forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathoptim.html
1850 lines (1593 loc) · 197 KB
/
optim.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 name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.optim — PyTorch master documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/optim.html"/>
<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="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/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="Quantization" href="quantization.html" />
<link rel="prev" title="torch.onnx" href="onnx.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/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" 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.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<div class="ecosystem-dropdown">
<a id="dropdownMenuButton" data-toggle="ecosystem-dropdown">
Ecosystem
</a>
<div class="ecosystem-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/hub"">
<span class=dropdown-title>Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class=dropdown-title>Tools & Libraries</span>
<p>Explore the ecosystem of tools and libraries</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<div class="resources-dropdown">
<a id="resourcesDropdownButton" data-toggle="resources-dropdown">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/resources"">
<span class=dropdown-title>Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class=dropdown-title>About</span>
<p>Learn about PyTorch’s features and capabilities</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</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">
<a href='http://pytorch.org/docs/versions.html'>1.4.0 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption"><span class="caption-text">Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/distributed_autograd.html">Distributed Autograd Design</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/rref.html">Remote Reference Protocol</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/windows.html">Windows FAQ</a></li>
<li class="toctree-l1"><a class="reference external" href="http://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
<p class="caption"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/cppdocs/">C++ API</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
</ul>
<p class="caption"><span class="caption-text">Python API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx.html">torch.onnx</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="__config__.html">torch.__config__</a></li>
</ul>
<p class="caption"><span class="caption-text">torchvision Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision/index.html">torchvision</a></li>
</ul>
<p class="caption"><span class="caption-text">torchaudio Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio">torchaudio</a></li>
</ul>
<p class="caption"><span class="caption-text">torchtext Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text">torchtext</a></li>
</ul>
<p class="caption"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/governance.html">PyTorch Governance</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Persons of Interest</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">
Docs
</a> >
</li>
<li>torch.optim</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/optim.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="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="section" id="module-torch.optim">
<span id="torch-optim"></span><h1>torch.optim<a class="headerlink" href="#module-torch.optim" title="Permalink to this headline">¶</a></h1>
<p><a class="reference internal" href="#module-torch.optim" title="torch.optim"><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.optim</span></code></a> is a package implementing various optimization algorithms.
Most commonly used methods are already supported, and the interface is general
enough, so that more sophisticated ones can be also easily integrated in the
future.</p>
<div class="section" id="how-to-use-an-optimizer">
<h2>How to use an optimizer<a class="headerlink" href="#how-to-use-an-optimizer" title="Permalink to this headline">¶</a></h2>
<p>To use <a class="reference internal" href="#module-torch.optim" title="torch.optim"><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.optim</span></code></a> you have to construct an optimizer object, that will hold
the current state and will update the parameters based on the computed gradients.</p>
<div class="section" id="constructing-it">
<h3>Constructing it<a class="headerlink" href="#constructing-it" title="Permalink to this headline">¶</a></h3>
<p>To construct an <a class="reference internal" href="#torch.optim.Optimizer" title="torch.optim.Optimizer"><code class="xref py py-class docutils literal notranslate"><span class="pre">Optimizer</span></code></a> you have to give it an iterable containing the
parameters (all should be <code class="xref py py-class docutils literal notranslate"><span class="pre">Variable</span></code> s) to optimize. Then,
you can specify optimizer-specific options such as the learning rate, weight decay, etc.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>If you need to move a model to GPU via <code class="docutils literal notranslate"><span class="pre">.cuda()</span></code>, please do so before
constructing optimizers for it. Parameters of a model after <code class="docutils literal notranslate"><span class="pre">.cuda()</span></code> will
be different objects with those before the call.</p>
<p>In general, you should make sure that optimized parameters live in
consistent locations when optimizers are constructed and used.</p>
</div>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">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">model</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="mf">0.01</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">)</span>
<span class="n">optimizer</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">Adam</span><span class="p">([</span><span class="n">var1</span><span class="p">,</span> <span class="n">var2</span><span class="p">],</span> <span class="n">lr</span><span class="o">=</span><span class="mf">0.0001</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="per-parameter-options">
<h3>Per-parameter options<a class="headerlink" href="#per-parameter-options" title="Permalink to this headline">¶</a></h3>
<p><a class="reference internal" href="#torch.optim.Optimizer" title="torch.optim.Optimizer"><code class="xref py py-class docutils literal notranslate"><span class="pre">Optimizer</span></code></a> s also support specifying per-parameter options. To do this, instead
of passing an iterable of <code class="xref py py-class docutils literal notranslate"><span class="pre">Variable</span></code> s, pass in an iterable of
<code class="xref py py-class docutils literal notranslate"><span class="pre">dict</span></code> s. Each of them will define a separate parameter group, and should contain
a <code class="docutils literal notranslate"><span class="pre">params</span></code> key, containing a list of parameters belonging to it. Other keys
should match the keyword arguments accepted by the optimizers, and will be used
as optimization options for this group.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>You can still pass options as keyword arguments. They will be used as
defaults, in the groups that didn’t override them. This is useful when you
only want to vary a single option, while keeping all others consistent
between parameter groups.</p>
</div>
<p>For example, this is very useful when one wants to specify per-layer learning rates:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">([</span>
<span class="p">{</span><span class="s1">'params'</span><span class="p">:</span> <span class="n">model</span><span class="o">.</span><span class="n">base</span><span class="o">.</span><span class="n">parameters</span><span class="p">()},</span>
<span class="p">{</span><span class="s1">'params'</span><span class="p">:</span> <span class="n">model</span><span class="o">.</span><span class="n">classifier</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="s1">'lr'</span><span class="p">:</span> <span class="mf">1e-3</span><span class="p">}</span>
<span class="p">],</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-2</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">)</span>
</pre></div>
</div>
<p>This means that <code class="docutils literal notranslate"><span class="pre">model.base</span></code>’s parameters will use the default learning rate of <code class="docutils literal notranslate"><span class="pre">1e-2</span></code>,
<code class="docutils literal notranslate"><span class="pre">model.classifier</span></code>’s parameters will use a learning rate of <code class="docutils literal notranslate"><span class="pre">1e-3</span></code>, and a momentum of
<code class="docutils literal notranslate"><span class="pre">0.9</span></code> will be used for all parameters.</p>
</div>
<div class="section" id="taking-an-optimization-step">
<h3>Taking an optimization step<a class="headerlink" href="#taking-an-optimization-step" title="Permalink to this headline">¶</a></h3>
<p>All optimizers implement a <a class="reference internal" href="#torch.optim.Optimizer.step" title="torch.optim.Optimizer.step"><code class="xref py py-func docutils literal notranslate"><span class="pre">step()</span></code></a> method, that updates the
parameters. It can be used in two ways:</p>
<div class="section" id="optimizer-step">
<h4><code class="docutils literal notranslate"><span class="pre">optimizer.step()</span></code><a class="headerlink" href="#optimizer-step" title="Permalink to this headline">¶</a></h4>
<p>This is a simplified version supported by most optimizers. The function can be
called once the gradients are computed using e.g.
<code class="xref py py-func docutils literal notranslate"><span class="pre">backward()</span></code>.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">for</span> <span class="nb">input</span><span class="p">,</span> <span class="n">target</span> <span class="ow">in</span> <span class="n">dataset</span><span class="p">:</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">loss_fn</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
</pre></div>
</div>
</div>
<div class="section" id="optimizer-step-closure">
<h4><code class="docutils literal notranslate"><span class="pre">optimizer.step(closure)</span></code><a class="headerlink" href="#optimizer-step-closure" title="Permalink to this headline">¶</a></h4>
<p>Some optimization algorithms such as Conjugate Gradient and LBFGS need to
reevaluate the function multiple times, so you have to pass in a closure that
allows them to recompute your model. The closure should clear the gradients,
compute the loss, and return it.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">for</span> <span class="nb">input</span><span class="p">,</span> <span class="n">target</span> <span class="ow">in</span> <span class="n">dataset</span><span class="p">:</span>
<span class="k">def</span> <span class="nf">closure</span><span class="p">():</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">loss_fn</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="k">return</span> <span class="n">loss</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">(</span><span class="n">closure</span><span class="p">)</span>
</pre></div>
</div>
</div>
</div>
</div>
<div class="section" id="algorithms">
<span id="optimizer-algorithms"></span><h2>Algorithms<a class="headerlink" href="#algorithms" title="Permalink to this headline">¶</a></h2>
<dl class="class">
<dt id="torch.optim.Optimizer">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Optimizer</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">defaults</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer" title="Permalink to this definition">¶</a></dt>
<dd><p>Base class for all optimizers.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Parameters need to be specified as collections that have a deterministic
ordering that is consistent between runs. Examples of objects that don’t
satisfy those properties are sets and iterators over values of dictionaries.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – an iterable of <a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> s or
<code class="xref py py-class docutils literal notranslate"><span class="pre">dict</span></code> s. Specifies what Tensors should be optimized.</p></li>
<li><p><strong>defaults</strong> – (dict): a dict containing default values of optimization
options (used when a parameter group doesn’t specify them).</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Optimizer.add_param_group">
<code class="sig-name descname">add_param_group</code><span class="sig-paren">(</span><em class="sig-param">param_group</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer.add_param_group"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer.add_param_group" title="Permalink to this definition">¶</a></dt>
<dd><p>Add a param group to the <a class="reference internal" href="#torch.optim.Optimizer" title="torch.optim.Optimizer"><code class="xref py py-class docutils literal notranslate"><span class="pre">Optimizer</span></code></a> s <cite>param_groups</cite>.</p>
<p>This can be useful when fine tuning a pre-trained network as frozen layers can be made
trainable and added to the <a class="reference internal" href="#torch.optim.Optimizer" title="torch.optim.Optimizer"><code class="xref py py-class docutils literal notranslate"><span class="pre">Optimizer</span></code></a> as training progresses.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>param_group</strong> (<em>dict</em>) – Specifies what Tensors should be optimized along with group</p></li>
<li><p><strong>optimization options.</strong> (<em>specific</em>) – </p></li>
</ul>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.optim.Optimizer.load_state_dict">
<code class="sig-name descname">load_state_dict</code><span class="sig-paren">(</span><em class="sig-param">state_dict</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer.load_state_dict"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer.load_state_dict" title="Permalink to this definition">¶</a></dt>
<dd><p>Loads the optimizer state.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>state_dict</strong> (<em>dict</em>) – optimizer state. Should be an object returned
from a call to <a class="reference internal" href="#torch.optim.Optimizer.state_dict" title="torch.optim.Optimizer.state_dict"><code class="xref py py-meth docutils literal notranslate"><span class="pre">state_dict()</span></code></a>.</p>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.optim.Optimizer.state_dict">
<code class="sig-name descname">state_dict</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer.state_dict"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer.state_dict" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the state of the optimizer as a <code class="xref py py-class docutils literal notranslate"><span class="pre">dict</span></code>.</p>
<p>It contains two entries:</p>
<ul class="simple">
<li><dl class="simple">
<dt>state - a dict holding current optimization state. Its content</dt><dd><p>differs between optimizer classes.</p>
</dd>
</dl>
</li>
<li><p>param_groups - a dict containing all parameter groups</p></li>
</ul>
</dd></dl>
<dl class="method">
<dt id="torch.optim.Optimizer.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step (parameter update).</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em>) – A closure that reevaluates the model and
returns the loss. Optional for most optimizers.</p>
</dd>
</dl>
</dd></dl>
<dl class="method">
<dt id="torch.optim.Optimizer.zero_grad">
<code class="sig-name descname">zero_grad</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/optimizer.html#Optimizer.zero_grad"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Optimizer.zero_grad" title="Permalink to this definition">¶</a></dt>
<dd><p>Clears the gradients of all optimized <a class="reference internal" href="tensors.html#torch.Tensor" title="torch.Tensor"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> s.</p>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.Adadelta">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Adadelta</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=1.0</em>, <em class="sig-param">rho=0.9</em>, <em class="sig-param">eps=1e-06</em>, <em class="sig-param">weight_decay=0</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adadelta.html#Adadelta"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adadelta" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements Adadelta algorithm.</p>
<p>It has been proposed in <a class="reference external" href="https://arxiv.org/abs/1212.5701">ADADELTA: An Adaptive Learning Rate Method</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>rho</strong> (<em>python:float</em><em>, </em><em>optional</em>) – coefficient used for computing a running average
of squared gradients (default: 0.9)</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-6)</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – coefficient that scale delta before it is applied
to the parameters (default: 1.0)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Adadelta.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adadelta.html#Adadelta.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adadelta.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.Adagrad">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Adagrad</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.01</em>, <em class="sig-param">lr_decay=0</em>, <em class="sig-param">weight_decay=0</em>, <em class="sig-param">initial_accumulator_value=0</em>, <em class="sig-param">eps=1e-10</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adagrad.html#Adagrad"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adagrad" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements Adagrad algorithm.</p>
<p>It has been proposed in <a class="reference external" href="http://jmlr.org/papers/v12/duchi11a.html">Adaptive Subgradient Methods for Online Learning
and Stochastic Optimization</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-2)</p></li>
<li><p><strong>lr_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate decay (default: 0)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-10)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Adagrad.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adagrad.html#Adagrad.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adagrad.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.Adam">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Adam</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.001</em>, <em class="sig-param">betas=(0.9</em>, <em class="sig-param">0.999)</em>, <em class="sig-param">eps=1e-08</em>, <em class="sig-param">weight_decay=0</em>, <em class="sig-param">amsgrad=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adam.html#Adam"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adam" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements Adam algorithm.</p>
<p>It has been proposed in <a class="reference external" href="https://arxiv.org/abs/1412.6980">Adam: A Method for Stochastic Optimization</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-3)</p></li>
<li><p><strong>betas</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-8)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
<li><p><strong>amsgrad</strong> (<em>boolean</em><em>, </em><em>optional</em>) – whether to use the AMSGrad variant of this
algorithm from the paper <a class="reference external" href="https://openreview.net/forum?id=ryQu7f-RZ">On the Convergence of Adam and Beyond</a>
(default: False)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Adam.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adam.html#Adam.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adam.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.AdamW">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">AdamW</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.001</em>, <em class="sig-param">betas=(0.9</em>, <em class="sig-param">0.999)</em>, <em class="sig-param">eps=1e-08</em>, <em class="sig-param">weight_decay=0.01</em>, <em class="sig-param">amsgrad=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adamw.html#AdamW"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.AdamW" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements AdamW algorithm.</p>
<p>The original Adam algorithm was proposed in <a class="reference external" href="https://arxiv.org/abs/1412.6980">Adam: A Method for Stochastic Optimization</a>.
The AdamW variant was proposed in <a class="reference external" href="https://arxiv.org/abs/1711.05101">Decoupled Weight Decay Regularization</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-3)</p></li>
<li><p><strong>betas</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-8)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay coefficient (default: 1e-2)</p></li>
<li><p><strong>amsgrad</strong> (<em>boolean</em><em>, </em><em>optional</em>) – whether to use the AMSGrad variant of this
algorithm from the paper <a class="reference external" href="https://openreview.net/forum?id=ryQu7f-RZ">On the Convergence of Adam and Beyond</a>
(default: False)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.AdamW.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adamw.html#AdamW.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.AdamW.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.SparseAdam">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">SparseAdam</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.001</em>, <em class="sig-param">betas=(0.9</em>, <em class="sig-param">0.999)</em>, <em class="sig-param">eps=1e-08</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/sparse_adam.html#SparseAdam"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.SparseAdam" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements lazy version of Adam algorithm suitable for sparse tensors.</p>
<p>In this variant, only moments that show up in the gradient get updated, and
only those portions of the gradient get applied to the parameters.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-3)</p></li>
<li><p><strong>betas</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-8)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.SparseAdam.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/sparse_adam.html#SparseAdam.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.SparseAdam.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.Adamax">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Adamax</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.002</em>, <em class="sig-param">betas=(0.9</em>, <em class="sig-param">0.999)</em>, <em class="sig-param">eps=1e-08</em>, <em class="sig-param">weight_decay=0</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adamax.html#Adamax"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adamax" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements Adamax algorithm (a variant of Adam based on infinity norm).</p>
<p>It has been proposed in <a class="reference external" href="https://arxiv.org/abs/1412.6980">Adam: A Method for Stochastic Optimization</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 2e-3)</p></li>
<li><p><strong>betas</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – coefficients used for computing
running averages of gradient and its square</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-8)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Adamax.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/adamax.html#Adamax.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Adamax.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.ASGD">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">ASGD</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.01</em>, <em class="sig-param">lambd=0.0001</em>, <em class="sig-param">alpha=0.75</em>, <em class="sig-param">t0=1000000.0</em>, <em class="sig-param">weight_decay=0</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/asgd.html#ASGD"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.ASGD" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements Averaged Stochastic Gradient Descent.</p>
<p>It has been proposed in <a class="reference external" href="http://dl.acm.org/citation.cfm?id=131098">Acceleration of stochastic approximation by
averaging</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-2)</p></li>
<li><p><strong>lambd</strong> (<em>python:float</em><em>, </em><em>optional</em>) – decay term (default: 1e-4)</p></li>
<li><p><strong>alpha</strong> (<em>python:float</em><em>, </em><em>optional</em>) – power for eta update (default: 0.75)</p></li>
<li><p><strong>t0</strong> (<em>python:float</em><em>, </em><em>optional</em>) – point at which to start averaging (default: 1e6)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.ASGD.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/asgd.html#ASGD.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.ASGD.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.LBFGS">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">LBFGS</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=1</em>, <em class="sig-param">max_iter=20</em>, <em class="sig-param">max_eval=None</em>, <em class="sig-param">tolerance_grad=1e-07</em>, <em class="sig-param">tolerance_change=1e-09</em>, <em class="sig-param">history_size=100</em>, <em class="sig-param">line_search_fn=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/lbfgs.html#LBFGS"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.LBFGS" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements L-BFGS algorithm, heavily inspired by <cite>minFunc
<https://www.cs.ubc.ca/~schmidtm/Software/minFunc.html></cite>.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This optimizer doesn’t support per-parameter options and parameter
groups (there can be only one).</p>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Right now all parameters have to be on a single device. This will be
improved in the future.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This is a very memory intensive optimizer (it requires additional
<code class="docutils literal notranslate"><span class="pre">param_bytes</span> <span class="pre">*</span> <span class="pre">(history_size</span> <span class="pre">+</span> <span class="pre">1)</span></code> bytes). If it doesn’t fit in memory
try reducing the history size, or use a different algorithm.</p>
</div>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>lr</strong> (<em>python:float</em>) – learning rate (default: 1)</p></li>
<li><p><strong>max_iter</strong> (<em>python:int</em>) – maximal number of iterations per optimization step
(default: 20)</p></li>
<li><p><strong>max_eval</strong> (<em>python:int</em>) – maximal number of function evaluations per optimization
step (default: max_iter * 1.25).</p></li>
<li><p><strong>tolerance_grad</strong> (<em>python:float</em>) – termination tolerance on first order optimality
(default: 1e-5).</p></li>
<li><p><strong>tolerance_change</strong> (<em>python:float</em>) – termination tolerance on function
value/parameter changes (default: 1e-9).</p></li>
<li><p><strong>history_size</strong> (<em>python:int</em>) – update history size (default: 100).</p></li>
<li><p><strong>line_search_fn</strong> (<em>str</em>) – either ‘strong_wolfe’ or None (default: None).</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.LBFGS.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/lbfgs.html#LBFGS.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.LBFGS.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.RMSprop">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">RMSprop</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.01</em>, <em class="sig-param">alpha=0.99</em>, <em class="sig-param">eps=1e-08</em>, <em class="sig-param">weight_decay=0</em>, <em class="sig-param">momentum=0</em>, <em class="sig-param">centered=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/rmsprop.html#RMSprop"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.RMSprop" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements RMSprop algorithm.</p>
<p>Proposed by G. Hinton in his
<a class="reference external" href="http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf">course</a>.</p>
<p>The centered version first appears in <a class="reference external" href="https://arxiv.org/pdf/1308.0850v5.pdf">Generating Sequences
With Recurrent Neural Networks</a>.</p>
<p>The implementation here takes the square root of the gradient average before
adding epsilon (note that TensorFlow interchanges these two operations). The effective
learning rate is thus <span class="math"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>α</mi><mi mathvariant="normal">/</mi><mo stretchy="false">(</mo><msqrt><mi>v</mi></msqrt><mo>+</mo><mi>ϵ</mi><mo stretchy="false">)</mo></mrow><annotation encoding="application/x-tex">\alpha/(\sqrt{v} + \epsilon)</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:1.05028em;vertical-align:-0.25em;"></span><span class="mord mathdefault" style="margin-right:0.0037em;">α</span><span class="mord">/</span><span class="mopen">(</span><span class="mord sqrt"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8002800000000001em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em;"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span></span></span><span style="top:-2.76028em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em;"><svg width='400em' height='1.08em' viewBox='0 0 400000 1080' preserveAspectRatio='xMinYMin slice'><path d='M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z'/></svg></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.23972em;"><span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord mathdefault">ϵ</span><span class="mclose">)</span></span></span></span>
</span> where <span class="math"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>α</mi></mrow><annotation encoding="application/x-tex">\alpha</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathdefault" style="margin-right:0.0037em;">α</span></span></span></span>
</span>
is the scheduled learning rate and <span class="math"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>v</mi></mrow><annotation encoding="application/x-tex">v</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathdefault" style="margin-right:0.03588em;">v</span></span></span></span>
</span> is the weighted moving average
of the squared gradient.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-2)</p></li>
<li><p><strong>momentum</strong> (<em>python:float</em><em>, </em><em>optional</em>) – momentum factor (default: 0)</p></li>
<li><p><strong>alpha</strong> (<em>python:float</em><em>, </em><em>optional</em>) – smoothing constant (default: 0.99)</p></li>
<li><p><strong>eps</strong> (<em>python:float</em><em>, </em><em>optional</em>) – term added to the denominator to improve
numerical stability (default: 1e-8)</p></li>
<li><p><strong>centered</strong> (<em>bool</em><em>, </em><em>optional</em>) – if <code class="docutils literal notranslate"><span class="pre">True</span></code>, compute the centered RMSProp,
the gradient is normalized by an estimation of its variance</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.RMSprop.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/rmsprop.html#RMSprop.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.RMSprop.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.Rprop">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">Rprop</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=0.01</em>, <em class="sig-param">etas=(0.5</em>, <em class="sig-param">1.2)</em>, <em class="sig-param">step_sizes=(1e-06</em>, <em class="sig-param">50)</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/rprop.html#Rprop"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Rprop" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements the resilient backpropagation algorithm.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em><em>, </em><em>optional</em>) – learning rate (default: 1e-2)</p></li>
<li><p><strong>etas</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – pair of (etaminus, etaplis), that
are multiplicative increase and decrease factors
(default: (0.5, 1.2))</p></li>
<li><p><strong>step_sizes</strong> (<em>Tuple</em><em>[</em><em>python:float</em><em>, </em><em>python:float</em><em>]</em><em>, </em><em>optional</em>) – a pair of minimal and
maximal allowed step sizes (default: (1e-6, 50))</p></li>
</ul>
</dd>
</dl>
<dl class="method">
<dt id="torch.optim.Rprop.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/rprop.html#Rprop.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.Rprop.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="class">
<dt id="torch.optim.SGD">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.</code><code class="sig-name descname">SGD</code><span class="sig-paren">(</span><em class="sig-param">params</em>, <em class="sig-param">lr=<required parameter></em>, <em class="sig-param">momentum=0</em>, <em class="sig-param">dampening=0</em>, <em class="sig-param">weight_decay=0</em>, <em class="sig-param">nesterov=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/sgd.html#SGD"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.SGD" title="Permalink to this definition">¶</a></dt>
<dd><p>Implements stochastic gradient descent (optionally with momentum).</p>
<p>Nesterov momentum is based on the formula from
<a class="reference external" href="http://www.cs.toronto.edu/%7Ehinton/absps/momentum.pdf">On the importance of initialization and momentum in deep learning</a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>params</strong> (<em>iterable</em>) – iterable of parameters to optimize or dicts defining
parameter groups</p></li>
<li><p><strong>lr</strong> (<em>python:float</em>) – learning rate</p></li>
<li><p><strong>momentum</strong> (<em>python:float</em><em>, </em><em>optional</em>) – momentum factor (default: 0)</p></li>
<li><p><strong>weight_decay</strong> (<em>python:float</em><em>, </em><em>optional</em>) – weight decay (L2 penalty) (default: 0)</p></li>
<li><p><strong>dampening</strong> (<em>python:float</em><em>, </em><em>optional</em>) – dampening for momentum (default: 0)</p></li>
<li><p><strong>nesterov</strong> (<em>bool</em><em>, </em><em>optional</em>) – enables Nesterov momentum (default: False)</p></li>
</ul>
</dd>
</dl>
<p class="rubric">Example</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">optimizer</span> <span class="o">=</span> <span class="n">torch</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">model</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="mf">0.1</span><span class="p">,</span> <span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">loss_fn</span><span class="p">(</span><span class="n">model</span><span class="p">(</span><span class="nb">input</span><span class="p">),</span> <span class="n">target</span><span class="p">)</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The implementation of SGD with Momentum/Nesterov subtly differs from
Sutskever et. al. and implementations in some other frameworks.</p>
<p>Considering the specific case of Momentum, the update can be written as</p>
<div class="math">
<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>v</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>=</mo><mi>μ</mi><mo>∗</mo><msub><mi>v</mi><mi>t</mi></msub><mo>+</mo><msub><mi>g</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mspace linebreak="newline"></mspace><msub><mi>p</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>p</mi><mi>t</mi></msub><mo>−</mo><mi>l</mi><mi>r</mi><mo>∗</mo><msub><mi>v</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><annotation encoding="application/x-tex">v_{t+1} = \mu * v_{t} + g_{t+1} \\
p_{t+1} = p_{t} - lr * v_{t+1}
</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:0.6597200000000001em;vertical-align:-0.19444em;"></span><span class="mord mathdefault">μ</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">∗</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.73333em;vertical-align:-0.15em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">g</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span><span class="mspace newline"></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault">p</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:0.7777700000000001em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathdefault">p</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord mathdefault" style="margin-right:0.01968em;">l</span><span class="mord mathdefault" style="margin-right:0.02778em;">r</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">∗</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span></span></span></span>
</div><p>where p, g, v and <span class="math"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>μ</mi></mrow><annotation encoding="application/x-tex">\mu</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.625em;vertical-align:-0.19444em;"></span><span class="mord mathdefault">μ</span></span></span></span>
</span> denote the parameters, gradient,
velocity, and momentum respectively.</p>
<p>This is in contrast to Sutskever et. al. and
other frameworks which employ an update of the form</p>
<div class="math">
<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>v</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>=</mo><mi>μ</mi><mo>∗</mo><msub><mi>v</mi><mi>t</mi></msub><mo>+</mo><mi>l</mi><mi>r</mi><mo>∗</mo><msub><mi>g</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mspace linebreak="newline"></mspace><msub><mi>p</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>p</mi><mi>t</mi></msub><mo>−</mo><msub><mi>v</mi><mrow><mi>t</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><annotation encoding="application/x-tex">v_{t+1} = \mu * v_{t} + lr * g_{t+1} \\
p_{t+1} = p_{t} - v_{t+1}
</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:0.6597200000000001em;vertical-align:-0.19444em;"></span><span class="mord mathdefault">μ</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">∗</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.73333em;vertical-align:-0.15em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord mathdefault" style="margin-right:0.01968em;">l</span><span class="mord mathdefault" style="margin-right:0.02778em;">r</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">∗</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">g</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span><span class="mspace newline"></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault">p</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:0.7777700000000001em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathdefault">p</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.638891em;vertical-align:-0.208331em;"></span><span class="mord"><span class="mord mathdefault" style="margin-right:0.03588em;">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathdefault mtight">t</span><span class="mbin mtight">+</span><span class="mord mtight">1</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span></span></span></span>
</div><p>The Nesterov version is analogously modified.</p>
</div>
<dl class="method">
<dt id="torch.optim.SGD.step">
<code class="sig-name descname">step</code><span class="sig-paren">(</span><em class="sig-param">closure=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/sgd.html#SGD.step"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.SGD.step" title="Permalink to this definition">¶</a></dt>
<dd><p>Performs a single optimization step.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>closure</strong> (<em>callable</em><em>, </em><em>optional</em>) – A closure that reevaluates the model
and returns the loss.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
</div>
<div class="section" id="how-to-adjust-learning-rate">
<h2>How to adjust Learning Rate<a class="headerlink" href="#how-to-adjust-learning-rate" title="Permalink to this headline">¶</a></h2>
<p><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.optim.lr_scheduler</span></code> provides several methods to adjust the learning
rate based on the number of epochs. <a class="reference internal" href="#torch.optim.lr_scheduler.ReduceLROnPlateau" title="torch.optim.lr_scheduler.ReduceLROnPlateau"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.optim.lr_scheduler.ReduceLROnPlateau</span></code></a>
allows dynamic learning rate reducing based on some validation measurements.</p>
<p>Learning rate scheduling should be applied after optimizer’s update; e.g., you
should write your code this way:</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">scheduler</span> <span class="o">=</span> <span class="o">...</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">100</span><span class="p">):</span>
<span class="gp">>>> </span> <span class="n">train</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">validate</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="gp">>>> </span> <span class="n">scheduler</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Prior to PyTorch 1.1.0, the learning rate scheduler was expected to be called before
the optimizer’s update; 1.1.0 changed this behavior in a BC-breaking way. If you use
the learning rate scheduler (calling <code class="docutils literal notranslate"><span class="pre">scheduler.step()</span></code>) before the optimizer’s update
(calling <code class="docutils literal notranslate"><span class="pre">optimizer.step()</span></code>), this will skip the first value of the learning rate schedule.
If you are unable to reproduce results after upgrading to PyTorch 1.1.0, please check
if you are calling <code class="docutils literal notranslate"><span class="pre">scheduler.step()</span></code> at the wrong time.</p>
</div>
<dl class="class">
<dt id="torch.optim.lr_scheduler.LambdaLR">
<em class="property">class </em><code class="sig-prename descclassname">torch.optim.lr_scheduler.</code><code class="sig-name descname">LambdaLR</code><span class="sig-paren">(</span><em class="sig-param">optimizer</em>, <em class="sig-param">lr_lambda</em>, <em class="sig-param">last_epoch=-1</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/optim/lr_scheduler.html#LambdaLR"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.optim.lr_scheduler.LambdaLR" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the learning rate of each parameter group to the initial lr
times a given function. When last_epoch=-1, sets initial lr as lr.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>optimizer</strong> (<a class="reference internal" href="#torch.optim.Optimizer" title="torch.optim.Optimizer"><em>Optimizer</em></a>) – Wrapped optimizer.</p></li>
<li><p><strong>lr_lambda</strong> (<em>function</em><em> or </em><em>list</em>) – A function which computes a multiplicative
factor given an integer parameter epoch, or a list of such
functions, one for each group in optimizer.param_groups.</p></li>
<li><p><strong>last_epoch</strong> (<em>python:int</em>) – The index of last epoch. Default: -1.</p></li>