network</a></li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_generation_tutorial.html#training">Training</a><ul>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_generation_tutorial.html#preparing-for-training">Preparing for Training</a></li>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_generation_tutorial.html#training-the-network">Training the Network</a></li>
<li class="toctree-l3"><a class="reference internal" href="char_rnn_generation_tutorial.html#plotting-the-losses">Plotting the Losses</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_generation_tutorial.html#sampling-the-network">Sampling the Network</a></li>
<li class="toctree-l2"><a class="reference internal" href="char_rnn_generation_tutorial.html#exercises">Exercises</a></li>
</ul>
</li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Translation with a Sequence to Sequence Network and Attention</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#loading-data-files">Loading data files</a></li>
<li class="toctree-l2"><a class="reference internal" href="#the-seq2seq-model">The Seq2Seq Model</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#the-encoder">The Encoder</a></li>
<li class="toctree-l3"><a class="reference internal" href="#the-decoder">The Decoder</a><ul>
<li class="toctree-l4"><a class="reference internal" href="#simple-decoder">Simple Decoder</a></li>
<li class="toctree-l4"><a class="reference internal" href="#attention-decoder">Attention Decoder</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#training">Training</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#preparing-training-data">Preparing Training Data</a></li>
<li class="toctree-l3"><a class="reference internal" href="#training-the-model">Training the Model</a></li>
<li class="toctree-l3"><a class="reference internal" href="#plotting-results">Plotting results</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#evaluation">Evaluation</a></li>
<li class="toctree-l2"><a class="reference internal" href="#training-and-evaluating">Training and Evaluating</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#visualizing-attention">Visualizing Attention</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#exercises">Exercises</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_q_learning.html">Reinforcement Learning (DQN) tutorial</a><ul>
<li class="toctree-l2"><a class="reference internal" href="reinforcement_q_learning.html#replay-memory">Replay Memory</a></li>
<li class="toctree-l2"><a class="reference internal" href="reinforcement_q_learning.html#dqn-algorithm">DQN algorithm</a><ul>
<li class="toctree-l3"><a class="reference internal" href="reinforcement_q_learning.html#q-network">Q-network</a></li>
<li class="toctree-l3"><a class="reference internal" href="reinforcement_q_learning.html#input-extraction">Input extraction</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="reinforcement_q_learning.html#training">Training</a><ul>
<li class="toctree-l3"><a class="reference internal" href="reinforcement_q_learning.html#hyperparameters-and-utilities">Hyperparameters and utilities</a></li>
<li class="toctree-l3"><a class="reference internal" href="reinforcement_q_learning.html#training-loop">Training loop</a></li>
</ul>
</li>
</ul>
</li>
</ul>
<p class="caption"><span class="caption-text">Advanced Tutorials</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../advanced/neural_style_tutorial.html">Neural Transfer with PyTorch</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../advanced/neural_style_tutorial.html#introduction">Introduction</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#neural-what">Neural what?</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#how-does-it-work">How does it work?</a><ul>
<li class="toctree-l4"><a class="reference internal" href="../advanced/neural_style_tutorial.html#ok-how-does-it-work">OK. How does it work?</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../advanced/neural_style_tutorial.html#pytorch-implementation">PyTorch implementation</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#packages">Packages</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#cuda">Cuda</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#load-images">Load images</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#display-images">Display images</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#content-loss">Content loss</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#style-loss">Style loss</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#load-the-neural-network">Load the neural network</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#input-image">Input image</a></li>
<li class="toctree-l3"><a class="reference internal" href="../advanced/neural_style_tutorial.html#gradient-descent">Gradient descent</a></li>
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/numpy_extensions_tutorial.html">Creating extensions using numpy and scipy</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../advanced/numpy_extensions_tutorial.html#parameter-less-example">Parameter-less example</a></li>
<li class="toctree-l2"><a class="reference internal" href="../advanced/numpy_extensions_tutorial.html#parametrized-example">Parametrized example</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/c_extension.html">Custom C extensions for pytorch</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../advanced/c_extension.html#step-1-prepare-your-c-code">Step 1. prepare your C code</a></li>
<li class="toctree-l2"><a class="reference internal" href="../advanced/c_extension.html#step-2-include-it-in-your-python-code">Step 2: Include it in your Python code</a></li>
</ul>
</li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" role="navigation" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../index.html">PyTorch Tutorials</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="../index.html">Docs</a> »</li>
<li>Translation with a Sequence to Sequence Network and Attention</li>
<li class="wy-breadcrumbs-aside">
<a href="../_sources/intermediate/seq2seq_translation_tutorial.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="translation-with-a-sequence-to-sequence-network-and-attention">
<span id="sphx-glr-intermediate-seq2seq-translation-tutorial-py"></span><h1>Translation with a Sequence to Sequence Network and Attention<a class="headerlink" href="#translation-with-a-sequence-to-sequence-network-and-attention" title="Permalink to this headline">¶</a></h1>
<p><strong>Author</strong>: <a class="reference external" href="https://github.com/spro/practical-pytorch">Sean Robertson</a></p>
<p>In this project we will be teaching a neural network to translate from
French to English.</p>
<div class="highlight-default"><div class="highlight"><pre><span></span>[KEY: > input, = target, < output]
> il est en train de peindre un tableau .
= he is painting a picture .
< he is painting a picture .
> pourquoi ne pas essayer ce vin delicieux ?
= why not try that delicious wine ?
< why not try that delicious wine ?
> elle n est pas poete mais romanciere .
= she is not a poet but a novelist .
< she not not a poet but a novelist .
> vous etes trop maigre .
= you re too skinny .
< you re all alone .
</pre></div>
</div>
<p>... to varying degrees of success.</p>
<p>This is made possible by the simple but powerful idea of the <a class="reference external" href="http://arxiv.org/abs/1409.3215">sequence
to sequence network</a>, in which two
recurrent neural networks work together to transform one sequence to
another. An encoder network condenses an input sequence into a vector,
and a decoder network unfolds that vector into a new sequence.</p>
<div class="figure">
<img alt="" src="../_images/seq2seq.png" />
</div>
<p>To improve upon this model we’ll use an <a class="reference external" href="https://arxiv.org/abs/1409.0473">attention
mechanism</a>, which lets the decoder
learn to focus over a specific range of the input sequence.</p>
<p><strong>Recommended Reading:</strong></p>
<p>I assume you have at least installed PyTorch, know Python, and
understand Tensors:</p>
<ul class="simple">
<li><a class="reference external" href="http://pytorch.org/">http://pytorch.org/</a> For installation instructions</li>
<li><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html"><span class="doc">Deep Learning with PyTorch: A 60 Minute Blitz</span></a> to get started with PyTorch in general</li>
<li><a class="reference internal" href="../beginner/pytorch_with_examples.html"><span class="doc">Learning PyTorch with Examples</span></a> for a wide and deep overview</li>
<li><a class="reference internal" href="../beginner/former_torchies_tutorial.html"><span class="doc">PyTorch for former Torch users</span></a> if you are former Lua Torch user</li>
</ul>
<p>It would also be useful to know about Sequence to Sequence networks and
how they work:</p>
<ul class="simple">
<li><a class="reference external" href="http://arxiv.org/abs/1406.1078">Learning Phrase Representations using RNN Encoder-Decoder for
Statistical Machine Translation</a></li>
<li><a class="reference external" href="http://arxiv.org/abs/1409.3215">Sequence to Sequence Learning with Neural
Networks</a></li>
<li><a class="reference external" href="https://arxiv.org/abs/1409.0473">Neural Machine Translation by Jointly Learning to Align and
Translate</a></li>
<li><a class="reference external" href="http://arxiv.org/abs/1506.05869">A Neural Conversational Model</a></li>
</ul>
<p>You will also find the previous tutorials on
<a class="reference internal" href="char_rnn_classification_tutorial.html"><span class="doc">Classifying Names with a Character-Level RNN</span></a>
and <a class="reference internal" href="char_rnn_generation_tutorial.html"><span class="doc">Generating Names with a Character-Level RNN</span></a>
helpful as those concepts are very similar to the Encoder and Decoder
models, respectively.</p>
<p>And for more, read the papers that introduced these topics:</p>
<ul class="simple">
<li><a class="reference external" href="http://arxiv.org/abs/1406.1078">Learning Phrase Representations using RNN Encoder-Decoder for
Statistical Machine Translation</a></li>
<li><a class="reference external" href="http://arxiv.org/abs/1409.3215">Sequence to Sequence Learning with Neural
Networks</a></li>
<li><a class="reference external" href="https://arxiv.org/abs/1409.0473">Neural Machine Translation by Jointly Learning to Align and
Translate</a></li>
<li><a class="reference external" href="http://arxiv.org/abs/1506.05869">A Neural Conversational Model</a></li>
</ul>
<p><strong>Requirements</strong></p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">__future__</span> <span class="kn">import</span> <span class="n">unicode_literals</span><span class="p">,</span> <span class="n">print_function</span><span class="p">,</span> <span class="n">division</span>
<span class="kn">from</span> <span class="nn">io</span> <span class="kn">import</span> <span class="nb">open</span>
<span class="kn">import</span> <span class="nn">unicodedata</span>
<span class="kn">import</span> <span class="nn">string</span>
<span class="kn">import</span> <span class="nn">re</span>
<span class="kn">import</span> <span class="nn">random</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torch.nn</span> <span class="kn">as</span> <span class="nn">nn</span>
<span class="kn">from</span> <span class="nn">torch.autograd</span> <span class="kn">import</span> <span class="n">Variable</span>
<span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">optim</span>
<span class="kn">import</span> <span class="nn">torch.nn.functional</span> <span class="kn">as</span> <span class="nn">F</span>
<span class="n">use_cuda</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span>
</pre></div>
</div>
<div class="section" id="loading-data-files">
<h2>Loading data files<a class="headerlink" href="#loading-data-files" title="Permalink to this headline">¶</a></h2>
<p>The data for this project is a set of many thousands of English to
French translation pairs.</p>
<p><a class="reference external" href="http://opendata.stackexchange.com/questions/3888/dataset-of-sentences-translated-into-many-languages">This question on Open Data Stack
Exchange</a>
pointed me to the open translation site <a class="reference external" href="http://tatoeba.org/">http://tatoeba.org/</a> which has
downloads available at <a class="reference external" href="http://tatoeba.org/eng/downloads">http://tatoeba.org/eng/downloads</a> - and better
yet, someone did the extra work of splitting language pairs into
individual text files here: <a class="reference external" href="http://www.manythings.org/anki/">http://www.manythings.org/anki/</a></p>
<p>The English to French pairs are too big to include in the repo, so
download to <code class="docutils literal"><span class="pre">data/eng-fra.txt</span></code> before continuing. The file is a tab
separated list of translation pairs:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">I</span> <span class="n">am</span> <span class="n">cold</span><span class="o">.</span> <span class="n">Je</span> <span class="n">suis</span> <span class="n">froid</span><span class="o">.</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Download the data from
<a class="reference external" href="https://download.pytorch.org/tutorial/data.zip">here</a>
and extract it to the current directory.</p>
</div>
<p>Similar to the character encoding used in the character-level RNN
tutorials, we will be representing each word in a language as a one-hot
vector, or giant vector of zeros except for a single one (at the index
of the word). Compared to the dozens of characters that might exist in a
language, there are many many more words, so the encoding vector is much
larger. We will however cheat a bit and trim the data to only use a few
thousand words per language.</p>
<div class="figure">
<img alt="" src="../_images/word-encoding.png" />
</div>
<p>We’ll need a unique index per word to use as the inputs and targets of
the networks later. To keep track of all this we will use a helper class
called <code class="docutils literal"><span class="pre">Lang</span></code> which has word → index (<code class="docutils literal"><span class="pre">word2index</span></code>) and index → word
(<code class="docutils literal"><span class="pre">index2word</span></code>) dictionaries, as well as a count of each word
<code class="docutils literal"><span class="pre">word2count</span></code> to use to later replace rare words.</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">SOS_token</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">EOS_token</span> <span class="o">=</span> <span class="mi">1</span>
<span class="k">class</span> <span class="nc">Lang</span><span class="p">:</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">name</span> <span class="o">=</span> <span class="n">name</span>
<span class="bp">self</span><span class="o">.</span><span class="n">word2index</span> <span class="o">=</span> <span class="p">{}</span>
<span class="bp">self</span><span class="o">.</span><span class="n">word2count</span> <span class="o">=</span> <span class="p">{}</span>
<span class="bp">self</span><span class="o">.</span><span class="n">index2word</span> <span class="o">=</span> <span class="p">{</span><span class="mi">0</span><span class="p">:</span> <span class="s2">"SOS"</span><span class="p">,</span> <span class="mi">1</span><span class="p">:</span> <span class="s2">"EOS"</span><span class="p">}</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_words</span> <span class="o">=</span> <span class="mi">2</span> <span class="c1"># Count SOS and EOS</span>
<span class="k">def</span> <span class="nf">addSentence</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">sentence</span><span class="p">):</span>
<span class="k">for</span> <span class="n">word</span> <span class="ow">in</span> <span class="n">sentence</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">' '</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">addWord</span><span class="p">(</span><span class="n">word</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">addWord</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">word</span><span class="p">):</span>
<span class="k">if</span> <span class="n">word</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">word2index</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">word2index</span><span class="p">[</span><span class="n">word</span><span class="p">]</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">n_words</span>
<span class="bp">self</span><span class="o">.</span><span class="n">word2count</span><span class="p">[</span><span class="n">word</span><span class="p">]</span> <span class="o">=</span> <span class="mi">1</span>
<span class="bp">self</span><span class="o">.</span><span class="n">index2word</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">n_words</span><span class="p">]</span> <span class="o">=</span> <span class="n">word</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_words</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">word2count</span><span class="p">[</span><span class="n">word</span><span class="p">]</span> <span class="o">+=</span> <span class="mi">1</span>
</pre></div>
</div>
<p>The files are all in Unicode, to simplify we will turn Unicode
characters to ASCII, make everything lowercase, and trim most
punctuation.</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="c1"># Turn a Unicode string to plain ASCII, thanks to</span>
<span class="c1"># http://stackoverflow.com/a/518232/2809427</span>
<span class="k">def</span> <span class="nf">unicodeToAscii</span><span class="p">(</span><span class="n">s</span><span class="p">):</span>
<span class="k">return</span> <span class="s1">''</span><span class="o">.</span><span class="n">join</span><span class="p">(</span>
<span class="n">c</span> <span class="k">for</span> <span class="n">c</span> <span class="ow">in</span> <span class="n">unicodedata</span><span class="o">.</span><span class="n">normalize</span><span class="p">(</span><span class="s1">'NFD'</span><span class="p">,</span> <span class="n">s</span><span class="p">)</span>
<span class="k">if</span> <span class="n">unicodedata</span><span class="o">.</span><span class="n">category</span><span class="p">(</span><span class="n">c</span><span class="p">)</span> <span class="o">!=</span> <span class="s1">'Mn'</span>
<span class="p">)</span>
<span class="c1"># Lowercase, trim, and remove non-letter characters</span>
<span class="k">def</span> <span class="nf">normalizeString</span><span class="p">(</span><span class="n">s</span><span class="p">):</span>
<span class="n">s</span> <span class="o">=</span> <span class="n">unicodeToAscii</span><span class="p">(</span><span class="n">s</span><span class="o">.</span><span class="n">lower</span><span class="p">()</span><span class="o">.</span><span class="n">strip</span><span class="p">())</span>
<span class="n">s</span> <span class="o">=</span> <span class="n">re</span><span class="o">.</span><span class="n">sub</span><span class="p">(</span><span class="sa">r</span><span class="s2">"([.!?])"</span><span class="p">,</span> <span class="sa">r</span><span class="s2">" \1"</span><span class="p">,</span> <span class="n">s</span><span class="p">)</span>
<span class="n">s</span> <span class="o">=</span> <span class="n">re</span><span class="o">.</span><span class="n">sub</span><span class="p">(</span><span class="sa">r</span><span class="s2">"[^a-zA-Z.!?]+"</span><span class="p">,</span> <span class="sa">r</span><span class="s2">" "</span><span class="p">,</span> <span class="n">s</span><span class="p">)</span>
<span class="k">return</span> <span class="n">s</span>
</pre></div>
</div>
<p>To read the data file we will split the file into lines, and then split
lines into pairs. The files are all English → Other Language, so if we
want to translate from Other Language → English I added the <code class="docutils literal"><span class="pre">reverse</span></code>
flag to reverse the pairs.</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">readLangs</span><span class="p">(</span><span class="n">lang1</span><span class="p">,</span> <span class="n">lang2</span><span class="p">,</span> <span class="n">reverse</span><span class="o">=</span><span class="bp">False</span><span class="p">):</span>
<span class="k">print</span><span class="p">(</span><span class="s2">"Reading lines..."</span><span class="p">)</span>
<span class="c1"># Read the file and split into lines</span>
<span class="n">lines</span> <span class="o">=</span> <span class="nb">open</span><span class="p">(</span><span class="s1">'data/</span><span class="si">%s</span><span class="s1">-</span><span class="si">%s</span><span class="s1">.txt'</span> <span class="o">%</span> <span class="p">(</span><span class="n">lang1</span><span class="p">,</span> <span class="n">lang2</span><span class="p">),</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">'utf-8'</span><span class="p">)</span><span class="o">.</span>\
<span class="n">read</span><span class="p">()</span><span class="o">.</span><span class="n">strip</span><span class="p">()</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">'</span><span class="se">\n</span><span class="s1">'</span><span class="p">)</span>
<span class="c1"># Split every line into pairs and normalize</span>
<span class="n">pairs</span> <span class="o">=</span> <span class="p">[[</span><span class="n">normalizeString</span><span class="p">(</span><span class="n">s</span><span class="p">)</span> <span class="k">for</span> <span class="n">s</span> <span class="ow">in</span> <span class="n">l</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">'</span><span class="se">\t</span><span class="s1">'</span><span class="p">)]</span> <span class="k">for</span> <span class="n">l</span> <span class="ow">in</span> <span class="n">lines</span><span class="p">]</span>
<span class="c1"># Reverse pairs, make Lang instances</span>
<span class="k">if</span> <span class="n">reverse</span><span class="p">:</span>
<span class="n">pairs</span> <span class="o">=</span> <span class="p">[</span><span class="nb">list</span><span class="p">(</span><span class="nb">reversed</span><span class="p">(</span><span class="n">p</span><span class="p">))</span> <span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="n">pairs</span><span class="p">]</span>
<span class="n">input_lang</span> <span class="o">=</span> <span class="n">Lang</span><span class="p">(</span><span class="n">lang2</span><span class="p">)</span>
<span class="n">output_lang</span> <span class="o">=</span> <span class="n">Lang</span><span class="p">(</span><span class="n">lang1</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">input_lang</span> <span class="o">=</span> <span class="n">Lang</span><span class="p">(</span><span class="n">lang1</span><span class="p">)</span>
<span class="n">output_lang</span> <span class="o">=</span> <span class="n">Lang</span><span class="p">(</span><span class="n">lang2</span><span class="p">)</span>
<span class="k">return</span> <span class="n">input_lang</span><span class="p">,</span> <span class="n">output_lang</span><span class="p">,</span> <span class="n">pairs</span>
</pre></div>
</div>
<p>Since there are a <em>lot</em> of example sentences and we want to train
something quickly, we’ll trim the data set to only relatively short and
simple sentences. Here the maximum length is 10 words (that includes
ending punctuation) and we’re filtering to sentences that translate to
the form “I am” or “He is” etc. (accounting for apostrophes replaced
earlier).</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">MAX_LENGTH</span> <span class="o">=</span> <span class="mi">10</span>
<span class="n">eng_prefixes</span> <span class="o">=</span> <span class="p">(</span>
<span class="s2">"i am "</span><span class="p">,</span> <span class="s2">"i m "</span><span class="p">,</span>
<span class="s2">"he is"</span><span class="p">,</span> <span class="s2">"he s "</span><span class="p">,</span>
<span class="s2">"she is"</span><span class="p">,</span> <span class="s2">"she s"</span><span class="p">,</span>
<span class="s2">"you are"</span><span class="p">,</span> <span class="s2">"you re "</span><span class="p">,</span>
<span class="s2">"we are"</span><span class="p">,</span> <span class="s2">"we re "</span><span class="p">,</span>
<span class="s2">"they are"</span><span class="p">,</span> <span class="s2">"they re "</span>
<span class="p">)</span>
<span class="k">def</span> <span class="nf">filterPair</span><span class="p">(</span><span class="n">p</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">len</span><span class="p">(</span><span class="n">p</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">' '</span><span class="p">))</span> <span class="o"><</span> <span class="n">MAX_LENGTH</span> <span class="ow">and</span> \
<span class="nb">len</span><span class="p">(</span><span class="n">p</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">' '</span><span class="p">))</span> <span class="o"><</span> <span class="n">MAX_LENGTH</span> <span class="ow">and</span> \
<span class="n">p</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="n">eng_prefixes</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">filterPairs</span><span class="p">(</span><span class="n">pairs</span><span class="p">):</span>
<span class="k">return</span> <span class="p">[</span><span class="n">pair</span> <span class="k">for</span> <span class="n">pair</span> <span class="ow">in</span> <span class="n">pairs</span> <span class="k">if</span> <span class="n">filterPair</span><span class="p">(</span><span class="n">pair</span><span class="p">)]</span>
</pre></div>
</div>
<p>The full process for preparing the data is:</p>
<ul class="simple">
<li>Read text file and split into lines, split lines into pairs</li>
<li>Normalize text, filter by length and content</li>
<li>Make word lists from sentences in pairs</li>
</ul>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">prepareData</span><span class="p">(</span><span class="n">lang1</span><span class="p">,</span> <span class="n">lang2</span><span class="p">,</span> <span class="n">reverse</span><span class="o">=</span><span class="bp">False</span><span class="p">):</span>
<span class="n">input_lang</span><span class="p">,</span> <span class="n">output_lang</span><span class="p">,</span> <span class="n">pairs</span> <span class="o">=</span> <span class="n">readLangs</span><span class="p">(</span><span class="n">lang1</span><span class="p">,</span> <span class="n">lang2</span><span class="p">,</span> <span class="n">reverse</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="s2">"Read </span><span class="si">%s</span><span class="s2"> sentence pairs"</span> <span class="o">%</span> <span class="nb">len</span><span class="p">(</span><span class="n">pairs</span><span class="p">))</span>
<span class="n">pairs</span> <span class="o">=</span> <span class="n">filterPairs</span><span class="p">(</span><span class="n">pairs</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="s2">"Trimmed to </span><span class="si">%s</span><span class="s2"> sentence pairs"</span> <span class="o">%</span> <span class="nb">len</span><span class="p">(</span><span class="n">pairs</span><span class="p">))</span>
<span class="k">print</span><span class="p">(</span><span class="s2">"Counting words..."</span><span class="p">)</span>
<span class="k">for</span> <span class="n">pair</span> <span class="ow">in</span> <span class="n">pairs</span><span class="p">:</span>
<span class="n">input_lang</span><span class="o">.</span><span class="n">addSentence</span><span class="p">(</span><span class="n">pair</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="n">output_lang</span><span class="o">.</span><span class="n">addSentence</span><span class="p">(</span><span class="n">pair</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
<span class="k">print</span><span class="p">(</span><span class="s2">"Counted words:"</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="n">input_lang</span><span class="o">.</span><span class="n">name</span><span class="p">,</span> <span class="n">input_lang</span><span class="o">.</span><span class="n">n_words</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="n">output_lang</span><span class="o">.</span><span class="n">name</span><span class="p">,</span> <span class="n">output_lang</span><span class="o">.</span><span class="n">n_words</span><span class="p">)</span>
<span class="k">return</span> <span class="n">input_lang</span><span class="p">,</span> <span class="n">output_lang</span><span class="p">,</span> <span class="n">pairs</span>
<span class="n">input_lang</span><span class="p">,</span> <span class="n">output_lang</span><span class="p">,</span> <span class="n">pairs</span> <span class="o">=</span> <span class="n">prepareData</span><span class="p">(</span><span class="s1">'eng'</span><span class="p">,</span> <span class="s1">'fra'</span><span class="p">,</span> <span class="bp">True</span><span class="p">)</span>
<span class="k">print</span><span class="p">(</span><span class="n">random</span><span class="o">.</span><span class="n">choice</span><span class="p">(</span><span class="n">pairs</span><span class="p">))</span>
</pre></div>
</div>
<p class="sphx-glr-script-out">Out:</p>
<div class="sphx-glr-script-out highlight-default"><div class="highlight"><pre><span></span><span class="n">Reading</span> <span class="n">lines</span><span class="o">...</span>
<span class="n">Read</span> <span class="mi">135842</span> <span class="n">sentence</span> <span class="n">pairs</span>
<span class="n">Trimmed</span> <span class="n">to</span> <span class="mi">10853</span> <span class="n">sentence</span> <span class="n">pairs</span>
<span class="n">Counting</span> <span class="n">words</span><span class="o">...</span>
<span class="n">Counted</span> <span class="n">words</span><span class="p">:</span>
<span class="n">fra</span> <span class="mi">4489</span>
<span class="n">eng</span> <span class="mi">2925</span>
<span class="p">[</span><span class="s1">'je redoute l examen .'</span><span class="p">,</span> <span class="s1">'i m dreading the exam .'</span><span class="p">]</span>
</pre></div>
</div>
</div>
<div class="section" id="the-seq2seq-model">
<h2>The Seq2Seq Model<a class="headerlink" href="#the-seq2seq-model" title="Permalink to this headline">¶</a></h2>
<p>A Recurrent Neural Network, or RNN, is a network that operates on a
sequence and uses its own output as input for subsequent steps.</p>
<p>A <a class="reference external" href="http://arxiv.org/abs/1409.3215">Sequence to Sequence network</a>, or
seq2seq network, or <a class="reference external" href="https://arxiv.org/pdf/1406.1078v3.pdf">Encoder Decoder
network</a>, is a model
consisting of two RNNs called the encoder and decoder. The encoder reads
an input sequence and outputs a single vector, and the decoder reads
that vector to produce an output sequence.</p>
<div class="figure">
<img alt="" src="../_images/seq2seq.png" />
</div>
<p>Unlike sequence prediction with a single RNN, where every input
corresponds to an output, the seq2seq model frees us from sequence
length and order, which makes it ideal for translation between two
languages.</p>
<p>Consider the sentence “Je ne suis pas le chat noir” → “I am not the
black cat”. Most of the words in the input sentence have a direct
translation in the output sentence, but are in slightly different
orders, e.g. “chat noir” and “black cat”. Because of the “ne/pas”
construction there is also one more word in the input sentence. It would
be difficult to produce a correct translation directly from the sequence
of input words.</p>
<p>With a seq2seq model the encoder creates a single vector which, in the
ideal case, encodes the “meaning” of the input sequence into a single
vector — a single point in some N dimensional space of sentences.</p>
<div class="section" id="the-encoder">
<h3>The Encoder<a class="headerlink" href="#the-encoder" title="Permalink to this headline">¶</a></h3>
<p>The encoder of a seq2seq network is a RNN that outputs some value for
every word from the input sentence. For every input word the encoder
outputs a vector and a hidden state, and uses the hidden state for the
next input word.</p>
<div class="figure">
<img alt="" src="../_images/encoder-network.png" />
</div>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">EncoderRNN</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">input_size</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">,</span> <span class="n">n_layers</span><span class="o">=</span><span class="mi">1</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">EncoderRNN</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span> <span class="o">=</span> <span class="n">n_layers</span>
<span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">=</span> <span class="n">hidden_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">embedding</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Embedding</span><span class="p">(</span><span class="n">input_size</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">gru</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">GRU</span><span class="p">(</span><span class="n">hidden_size</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">,</span> <span class="n">hidden</span><span class="p">):</span>
<span class="n">embedded</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">embedded</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span><span class="p">):</span>
<span class="n">output</span><span class="p">,</span> <span class="n">hidden</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">gru</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">hidden</span><span class="p">)</span>
<span class="k">return</span> <span class="n">output</span><span class="p">,</span> <span class="n">hidden</span>
<span class="k">def</span> <span class="nf">initHidden</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">))</span>
<span class="k">if</span> <span class="n">use_cuda</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span>
</pre></div>
</div>
</div>
<div class="section" id="the-decoder">
<h3>The Decoder<a class="headerlink" href="#the-decoder" title="Permalink to this headline">¶</a></h3>
<p>The decoder is another RNN that takes the encoder output vector(s) and
outputs a sequence of words to create the translation.</p>
<div class="section" id="simple-decoder">
<h4>Simple Decoder<a class="headerlink" href="#simple-decoder" title="Permalink to this headline">¶</a></h4>
<p>In the simplest seq2seq decoder we use only last output of the encoder.
This last output is sometimes called the <em>context vector</em> as it encodes
context from the entire sequence. This context vector is used as the
initial hidden state of the decoder.</p>
<p>At every step of decoding, the decoder is given an input token and
hidden state. The initial input token is the start-of-string <code class="docutils literal"><span class="pre"><SOS></span></code>
token, and the first hidden state is the context vector (the encoder’s
last hidden state).</p>
<div class="figure">
<img alt="" src="../_images/decoder-network.png" />
</div>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">DecoderRNN</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">,</span> <span class="n">output_size</span><span class="p">,</span> <span class="n">n_layers</span><span class="o">=</span><span class="mi">1</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">DecoderRNN</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span> <span class="o">=</span> <span class="n">n_layers</span>
<span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">=</span> <span class="n">hidden_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">embedding</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Embedding</span><span class="p">(</span><span class="n">output_size</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">gru</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">GRU</span><span class="p">(</span><span class="n">hidden_size</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">out</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="n">hidden_size</span><span class="p">,</span> <span class="n">output_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">softmax</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LogSoftmax</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">,</span> <span class="n">hidden</span><span class="p">):</span>
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span><span class="p">):</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="n">output</span><span class="p">)</span>
<span class="n">output</span><span class="p">,</span> <span class="n">hidden</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">gru</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">hidden</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">softmax</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">out</span><span class="p">(</span><span class="n">output</span><span class="p">[</span><span class="mi">0</span><span class="p">]))</span>
<span class="k">return</span> <span class="n">output</span><span class="p">,</span> <span class="n">hidden</span>
<span class="k">def</span> <span class="nf">initHidden</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">))</span>
<span class="k">if</span> <span class="n">use_cuda</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span>
</pre></div>
</div>
<p>I encourage you to train and observe the results of this model, but to
save space we’ll be going straight for the gold and introducing the
Attention Mechanism.</p>
</div>
<div class="section" id="attention-decoder">
<h4>Attention Decoder<a class="headerlink" href="#attention-decoder" title="Permalink to this headline">¶</a></h4>
<p>If only the context vector is passed betweeen the encoder and decoder,
that single vector carries the burden of encoding the entire sentence.</p>
<p>Attention allows the decoder network to “focus” on a different part of
the encoder’s outputs for every step of the decoder’s own outputs. First
we calculate a set of <em>attention weights</em>. These will be multiplied by
the encoder output vectors to create a weighted combination. The result
(called <code class="docutils literal"><span class="pre">attn_applied</span></code> in the code) should contain information about
that specific part of the input sequence, and thus help the decoder
choose the right output words.</p>
<div class="figure">
<img alt="" src="https://i.imgur.com/1152PYf.png" />
</div>
<p>Calculating the attention weights is done with another feed-forward
layer <code class="docutils literal"><span class="pre">attn</span></code>, using the decoder’s input and hidden state as inputs.
Because there are sentences of all sizes in the training data, to
actually create and train this layer we have to choose a maximum
sentence length (input length, for encoder outputs) that it can apply
to. Sentences of the maximum length will use all the attention weights,
while shorter sentences will only use the first few.</p>
<div class="figure">
<img alt="" src="../_images/attention-decoder-network.png" />
</div>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">AttnDecoderRNN</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_size</span><span class="p">,</span> <span class="n">output_size</span><span class="p">,</span> <span class="n">n_layers</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">dropout_p</span><span class="o">=</span><span class="mf">0.1</span><span class="p">,</span> <span class="n">max_length</span><span class="o">=</span><span class="n">MAX_LENGTH</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">AttnDecoderRNN</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">=</span> <span class="n">hidden_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">output_size</span> <span class="o">=</span> <span class="n">output_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span> <span class="o">=</span> <span class="n">n_layers</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dropout_p</span> <span class="o">=</span> <span class="n">dropout_p</span>
<span class="bp">self</span><span class="o">.</span><span class="n">max_length</span> <span class="o">=</span> <span class="n">max_length</span>
<span class="bp">self</span><span class="o">.</span><span class="n">embedding</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Embedding</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">output_size</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">attn</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">*</span> <span class="mi">2</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_length</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">attn_combine</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">*</span> <span class="mi">2</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dropout</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Dropout</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">dropout_p</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">gru</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">GRU</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">out</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">output_size</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">,</span> <span class="n">hidden</span><span class="p">,</span> <span class="n">encoder_output</span><span class="p">,</span> <span class="n">encoder_outputs</span><span class="p">):</span>
<span class="n">embedded</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span>
<span class="n">embedded</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dropout</span><span class="p">(</span><span class="n">embedded</span><span class="p">)</span>
<span class="n">attn_weights</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">softmax</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">attn</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">cat</span><span class="p">((</span><span class="n">embedded</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">hidden</span><span class="p">[</span><span class="mi">0</span><span class="p">]),</span> <span class="mi">1</span><span class="p">)))</span>
<span class="n">attn_applied</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">bmm</span><span class="p">(</span><span class="n">attn_weights</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="mi">0</span><span class="p">),</span>
<span class="n">encoder_outputs</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="mi">0</span><span class="p">))</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cat</span><span class="p">((</span><span class="n">embedded</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">attn_applied</span><span class="p">[</span><span class="mi">0</span><span class="p">]),</span> <span class="mi">1</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn_combine</span><span class="p">(</span><span class="n">output</span><span class="p">)</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span><span class="p">):</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">relu</span><span class="p">(</span><span class="n">output</span><span class="p">)</span>
<span class="n">output</span><span class="p">,</span> <span class="n">hidden</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">gru</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">hidden</span><span class="p">)</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">log_softmax</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">out</span><span class="p">(</span><span class="n">output</span><span class="p">[</span><span class="mi">0</span><span class="p">]))</span>
<span class="k">return</span> <span class="n">output</span><span class="p">,</span> <span class="n">hidden</span><span class="p">,</span> <span class="n">attn_weights</span>
<span class="k">def</span> <span class="nf">initHidden</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">))</span>
<span class="k">if</span> <span class="n">use_cuda</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last">There are other forms of attention that work around the length
limitation by using a relative position approach. Read about “local
attention” in <a class="reference external" href="https://arxiv.org/abs/1508.04025">Effective Approaches to Attention-based Neural Machine
Translation</a>.</p>
</div>
</div>
</div>
</div>
<div class="section" id="training">
<h2>Training<a class="headerlink" href="#training" title="Permalink to this headline">¶</a></h2>
<div class="section" id="preparing-training-data">
<h3>Preparing Training Data<a class="headerlink" href="#preparing-training-data" title="Permalink to this headline">¶</a></h3>
<p>To train, for each pair we will need an input tensor (indexes of the
words in the input sentence) and target tensor (indexes of the words in
the target sentence). While creating these vectors we will append the
EOS token to both sequences.</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">indexesFromSentence</span><span class="p">(</span><span class="n">lang</span><span class="p">,</span> <span class="n">sentence</span><span class="p">):</span>
<span class="k">return</span> <span class="p">[</span><span class="n">lang</span><span class="o">.</span><span class="n">word2index</span><span class="p">[</span><span class="n">word</span><span class="p">]</span> <span class="k">for</span> <span class="n">word</span> <span class="ow">in</span> <span class="n">sentence</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">' '</span><span class="p">)]</span>
<span class="k">def</span> <span class="nf">variableFromSentence</span><span class="p">(</span><span class="n">lang</span><span class="p">,</span> <span class="n">sentence</span><span class="p">):</span>
<span class="n">indexes</span> <span class="o">=</span> <span class="n">indexesFromSentence</span><span class="p">(</span><span class="n">lang</span><span class="p">,</span> <span class="n">sentence</span><span class="p">)</span>
<span class="n">indexes</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">EOS_token</span><span class="p">)</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">LongTensor</span><span class="p">(</span><span class="n">indexes</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">))</span>
<span class="k">if</span> <span class="n">use_cuda</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">result</span>
<span class="k">def</span> <span class="nf">variablesFromPair</span><span class="p">(</span><span class="n">pair</span><span class="p">):</span>
<span class="n">input_variable</span> <span class="o">=</span> <span class="n">variableFromSentence</span><span class="p">(</span><span class="n">input_lang</span><span class="p">,</span> <span class="n">pair</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="n">target_variable</span> <span class="o">=</span> <span class="n">variableFromSentence</span><span class="p">(</span><span class="n">output_lang</span><span class="p">,</span> <span class="n">pair</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
<span class="k">return</span> <span class="p">(</span><span class="n">input_variable</span><span class="p">,</span> <span class="n">target_variable</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="training-the-model">
<h3>Training the Model<a class="headerlink" href="#training-the-model" title="Permalink to this headline">¶</a></h3>
<p>To train we run the input sentence through the encoder, and keep track
of every output and the latest hidden state. Then the decoder is given
the <code class="docutils literal"><span class="pre"><SOS></span></code> token as its first input, and the last hidden state of the
encoder as its first hidden state.</p>
<p>“Teacher forcing” is the concept of using the real target outputs as
each next input, instead of using the decoder’s guess as the next input.
Using teacher forcing causes it to converge faster but <a class="reference external" href="http://minds.jacobs-university.de/sites/default/files/uploads/papers/ESNTutorialRev.pdf">when the trained
network is exploited, it may exhibit
instability</a>.</p>
<p>You can observe outputs of teacher-forced networks that read with
coherent grammar but wander far from the correct translation -
intuitively it has learned to represent the output grammar and can “pick
up” the meaning once the teacher tells it the first few words, but it
has not properly learned how to create the sentence from the translation
in the first place.</p>
<p>Because of the freedom PyTorch’s autograd gives us, we can randomly
choose to use teacher forcing or not with a simple if statement. Turn
<code class="docutils literal"><span class="pre">teacher_forcing_ratio</span></code> up to use more of it.</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">teacher_forcing_ratio</span> <span class="o">=</span> <span class="mf">0.5</span>
<span class="k">def</span> <span class="nf">train</span><span class="p">(</span><span class="n">input_variable</span><span class="p">,</span> <span class="n">target_variable</span><span class="p">,</span> <span class="n">encoder</span><span class="p">,</span> <span class="n">decoder</span><span class="p">,</span> <span class="n">encoder_optimizer</span><span class="p">,</span> <span class="n">decoder_optimizer</span><span class="p">,</span> <span class="n">criterion</span><span class="p">,</span> <span class="n">max_length</span><span class="o">=</span><span class="n">MAX_LENGTH</span><span class="p">):</span>
<span class="n">encoder_hidden</span> <span class="o">=</span> <span class="n">encoder</span><span class="o">.</span><span class="n">initHidden</span><span class="p">()</span>
<span class="n">encoder_optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">decoder_optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">input_length</span> <span class="o">=</span> <span class="n">input_variable</span><span class="o">.</span><span class="n">size</span><span class="p">()[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">target_length</span> <span class="o">=</span> <span class="n">target_variable</span><span class="o">.</span><span class="n">size</span><span class="p">()[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">encoder_outputs</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="n">max_length</span><span class="p">,</span> <span class="n">encoder</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">))</span>
<span class="n">encoder_outputs</span> <span class="o">=</span> <span class="n">encoder_outputs</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span> <span class="k">if</span> <span class="n">use_cuda</span> <span class="k">else</span> <span class="n">encoder_outputs</span>
<span class="n">loss</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">for</span> <span class="n">ei</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">input_length</span><span class="p">):</span>
<span class="n">encoder_output</span><span class="p">,</span> <span class="n">encoder_hidden</span> <span class="o">=</span> <span class="n">encoder</span><span class="p">(</span>
<span class="n">input_variable</span><span class="p">[</span><span class="n">ei</span><span class="p">],</span> <span class="n">encoder_hidden</span><span class="p">)</span>
<span class="n">encoder_outputs</span><span class="p">[</span><span class="n">ei</span><span class="p">]</span> <span class="o">=</span> <span class="n">encoder_output</span><span class="p">[</span><span class="mi">0</span><span class="p">][</span><span class="mi">0</span><span class="p">]</span>
<span class="n">decoder_input</span> <span class="o">=</span> <span class="n">Variable</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">LongTensor</span><span class="p">([[</span><span class="n">SOS_token</span><span class="p">]]))</span>
<span class="n">decoder_input</span> <span class="o">=</span> <span class="n">decoder_input</span><span class="o">.</span><span class="n">cuda</span><span class="p">()</span> <span class="k">if</span> <span class="n">use_cuda</span> <span class="k">else</span> <span class="n">decoder_input</span>
<span class="n">decoder_hidden</span> <span class="o">=</span> <span class="n">encoder_hidden</span>
<span class="n">use_teacher_forcing</span> <span class="o">=</span> <span class="bp">True</span> <span class="k">if</span> <span class="n">random</span><span class="o">.</span><span class="n">random</span><span class="p">()</span> <span class="o"><</span> <span class="n">teacher_forcing_ratio</span> <span class="k">else</span> <span class="bp">False</span>
<span class="k">if</span> <span class="n">use_teacher_forcing</span><span class="p">:</span>
<span class="c1"># Teacher forcing: Feed the target as the next input</span>
<span class="k">for</span> <span class="n">di</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">target_length</span><span class="p">):</span>