-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathseq2seq_translation_tutorial.html
1542 lines (1271 loc) · 133 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>기초부터 시작하는 NLP: Sequence to Sequence 네트워크와 Attention을 이용한 번역 — PyTorch Tutorials 1.10.2+cu102 documentation</title>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="../_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.0-beta/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.11/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/katex-math.css" type="text/css" />
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="torchtext 라이브러리로 텍스트 분류하기" href="../beginner/text_sentiment_ngrams_tutorial.html" />
<link rel="prev" title="기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 생성하기" href="char_rnn_generation_tutorial.html" />
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.kr/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.kr/get-started">시작하기</a>
</li>
<li class="active">
<a href="https://tutorials.pytorch.kr">튜토리얼</a>
</li>
<li>
<a href="https://pytorch.kr/hub">허브</a>
</li>
<li>
<a href="https://discuss.pytorch.kr">커뮤니티</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
1.10.2+cu102
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search Tutorials" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption"><span class="caption-text">파이토치(PyTorch) 레시피</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../recipes/recipes_index.html">모든 레시피 보기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../prototype/prototype_index.html">모든 프로토타입 레시피 보기</a></li>
</ul>
<p class="caption"><span class="caption-text">파이토치(PyTorch) 시작하기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/intro.html">파이토치(PyTorch) 기본 익히기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/quickstart_tutorial.html">빠른 시작(Quickstart)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/tensorqs_tutorial.html">텐서(Tensor)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/data_tutorial.html">Dataset과 DataLoader</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/transforms_tutorial.html">변형(Transform)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/buildmodel_tutorial.html">신경망 모델 구성하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/autogradqs_tutorial.html"><code class="docutils literal notranslate"><span class="pre">torch.autograd</span></code>를 사용한 자동 미분</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/optimization_tutorial.html">모델 매개변수 최적화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/saveloadrun_tutorial.html">모델 저장하고 불러오기</a></li>
</ul>
<p class="caption"><span class="caption-text">Introduction to PyTorch on YouTube</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt.html">Introduction to PyTorch - YouTube Series</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/introyt1_tutorial.html">Introduction to PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensors_deeper_tutorial.html">Introduction to PyTorch Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/autogradyt_tutorial.html">The Fundamentals of Autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/modelsyt_tutorial.html">Building Models with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensorboardyt_tutorial.html">PyTorch TensorBoard Support</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/trainingyt.html">Training with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/captumyt.html">Model Understanding with Captum</a></li>
</ul>
<p class="caption"><span class="caption-text">파이토치(PyTorch) 배우기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">PyTorch로 딥러닝하기: 60분만에 끝장내기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">예제로 배우는 파이토치(PyTorch)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/nn_tutorial.html"><cite>torch.nn</cite> 이 <em>실제로</em> 무엇인가요?</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard_tutorial.html">TensorBoard로 모델, 데이터, 학습 시각화하기</a></li>
</ul>
<p class="caption"><span class="caption-text">이미지/비디오</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision_tutorial.html">TorchVision 객체 검출 미세조정(Finetuning) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">컴퓨터 비전(Vision)을 위한 전이학습(Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/fgsm_tutorial.html">적대적 예제 생성(Adversarial Example Generation)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dcgan_faces_tutorial.html">DCGAN 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">배포를 위한 비전 트랜스포머(Vision Transformer) 모델 최적화하기</a></li>
</ul>
<p class="caption"><span class="caption-text">오디오</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_io_tutorial.html">Audio I/O</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_resampling_tutorial.html">Audio Resampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_data_augmentation_tutorial.html">Audio Data Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_extractions_tutorial.html">Audio Feature Extractions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_augmentation_tutorial.html">Audio Feature Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_datasets_tutorial.html">Audio Datasets</a></li>
<li class="toctree-l1"><a class="reference internal" href="speech_recognition_pipeline_tutorial.html">Speech Recognition with Wav2Vec2</a></li>
<li class="toctree-l1"><a class="reference internal" href="speech_command_classification_with_torchaudio_tutorial.html">Speech Command Classification with torchaudio</a></li>
<li class="toctree-l1"><a class="reference internal" href="text_to_speech_with_torchaudio.html">Text-to-speech with torchaudio</a></li>
<li class="toctree-l1"><a class="reference internal" href="forced_alignment_with_torchaudio_tutorial.html">Forced Alignment with Wav2Vec2</a></li>
</ul>
<p class="caption"><span class="caption-text">텍스트</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../beginner/transformer_tutorial.html">nn.Transformer 와 TorchText 로 시퀀스-투-시퀀스(Sequence-to-Sequence) 모델링하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_classification_tutorial.html">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 분류하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_generation_tutorial.html">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 생성하기</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">기초부터 시작하는 NLP: Sequence to Sequence 네트워크와 Attention을 이용한 번역</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/text_sentiment_ngrams_tutorial.html">torchtext 라이브러리로 텍스트 분류하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/translation_transformer.html">nn.Transformer와 torchtext로 언어 번역하기</a></li>
</ul>
<p class="caption"><span class="caption-text">강화학습</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_q_learning.html">강화 학습 (DQN) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="mario_rl_tutorial.html">Train a Mario-playing RL Agent</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch 모델을 프로덕션 환경에 배포하기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="flask_rest_api_tutorial.html">Flask를 사용하여 Python에서 PyTorch를 REST API로 배포하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/Intro_to_TorchScript_tutorial.html">TorchScript 소개</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_export.html">C++에서 TorchScript 모델 로딩하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/super_resolution_with_onnxruntime.html">(선택) PyTorch 모델을 ONNX으로 변환하고 ONNX 런타임에서 실행하기</a></li>
</ul>
<p class="caption"><span class="caption-text">Code Transforms with FX</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="fx_conv_bn_fuser.html">(베타) FX에서 합성곱/배치 정규화(Convolution/Batch Norm) 결합기(Fuser) 만들기</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx_profiling_tutorial.html">(beta) Building a Simple CPU Performance Profiler with FX</a></li>
</ul>
<p class="caption"><span class="caption-text">프론트엔드 API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="memory_format_tutorial.html">(베타) PyTorch를 사용한 Channels Last 메모리 형식</a></li>
<li class="toctree-l1"><a class="reference internal" href="forward_ad_usage.html">Forward-mode Automatic Differentiation (Beta)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_frontend.html">PyTorch C++ 프론트엔드 사용하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch-script-parallelism.html">TorchScript의 동적 병렬 처리(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_autograd.html">C++ 프론트엔드의 자동 미분 (autograd)</a></li>
</ul>
<p class="caption"><span class="caption-text">PyTorch 확장하기</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="custom_function_double_backward_tutorial.html">Double Backward with Custom Functions</a></li>
<li class="toctree-l1"><a class="reference internal" href="custom_function_conv_bn_tutorial.html">Fusing Convolution and Batch Norm using Custom Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_extension.html">Custom C++ and CUDA Extensions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_ops.html">Extending TorchScript with Custom C++ Operators</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_classes.html">커스텀 C++ 클래스로 TorchScript 확장하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dispatcher.html">Registering a Dispatched Operator in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/extend_dispatcher.html">Extending dispatcher for a new backend in C++</a></li>
</ul>
<p class="caption"><span class="caption-text">모델 최적화</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch 모듈 프로파일링 하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard_profiler_tutorial.html">PyTorch Profiler With TensorBoard</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hyperparameter_tuning_tutorial.html">Hyperparameter tuning with Ray Tune</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">배포를 위한 비전 트랜스포머(Vision Transformer) 모델 최적화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="parametrizations.html">Parametrizations Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="pruning_tutorial.html">가지치기 기법(Pruning) 튜토리얼</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dynamic_quantization_tutorial.html">(베타) LSTM 기반 단어 단위 언어 모델의 동적 양자화</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_bert_tutorial.html">(베타) BERT 모델 동적 양자화하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantized_transfer_learning_tutorial.html">(베타) 컴퓨터 비전 튜토리얼을 위한 양자화된 전이학습(Quantized Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/static_quantization_tutorial.html">(beta) Static Quantization with Eager Mode in PyTorch</a></li>
</ul>
<p class="caption"><span class="caption-text">병렬 및 분산 학습</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dist_overview.html">PyTorch Distributed Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_parallel_tutorial.html">단일 머신을 사용한 모델 병렬화 모범 사례</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_tutorial.html">분산 데이터 병렬 처리 시작하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_tuto.html">PyTorch로 분산 어플리케이션 개발하기</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_tutorial.html">Getting Started with Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_param_server_tutorial.html">Implementing a Parameter Server Using Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_pipeline_parallel_tutorial.html">Distributed Pipeline Parallelism Using RPC</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_async_execution.html">Implementing Batch RPC Processing Using Asynchronous Executions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/rpc_ddp_tutorial.html">분산 데이터 병렬(DDP)과 분산 RPC 프레임워크 결합</a></li>
<li class="toctree-l1"><a class="reference internal" href="pipeline_tutorial.html">파이프라인 병렬화로 트랜스포머 모델 학습시키기</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/ddp_pipeline.html">분산 데이터 병렬 처리와 병렬 처리 파이프라인을 사용한 트랜스포머 모델 학습</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/generic_join.html">Distributed Training with Uneven Inputs Using the Join Context Manager</a></li>
</ul>
<p class="caption"><span class="caption-text">Mobile</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_ios.html">iOS에서의 이미지 분할 DeepLapV3</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_android.html">안드로이드에서의 이미지 분할 DeepLapV3</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../index.html">
Tutorials
</a> >
</li>
<li>기초부터 시작하는 NLP: Sequence to Sequence 네트워크와 Attention을 이용한 번역</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/intermediate/seq2seq_translation_tutorial.rst.txt" rel="nofollow"><img src="../_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="pytorch-call-to-action-links">
<div id="tutorial-type">intermediate/seq2seq_translation_tutorial</div>
<div id="google-colab-link">
<img class="call-to-action-img" src="../_static/images/pytorch-colab.svg"/>
<div class="call-to-action-desktop-view">Run in Google Colab</div>
<div class="call-to-action-mobile-view">Colab</div>
</div>
<div id="download-notebook-link">
<img class="call-to-action-notebook-img" src="../_static/images/pytorch-download.svg"/>
<div class="call-to-action-desktop-view">Download Notebook</div>
<div class="call-to-action-mobile-view">Notebook</div>
</div>
<div id="github-view-link">
<img class="call-to-action-img" src="../_static/images/pytorch-github.svg"/>
<div class="call-to-action-desktop-view">View on GitHub</div>
<div class="call-to-action-mobile-view">GitHub</div>
</div>
</div>
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="sphx-glr-download-link-note admonition note">
<p class="first admonition-title">Note</p>
<p class="last">Click <a class="reference internal" href="#sphx-glr-download-intermediate-seq2seq-translation-tutorial-py"><span class="std std-ref">here</span></a> to download the full example code</p>
</div>
<div class="sphx-glr-example-title section" id="nlp-sequence-to-sequence-attention">
<span id="sphx-glr-intermediate-seq2seq-translation-tutorial-py"></span><h1>기초부터 시작하는 NLP: Sequence to Sequence 네트워크와 Attention을 이용한 번역<a class="headerlink" href="#nlp-sequence-to-sequence-attention" title="Permalink to this headline">¶</a></h1>
<dl class="docutils">
<dt><strong>Author</strong>: <a class="reference external" href="https://github.com/spro/practical-pytorch">Sean Robertson</a></dt>
<dd><strong>번역</strong>: <a class="reference external" href="https://github.com/adonisues">황성수</a></dd>
</dl>
<p>이 튜토리얼은 “기초부터 시작하는 NLP”의 세번째이자 마지막 편으로, NLP 모델링 작업을
위한 데이터 전처리에 사용할 자체 클래스와 함수들을 작성해보겠습니다.
이 튜토리얼을 마친 뒤에는 <cite>torchtext</cite> 가 어떻게 지금까지의 튜토리얼들에서의
전처리 과정을 다루는지를 이후 튜토리얼들에서 배울 수 있습니다.</p>
<p>이 프로젝트에서는 신경망이 불어를 영어로 번역하도록 가르칠 예정입니다.</p>
<div class="highlight-default notranslate"><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>… 성공율은 변할 수 있습니다.</p>
<p>하나의 시퀀스를 다른 시퀀스로 바꾸는 두개의 RNN이 함께 동작하는
<a class="reference external" href="https://arxiv.org/abs/1409.3215">sequence to sequence network</a> 의 간단하지만 강력한 아이디어가
이것(번역)을 가능하게 합니다. 인코더 네트워크는 입력 시퀀스를 벡터로 압축하고,
디코더 네트워크는 해당 벡터를 새로운 시퀀스로 펼칩니다.</p>
<div class="figure">
<img alt="" src="../_images/seq2seq.png" />
</div>
<p>이 모델을 개선하기 위해 <a class="reference external" href="https://arxiv.org/abs/1409.0473">Attention Mechanism</a> 을
사용하면 디코더가 입력 시퀀스의 특정 범위에 집중할 수 있도록 합니다.</p>
<p><strong>추천 자료:</strong></p>
<p>최소한 Pytorch를 설치했고, Python을 알고, Tensor를 이해한다고 가정합니다.:</p>
<ul class="simple">
<li><a class="reference external" href="http://pytorch.org/">http://pytorch.org/</a> 설치 안내를 위한 자료</li>
<li><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html"><span class="doc">PyTorch로 딥러닝하기: 60분만에 끝장내기</span></a> 일반적인 PyTorch 시작을 위한 자료</li>
<li><a class="reference internal" href="../beginner/pytorch_with_examples.html"><span class="doc">예제로 배우는 파이토치(PyTorch)</span></a> 넓고 깊은 통찰을 위한 자료</li>
<li><a class="reference internal" href="../beginner/former_torchies_tutorial.html"><span class="doc">Torch 사용자를 위한 PyTorch</span></a> 이전 Lua Torch 사용자를 위한 자료</li>
</ul>
<p>Sequence to Sequence 네트워크와 동작 방법에 관해서 아는 것은 유용합니다:</p>
<ul class="simple">
<li><a class="reference external" href="https://arxiv.org/abs/1406.1078">Learning Phrase Representations using RNN Encoder-Decoder for
Statistical Machine Translation</a></li>
<li><a class="reference external" href="https://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="https://arxiv.org/abs/1506.05869">A Neural Conversational Model</a></li>
</ul>
<p>이전 튜토리얼에 있는
<a class="reference internal" href="char_rnn_classification_tutorial.html"><span class="doc">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 분류하기</span></a>
와 <a class="reference internal" href="char_rnn_generation_tutorial.html"><span class="doc">기초부터 시작하는 NLP: 문자-단위 RNN으로 이름 생성하기</span></a> 는
각각 인코더, 디코더 모델과 비슷한 컨센을 가지기 때문에 도움이 됩니다.</p>
<p><strong>요구 사항</strong></p>
<div class="highlight-default notranslate"><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="k">as</span> <span class="nn">nn</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="k">as</span> <span class="nn">F</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s2">"cuda"</span> <span class="k">if</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> <span class="k">else</span> <span class="s2">"cpu"</span><span class="p">)</span>
</pre></div>
</div>
<div class="section" id="id2">
<h2>데이터 파일 로딩<a class="headerlink" href="#id2" title="Permalink to this headline">¶</a></h2>
<p>이 프로젝트의 데이터는 수천 개의 영어-프랑스어 번역 쌍입니다.</p>
<p><a class="reference external" href="https://opendata.stackexchange.com/questions/3888/dataset-of-sentences-translated-into-many-languages">Open Data Stack Exchange</a>
에 관한 이 질문은 <a class="reference external" href="https://tatoeba.org/eng/downloads">https://tatoeba.org/eng/downloads</a> 에서 다운 로드가 가능한
공개 번역 사이트 <a class="reference external" href="https://tatoeba.org/">https://tatoeba.org/</a> 를 알려 주었습니다. 더 나은 방법으로
언어 쌍을 개별 텍스트 파일로 분할하는 추가 작업을 수행한
<a class="reference external" href="https://www.manythings.org/anki/">https://www.manythings.org/anki/</a> 가 있습니다:</p>
<p>영어-프랑스어 쌍이 너무 커서 저장소에 포함 할 수 없기 때문에
계속하기 전에 <code class="docutils literal notranslate"><span class="pre">data/eng-fra.txt</span></code> 로 다운로드하십시오.
이 파일은 탭으로 구분된 번역 쌍 목록입니다:</p>
<div class="highlight-default notranslate"><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">J</span><span class="s1">'ai froid.</span>
</pre></div>
</div>
<div class="admonition note">
<p class="first admonition-title">Note</p>
<p class="last"><a class="reference external" href="https://download.pytorch.org/tutorial/data.zip">여기</a>
에서 데이터를 다운 받고 현재 디렉토리에 압축을 푸십시오.</p>
</div>
<p>문자 단위 RNN 튜토리얼에서 사용된 문자 인코딩과 유사하게, 언어의 각
단어들을 One-Hot 벡터 또는 그 단어의 주소에만 단 하나의 1을 제외하고
모두 0인 큰 벡터로 표현합니다. 한 가지 언어에 있는 수십 개의 문자와
달리 번역에는 아주 많은 단어들이 있기 때문에 인코딩 벡터는 매우 더 큽니다.
그러나 우리는 약간의 트릭를 써서 언어 당 수천 단어 만
사용하도록 데이터를 다듬을 것입니다.</p>
<div class="figure">
<img alt="" src="../_images/word-encoding.png" />
</div>
<p>나중에 네트워크의 입력 및 목표로 사용하려면 단어 당 고유 번호가
필요합니다. 이 모든 것을 추적하기 위해 우리는
단어→색인(<code class="docutils literal notranslate"><span class="pre">word2index</span></code>)과 색인→단어(<code class="docutils literal notranslate"><span class="pre">index2word</span></code>) 사전,
그리고 나중에 희귀 단어를 대체하는데 사용할 각 단어의 빈도
<code class="docutils literal notranslate"><span class="pre">word2count</span></code> 를 가진 <code class="docutils literal notranslate"><span class="pre">Lang</span></code> 이라는 헬퍼 클래스를 사용합니다.</p>
<div class="highlight-default notranslate"><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"># SOS 와 EOS 포함</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>파일은 모두 유니 코드로 되어있어 간단하게하기 위해 유니 코드 문자를
ASCII로 변환하고, 모든 문자를 소문자로 만들고, 대부분의 구두점을
지워줍니다.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># 유니 코드 문자열을 일반 ASCII로 변환하십시오.</span>
<span class="c1"># https://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"># 소문자, 다듬기, 그리고 문자가 아닌 문자 제거</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 notranslate"><span class="pre">reverse</span></code>
flag to reverse the pairs.</p>
<div class="highlight-default notranslate"><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="kc">False</span><span class="p">):</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Reading lines..."</span><span class="p">)</span>
<span class="c1"># 파일을 읽고 줄로 분리</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"># 모든 줄을 쌍으로 분리하고 정규화</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"># 쌍을 뒤집고, Lang 인스턴스 생성</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><em>많은</em> 예제 문장이 있고 신속하게 학습하기를 원하기 때문에
비교적 짧고 간단한 문장으로만 데이터 셋을 정리할 것입니다. 여기서
최대 길이는 10 단어 (종료 문장 부호 포함)이며 “I am” 또는
“He is” 등의 형태로 번역되는 문장으로 필터링됩니다.(이전에
아포스트로피는 대체 됨)</p>
<div class="highlight-default notranslate"><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>데이터 준비를 위한 전체 과정:</p>
<ul class="simple">
<li>텍스트 파일을 읽고 줄로 분리하고, 줄을 쌍으로 분리합니다.</li>
<li>텍스트를 정규화 하고 길이와 내용으로 필터링 합니다.</li>
<li>쌍을 이룬 문장들로 단어 리스트를 생성합니다.</li>
</ul>
<div class="highlight-default notranslate"><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="kc">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="nb">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="nb">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="nb">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="nb">print</span><span class="p">(</span><span class="s2">"Counted words:"</span><span class="p">)</span>
<span class="nb">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="nb">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="kc">True</span><span class="p">)</span>
<span class="nb">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-none notranslate"><div class="highlight"><pre><span></span>Reading lines...
Read 135842 sentence pairs
Trimmed to 10599 sentence pairs
Counting words...
Counted words:
fra 4345
eng 2803
['je vais maintenant raccrocher .', 'i m going to hang up now .']
</pre></div>
</div>
</div>
<div class="section" id="seq2seq">
<h2>Seq2Seq 모델<a class="headerlink" href="#seq2seq" title="Permalink to this headline">¶</a></h2>
<p>Recurrent Neural Network(RNN)는 시퀀스에서 작동하고 다음 단계의
입력으로 자신의 출력을 사용하는 네트워크입니다.</p>
<p><a class="reference external" href="https://arxiv.org/abs/1409.3215">Sequence to Sequence network</a>, 또는
Seq2Seq 네트워크, 또는 <a class="reference external" href="https://arxiv.org/pdf/1406.1078v3.pdf">Encoder Decoder
network</a> 는 인코더 및
디코더라고 하는 두 개의 RNN으로 구성된 모델입니다.
인코더는 입력 시퀀스를 읽고 단일 벡터를 출력하고,
디코더는 해당 벡터를 읽어 출력 시퀀스를 생성합니다.</p>
<div class="figure">
<img alt="" src="../_images/seq2seq.png" />
</div>
<p>모든 입력에 해당하는 출력이 있는 단일 RNN의 시퀀스 예측과 달리
Seq2Seq 모델은 시퀀스 길이와 순서를 자유롭게하기 때문에
두 언어 사이의 번역에 이상적입니다.</p>
<p>다음 문장 “Je ne suis pas le chat noir” → “I am not the black cat”
를 살펴 봅시다. 입력 문장의 단어 대부분은 출력 문장에서
직역(“chat noir” 와 “black cat”)되지만 약간 다른 순서도 있습니다.
“ne/pas” 구조로 인해 입력 문장에 단어가 하나 더 있습니다.
입력 단어의 시퀀스를 직역해서 정확한 번역을 만드는
것은 어려울 것입니다.</p>
<p>Seq2Seq 모델을 사용하면 인코더는 하나의 벡터를 생성합니다.
이상적인 경우에 입력 시퀀스의 “의미”를 문장의 N 차원 공간에 있는
단일 지점인 단일 벡터으로 인코딩합니다.</p>
<div class="section" id="id4">
<h3>인코더<a class="headerlink" href="#id4" title="Permalink to this headline">¶</a></h3>
<p>Seq2Seq 네트워크의 인코더는 입력 문장의 모든 단어에 대해 어떤 값을
출력하는 RNN입니다. 모든 입력 단어에 대해 인코더는 벡터와
은닉 상태를 출력하고 다음 입력 단어를 위해 그 은닉 상태를 사용합니다.</p>
<div class="figure">
<img alt="" src="../_images/encoder-network.png" />
</div>
<div class="highlight-default notranslate"><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="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">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="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="k">return</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="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id5">
<h3>디코더<a class="headerlink" href="#id5" title="Permalink to this headline">¶</a></h3>
<p>디코더는 인코더 출력 벡터를 받아서 번역을 생성하기 위한 단어 시퀀스를
출력합니다.</p>
<div class="section" id="id6">
<h4>간단한 디코더<a class="headerlink" href="#id6" title="Permalink to this headline">¶</a></h4>
<p>가장 간단한 Seq2Seq 디코더는 인코더의 마지막 출력만을 이용합니다.
이 마지막 출력은 전체 시퀀스에서 문맥을 인코드하기 때문에
<em>문맥 벡터(context vector)</em> 로 불립니다. 이 문맥 벡터는 디코더의 초기 은닉 상태로
사용 됩니다.</p>
<p>디코딩의 매 단계에서 디코더에게 입력 토큰과 은닉 상태가 주어집니다.
초기 입력 토큰은 문자열-시작 (start-of-string) <code class="docutils literal notranslate"><span class="pre"><SOS></span></code> 토큰이고,
첫 은닉 상태는 문맥 벡터(인코더의 마지막 은닉 상태) 입니다.</p>
<div class="figure">
<img alt="" src="../_images/decoder-network.png" />
</div>
<div class="highlight-default notranslate"><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="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">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="n">dim</span><span class="o">=</span><span class="mi">1</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="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="k">return</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="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</span>
</pre></div>
</div>
<p>이 모델의 결과를 학습하고 관찰하는 것을 권장하지만,
공간을 절약하기 위해 최종 목적지로 바로 이동해서
Attention 메커니즘을 소개 할 것입니다.</p>
</div>
<div class="section" id="attention">
<h4>Attention 디코더<a class="headerlink" href="#attention" title="Permalink to this headline">¶</a></h4>
<p>문맥 벡터만 인코더와 디코더 사이로 전달 된다면, 단일 벡터가 전체 문장을
인코딩 해야하는 부담을 가지게 됩니다.</p>
<p>Attention은 디코더 네트워크가 자기 출력의 모든 단계에서 인코더 출력의
다른 부분에 “집중” 할 수 있게 합니다. 첫째 <em>Attention 가중치</em> 의 세트를
계산합니다. 이것은 가중치 조합을 만들기 위해서 인코더 출력 벡터와
곱해집니다. 그 결과(코드에서 <code class="docutils literal notranslate"><span class="pre">attn_applied</span></code>)는 입력 시퀀스의
특정 부분에 관한 정보를 포함해야하고 따라서 디코더가 알맞은 출력
단어를 선택하는 것을 도와줍니다.</p>
<div class="figure">
<img alt="" src="https://i.imgur.com/1152PYf.png" />
</div>
<p>어텐션 가중치 계산은 디코더의 입력 및 은닉 상태를 입력으로
사용하는 다른 feed-forwad 계층인 <code class="docutils literal notranslate"><span class="pre">attn</span></code> 으로 수행됩니다.
학습 데이터에는 모든 크기의 문장이 있기 때문에 이 계층을 실제로
만들고 학습시키려면 적용 할 수 있는 최대 문장 길이 (인코더 출력을 위한 입력 길이)를
선택해야 합니다. 최대 길이의 문장은 모든 Attention 가중치를 사용하지만
더 짧은 문장은 처음 몇 개만 사용합니다.</p>
<div class="figure">
<img alt="" src="../_images/attention-decoder-network.png" />
</div>
<div class="highlight-default notranslate"><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">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">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_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">dim</span><span class="o">=</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="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="n">dim</span><span class="o">=</span><span class="mi">1</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="k">return</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="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</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="id7">
<h2>학습<a class="headerlink" href="#id7" title="Permalink to this headline">¶</a></h2>
<div class="section" id="id8">
<h3>학습 데이터 준비<a class="headerlink" href="#id8" title="Permalink to this headline">¶</a></h3>
<p>학습을 위해서, 각 쌍마다 입력 Tensor(입력 문장의 단어 주소)와
목표 Tensor(목표 문장의 단어 주소)가 필요합니다. 이 벡터들을
생성하는 동안 두 시퀀스에 EOS 토큰을 추가 합니다.</p>
<div class="highlight-default notranslate"><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">tensorFromSentence</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="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">indexes</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">long</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device</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">def</span> <span class="nf">tensorsFromPair</span><span class="p">(</span><span class="n">pair</span><span class="p">):</span>
<span class="n">input_tensor</span> <span class="o">=</span> <span class="n">tensorFromSentence</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_tensor</span> <span class="o">=</span> <span class="n">tensorFromSentence</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_tensor</span><span class="p">,</span> <span class="n">target_tensor</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id9">
<h3>모델 학습<a class="headerlink" href="#id9" title="Permalink to this headline">¶</a></h3>
<p>학습을 위해서 인코더에 입력 문장을 넣고 모든 출력과 최신 은닉 상태를
추적합니다. 그런 다음 디코더에 첫 번째 입력으로 <code class="docutils literal notranslate"><span class="pre"><SOS></span></code> 토큰과
인코더의 마지막 은닉 상태가 첫번쩨 은닉 상태로 제공됩니다.</p>
<p>“Teacher forcing”은 다음 입력으로 디코더의 예측을 사용하는 대신
실제 목표 출력을 다음 입력으로 사용하는 컨셉입니다.
“Teacher forcing”을 사용하면 수렴이 빨리되지만 <a class="reference external" href="http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.378.4095&rep=rep1&type=pdf">학습된 네트워크가
잘못 사용될 때 불안정성을 보입니다.</a>.</p>
<p>Teacher-forced 네트워크의 출력이 일관된 문법으로 읽지만 정확한
번역과는 거리가 멀다는 것을 볼 수 있습니다. 직관적으로 출력 문법을
표현하는 법을 배우고 교사가 처음 몇 단어를 말하면 의미를 “선택” 할 수 있지만,
번역에서 처음으로 문장을 만드는 법은 잘 배우지 못합니다.</p>
<p>PyTorch의 autograd 가 제공하는 자유 덕분에 간단한 If 문으로
Teacher Forcing을 사용할지 아니면 사용하지 않을지를 선택할 수 있습니다.
더 많이 사용하려면 <code class="docutils literal notranslate"><span class="pre">teacher_forcing_ratio</span></code> 를 확인하십시오.</p>
<div class="highlight-default notranslate"><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_tensor</span><span class="p">,</span> <span class="n">target_tensor</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_tensor</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_tensor</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">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">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</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_tensor</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">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([[</span><span class="n">SOS_token</span><span class="p">]],</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</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="kc">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="kc">False</span>
<span class="k">if</span> <span class="n">use_teacher_forcing</span><span class="p">:</span>
<span class="c1"># Teacher forcing 포함: 목표를 다음 입력으로 전달</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>
<span class="n">decoder_output</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">decoder_attention</span> <span class="o">=</span> <span class="n">decoder</span><span class="p">(</span>
<span class="n">decoder_input</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">encoder_outputs</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">+=</span> <span class="n">criterion</span><span class="p">(</span><span class="n">decoder_output</span><span class="p">,</span> <span class="n">target_tensor</span><span class="p">[</span><span class="n">di</span><span class="p">])</span>
<span class="n">decoder_input</span> <span class="o">=</span> <span class="n">target_tensor</span><span class="p">[</span><span class="n">di</span><span class="p">]</span> <span class="c1"># Teacher forcing</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># Teacher forcing 미포함: 자신의 예측을 다음 입력으로 사용</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>
<span class="n">decoder_output</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">decoder_attention</span> <span class="o">=</span> <span class="n">decoder</span><span class="p">(</span>
<span class="n">decoder_input</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">encoder_outputs</span><span class="p">)</span>
<span class="n">topv</span><span class="p">,</span> <span class="n">topi</span> <span class="o">=</span> <span class="n">decoder_output</span><span class="o">.</span><span class="n">topk</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="n">decoder_input</span> <span class="o">=</span> <span class="n">topi</span><span class="o">.</span><span class="n">squeeze</span><span class="p">()</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span> <span class="c1"># 입력으로 사용할 부분을 히스토리에서 분리</span>
<span class="n">loss</span> <span class="o">+=</span> <span class="n">criterion</span><span class="p">(</span><span class="n">decoder_output</span><span class="p">,</span> <span class="n">target_tensor</span><span class="p">[</span><span class="n">di</span><span class="p">])</span>
<span class="k">if</span> <span class="n">decoder_input</span><span class="o">.</span><span class="n">item</span><span class="p">()</span> <span class="o">==</span> <span class="n">EOS_token</span><span class="p">:</span>
<span class="k">break</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">encoder_optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="n">decoder_optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="k">return</span> <span class="n">loss</span><span class="o">.</span><span class="n">item</span><span class="p">()</span> <span class="o">/</span> <span class="n">target_length</span>
</pre></div>
</div>
<p>이것은 현재 시간과 진행률%을 고려해 경과된 시간과 남은 예상
시간을 출력하는 헬퍼 함수입니다.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">time</span>
<span class="kn">import</span> <span class="nn">math</span>
<span class="k">def</span> <span class="nf">asMinutes</span><span class="p">(</span><span class="n">s</span><span class="p">):</span>
<span class="n">m</span> <span class="o">=</span> <span class="n">math</span><span class="o">.</span><span class="n">floor</span><span class="p">(</span><span class="n">s</span> <span class="o">/</span> <span class="mi">60</span><span class="p">)</span>
<span class="n">s</span> <span class="o">-=</span> <span class="n">m</span> <span class="o">*</span> <span class="mi">60</span>
<span class="k">return</span> <span class="s1">'</span><span class="si">%d</span><span class="s1">m </span><span class="si">%d</span><span class="s1">s'</span> <span class="o">%</span> <span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="n">s</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">timeSince</span><span class="p">(</span><span class="n">since</span><span class="p">,</span> <span class="n">percent</span><span class="p">):</span>
<span class="n">now</span> <span class="o">=</span> <span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span>
<span class="n">s</span> <span class="o">=</span> <span class="n">now</span> <span class="o">-</span> <span class="n">since</span>
<span class="n">es</span> <span class="o">=</span> <span class="n">s</span> <span class="o">/</span> <span class="p">(</span><span class="n">percent</span><span class="p">)</span>
<span class="n">rs</span> <span class="o">=</span> <span class="n">es</span> <span class="o">-</span> <span class="n">s</span>
<span class="k">return</span> <span class="s1">'</span><span class="si">%s</span><span class="s1"> (- </span><span class="si">%s</span><span class="s1">)'</span> <span class="o">%</span> <span class="p">(</span><span class="n">asMinutes</span><span class="p">(</span><span class="n">s</span><span class="p">),</span> <span class="n">asMinutes</span><span class="p">(</span><span class="n">rs</span><span class="p">))</span>
</pre></div>
</div>
<p>전체 학습 과정은 다음과 같습니다:</p>
<ul class="simple">
<li>타이머 시작</li>
<li>optimizers와 criterion 초기화</li>
<li>학습 쌍의 세트 생성</li>
<li>도식화를 위한 빈 손실 배열 시작</li>
</ul>
<p>그런 다음 우리는 여러 번 <code class="docutils literal notranslate"><span class="pre">train</span></code> 을 호출하며 때로는 진행률
(예제의 %, 현재까지의 예상 시간)과 평균 손실을 출력합니다.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">trainIters</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">n_iters</span><span class="p">,</span> <span class="n">print_every</span><span class="o">=</span><span class="mi">1000</span><span class="p">,</span> <span class="n">plot_every</span><span class="o">=</span><span class="mi">100</span><span class="p">,</span> <span class="n">learning_rate</span><span class="o">=</span><span class="mf">0.01</span><span class="p">):</span>
<span class="n">start</span> <span class="o">=</span> <span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span>
<span class="n">plot_losses</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">print_loss_total</span> <span class="o">=</span> <span class="mi">0</span> <span class="c1"># print_every 마다 초기화</span>
<span class="n">plot_loss_total</span> <span class="o">=</span> <span class="mi">0</span> <span class="c1"># plot_every 마다 초기화</span>
<span class="n">encoder_optimizer</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">encoder</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="n">learning_rate</span><span class="p">)</span>
<span class="n">decoder_optimizer</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">decoder</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="n">learning_rate</span><span class="p">)</span>
<span class="n">training_pairs</span> <span class="o">=</span> <span class="p">[</span><span class="n">tensorsFromPair</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>
<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="n">n_iters</span><span class="p">)]</span>
<span class="n">criterion</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">NLLLoss</span><span class="p">()</span>
<span class="k">for</span> <span class="nb">iter</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">n_iters</span> <span class="o">+</span> <span class="mi">1</span><span class="p">):</span>
<span class="n">training_pair</span> <span class="o">=</span> <span class="n">training_pairs</span><span class="p">[</span><span class="nb">iter</span> <span class="o">-</span> <span class="mi">1</span><span class="p">]</span>
<span class="n">input_tensor</span> <span class="o">=</span> <span class="n">training_pair</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">target_tensor</span> <span class="o">=</span> <span class="n">training_pair</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">train</span><span class="p">(</span><span class="n">input_tensor</span><span class="p">,</span> <span class="n">target_tensor</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">print_loss_total</span> <span class="o">+=</span> <span class="n">loss</span>
<span class="n">plot_loss_total</span> <span class="o">+=</span> <span class="n">loss</span>
<span class="k">if</span> <span class="nb">iter</span> <span class="o">%</span> <span class="n">print_every</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">print_loss_avg</span> <span class="o">=</span> <span class="n">print_loss_total</span> <span class="o">/</span> <span class="n">print_every</span>
<span class="n">print_loss_total</span> <span class="o">=</span> <span class="mi">0</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'</span><span class="si">%s</span><span class="s1"> (</span><span class="si">%d</span><span class="s1"> </span><span class="si">%d%%</span><span class="s1">) </span><span class="si">%.4f</span><span class="s1">'</span> <span class="o">%</span> <span class="p">(</span><span class="n">timeSince</span><span class="p">(</span><span class="n">start</span><span class="p">,</span> <span class="nb">iter</span> <span class="o">/</span> <span class="n">n_iters</span><span class="p">),</span>
<span class="nb">iter</span><span class="p">,</span> <span class="nb">iter</span> <span class="o">/</span> <span class="n">n_iters</span> <span class="o">*</span> <span class="mi">100</span><span class="p">,</span> <span class="n">print_loss_avg</span><span class="p">))</span>
<span class="k">if</span> <span class="nb">iter</span> <span class="o">%</span> <span class="n">plot_every</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">plot_loss_avg</span> <span class="o">=</span> <span class="n">plot_loss_total</span> <span class="o">/</span> <span class="n">plot_every</span>
<span class="n">plot_losses</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">plot_loss_avg</span><span class="p">)</span>
<span class="n">plot_loss_total</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">showPlot</span><span class="p">(</span><span class="n">plot_losses</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id10">
<h3>결과 도식화<a class="headerlink" href="#id10" title="Permalink to this headline">¶</a></h3>
<p>matplotlib로 학습 중에 저장된 손실 값 <code class="docutils literal notranslate"><span class="pre">plot_losses</span></code> 의 배열을
사용하여 도식화합니다.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="n">plt</span><span class="o">.</span><span class="n">switch_backend</span><span class="p">(</span><span class="s1">'agg'</span><span class="p">)</span>
<span class="kn">import</span> <span class="nn">matplotlib.ticker</span> <span class="k">as</span> <span class="nn">ticker</span>
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="k">def</span> <span class="nf">showPlot</span><span class="p">(</span><span class="n">points</span><span class="p">):</span>
<span class="n">plt</span><span class="o">.</span><span class="n">figure</span><span class="p">()</span>
<span class="n">fig</span><span class="p">,</span> <span class="n">ax</span> <span class="o">=</span> <span class="n">plt</span><span class="o">.</span><span class="n">subplots</span><span class="p">()</span>
<span class="c1"># 주기적인 간격에 이 locator가 tick을 설정</span>
<span class="n">loc</span> <span class="o">=</span> <span class="n">ticker</span><span class="o">.</span><span class="n">MultipleLocator</span><span class="p">(</span><span class="n">base</span><span class="o">=</span><span class="mf">0.2</span><span class="p">)</span>
<span class="n">ax</span><span class="o">.</span><span class="n">yaxis</span><span class="o">.</span><span class="n">set_major_locator</span><span class="p">(</span><span class="n">loc</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">plot</span><span class="p">(</span><span class="n">points</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="id11">
<h2>평가<a class="headerlink" href="#id11" title="Permalink to this headline">¶</a></h2>
<p>평가는 대부분 학습과 동일하지만 목표가 없으므로 각 단계마다 디코더의
예측을 되돌려 전달합니다.
단어를 예측할 때마다 그 단어를 출력 문자열에 추가합니다.
만약 EOS 토큰을 예측하면 거기에서 멈춥니다.
나중에 도식화를 위해서 디코더의 Attention 출력을 저장합니다.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">evaluate</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">sentence</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="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="n">input_tensor</span> <span class="o">=</span> <span class="n">tensorFromSentence</span><span class="p">(</span><span class="n">input_lang</span><span class="p">,</span> <span class="n">sentence</span><span class="p">)</span>
<span class="n">input_length</span> <span class="o">=</span> <span class="n">input_tensor</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_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_outputs</span> <span class="o">=</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">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</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_tensor</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">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([[</span><span class="n">SOS_token</span><span class="p">]],</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</span> <span class="c1"># SOS</span>
<span class="n">decoder_hidden</span> <span class="o">=</span> <span class="n">encoder_hidden</span>
<span class="n">decoded_words</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">decoder_attentions</span> <span class="o">=</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">max_length</span><span class="p">)</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">max_length</span><span class="p">):</span>
<span class="n">decoder_output</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">decoder_attention</span> <span class="o">=</span> <span class="n">decoder</span><span class="p">(</span>
<span class="n">decoder_input</span><span class="p">,</span> <span class="n">decoder_hidden</span><span class="p">,</span> <span class="n">encoder_outputs</span><span class="p">)</span>
<span class="n">decoder_attentions</span><span class="p">[</span><span class="n">di</span><span class="p">]</span> <span class="o">=</span> <span class="n">decoder_attention</span><span class="o">.</span><span class="n">data</span>
<span class="n">topv</span><span class="p">,</span> <span class="n">topi</span> <span class="o">=</span> <span class="n">decoder_output</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">topk</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="k">if</span> <span class="n">topi</span><span class="o">.</span><span class="n">item</span><span class="p">()</span> <span class="o">==</span> <span class="n">EOS_token</span><span class="p">:</span>
<span class="n">decoded_words</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="s1">'<EOS>'</span><span class="p">)</span>
<span class="k">break</span>