-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodels.html
1977 lines (1702 loc) · 126 KB
/
models.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>
<html >
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1"><meta name="generator" content="Docutils 0.17.1: http://docutils.sourceforge.net/" />
<title>Models</title>
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="_static/theme.css " type="text/css" />
<link rel="stylesheet" href="_static/css/custom.css" type="text/css" />
<link rel="stylesheet" href="_static/css/waveform-list.css" type="text/css" />
<!-- sphinx script_files -->
<script data-url_root="./" id="documentation_options" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<!-- bundled in js (rollup iife) -->
<!-- <script src="_static/theme-vendors.js"></script> -->
<script src="_static/theme.js" defer></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Training" href="training.html" />
<link rel="prev" title="Feature" href="feature.html" />
</head>
<body>
<div id="app">
<div class="theme-container" :class="pageClasses"><navbar @toggle-sidebar="toggleSidebar">
<router-link to="index.html" class="home-link">
<span class="site-name">omnizart</span>
</router-link>
<div class="links">
<navlinks class="can-hide">
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link ">
Contents
</a>
</div>
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link ">
Command Line Interface
</a>
</div>
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link router-link-active">
API Reference
</a>
</div>
</navlinks>
</div>
</navbar>
<div class="sidebar-mask" @click="toggleSidebar(false)">
</div>
<sidebar @toggle-sidebar="toggleSidebar">
<navlinks>
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link ">
Contents
</a>
</div>
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link ">
Command Line Interface
</a>
</div>
<div class="nav-item">
<a href="index.html#sound-samples"
class="nav-link router-link-active">
API Reference
</a>
</div>
</navlinks><div id="searchbox" class="searchbox" role="search">
<div class="caption"><span class="caption-text">Quick search</span>
<div class="searchformwrapper">
<form class="search" action="search.html" method="get">
<input type="text" name="q" />
<input type="submit" value="Search" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
</div><div class="sidebar-links" role="navigation" aria-label="main navigation">
<div class="sidebar-group">
<p class="caption">
<span class="caption-text"><a href="index.html#sound-samples">Contents</a></span>
</p>
<ul class="">
<li class="toctree-l1 ">
<a href="quick-start.html" class="reference internal ">Quick Start</a>
</li>
<li class="toctree-l1 ">
<a href="tutorial.html" class="reference internal ">Tutorial</a>
</li>
<li class="toctree-l1 ">
<a href="demo.html" class="reference internal ">Demonstration</a>
</li>
</ul>
</div>
<div class="sidebar-group">
<p class="caption">
<span class="caption-text"><a href="index.html#sound-samples">Command Line Interface</a></span>
</p>
<ul class="">
<li class="toctree-l1 ">
<a href="music/cli.html" class="reference internal ">omnizart music</a>
</li>
<li class="toctree-l1 ">
<a href="drum/cli.html" class="reference internal ">omnizart drum</a>
</li>
<li class="toctree-l1 ">
<a href="chord/cli.html" class="reference internal ">omnizart chord</a>
</li>
<li class="toctree-l1 ">
<a href="vocal/cli.html" class="reference internal ">omnizart vocal</a>
</li>
<li class="toctree-l1 ">
<a href="vocal-contour/cli.html" class="reference internal ">omnizart vocal-contour</a>
</li>
<li class="toctree-l1 ">
<a href="beat/cli.html" class="reference internal ">omnizart beat</a>
</li>
<li class="toctree-l1 ">
<a href="patch-cnn/cli.html" class="reference internal ">omnizart patch-cnn</a>
</li>
</ul>
</div>
<div class="sidebar-group">
<p class="caption">
<span class="caption-text"><a href="index.html#sound-samples">API Reference</a></span>
</p>
<ul class="current">
<li class="toctree-l1 ">
<a href="music/api.html" class="reference internal ">Music Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="drum/api.html" class="reference internal ">Drum Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="chord/api.html" class="reference internal ">Chord Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="vocal/api.html" class="reference internal ">Vocal Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="vocal-contour/api.html" class="reference internal ">Vocal-Contour Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="patch-cnn/api.html" class="reference internal ">Patch-CNN Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="beat/api.html" class="reference internal ">Beat Transcription</a>
</li>
<li class="toctree-l1 ">
<a href="feature.html" class="reference internal ">Feature</a>
</li>
<li class="toctree-l1 current">
<a href="#" class="reference internal current">Models</a>
<ul>
<li class="toctree-l2"><a href="#module-omnizart.models.u_net" class="reference internal">U-Net</a></li>
<li class="toctree-l2"><a href="#module-omnizart.models.t2t" class="reference internal">Tensor2Tensor</a></li>
<li class="toctree-l2"><a href="#module-omnizart.models.spectral_norm_net" class="reference internal">Spectral Normalization Model</a></li>
<li class="toctree-l2"><a href="#module-omnizart.models.chord_model" class="reference internal">Chord Transformer</a></li>
<li class="toctree-l2"><a href="#module-omnizart.models.pyramid_net" class="reference internal">Pyramid Net</a></li>
<li class="toctree-l2"><a href="#module-omnizart.models.utils" class="reference internal">Utils</a></li>
</ul>
</li>
<li class="toctree-l1 ">
<a href="training.html" class="reference internal ">Training</a>
</li>
<li class="toctree-l1 ">
<a href="base.html" class="reference internal ">Base Classes</a>
</li>
<li class="toctree-l1 ">
<a href="constants.html" class="reference internal ">Constants</a>
</li>
<li class="toctree-l1 ">
<a href="utils.html" class="reference internal ">Utilities</a>
</li>
</ul>
</div>
</div>
</sidebar>
<page>
<div class="body-header" role="navigation" aria-label="navigation">
<ul class="breadcrumbs">
<li><a href="index.html">Docs</a> »</li>
<li>Models</li>
</ul>
<ul class="page-nav">
<li class="prev">
<a href="feature.html"
title="previous chapter">← Feature</a>
</li>
<li class="next">
<a href="training.html"
title="next chapter">Training →</a>
</li>
</ul>
</div>
<hr>
<div class="content" role="main" v-pre>
<section id="models">
<h1>Models<a class="headerlink" href="#models" title="Permalink to this headline">¶</a></h1>
<span class="target" id="module-omnizart.models"></span><p>Definitions of models and corresponding layers</p>
<p>This folder contains all the definition of model architectures, including common layers that
could share among different models.</p>
<section id="module-omnizart.models.u_net">
<span id="u-net"></span><h2>U-Net<a class="headerlink" href="#module-omnizart.models.u_net" title="Permalink to this headline">¶</a></h2>
<p>Definition of customized U-Net like model architecture.</p>
<dl class="py class">
<dt class="sig sig-object py" id="omnizart.models.u_net.MultiHeadAttention">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">omnizart.models.u_net.</span></span><span class="sig-name descname"><span class="pre">MultiHeadAttention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.MultiHeadAttention" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">tensorflow.python.keras.engine.base_layer.Layer</span></code></p>
<p>Attention layer for 2D input feature.</p>
<p>As the attention mechanism consumes a large amount of memory, here we leverage a divide-and-conquer approach
implemented in the <code class="docutils literal notranslate"><span class="pre">tensor2tensor</span></code> repository. The input feature is first partitioned into smaller parts before
being passed to do self-attention computation. The processed outputs are then assembled back into the same
size as the input.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>out_channel: int</strong></dt><dd><p>Number of output channels.</p>
</dd>
<dt><strong>d_model: int</strong></dt><dd><p>Dimension of embeddings for each position of input feature.</p>
</dd>
<dt><strong>n_heads: int</strong></dt><dd><p>Number of heads for multi-head attention computation. Should be division of <cite>d_model</cite>.</p>
</dd>
<dt><strong>query_shape: Tuple(int, int)</strong></dt><dd><p>Size of each partition.</p>
</dd>
<dt><strong>memory_flange: Tuple(int, int)</strong></dt><dd><p>Additional overlapping size to be extended to each partition, indicating the final size to be
computed is: (query_shape[0]+memory_flange[0]) x (query_shape[1]+memory_flange[1])</p>
</dd>
</dl>
</dd>
</dl>
<p class="rubric">References</p>
<p>This approach is originated from <a class="reference internal" href="#r80975e0fd81b-1" id="id1">[1]</a>.</p>
<dl class="citation">
<dt class="label" id="r80975e0fd81b-1"><span class="brackets"><a class="fn-backref" href="#id1">1</a></span></dt>
<dd><p>Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran,
“Image Transformer,” in Proceedings of the 35th International Conference on Machine Learning (ICML), 2018</p>
</dd>
</dl>
<p class="rubric">Methods</p>
<table class="longtable docutils align-default">
<colgroup>
<col style="width: 10%" />
<col style="width: 90%" />
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="#omnizart.models.u_net.MultiHeadAttention.call" title="omnizart.models.u_net.MultiHeadAttention.call"><code class="xref py py-obj docutils literal notranslate"><span class="pre">call</span></code></a>(inputs)</p></td>
<td><p>This is where the layer's logic lives.</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="#omnizart.models.u_net.MultiHeadAttention.get_config" title="omnizart.models.u_net.MultiHeadAttention.get_config"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_config</span></code></a>()</p></td>
<td><p>Returns the config of the layer.</p></td>
</tr>
</tbody>
</table>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.u_net.MultiHeadAttention.call">
<span class="sig-name descname"><span class="pre">call</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.MultiHeadAttention.call" title="Permalink to this definition">¶</a></dt>
<dd><p>This is where the layer’s logic lives.</p>
<p>Note here that <cite>call()</cite> method in <cite>tf.keras</cite> is little bit different
from <cite>keras</cite> API. In <cite>keras</cite> API, you can pass support masking for
layers as additional arguments. Whereas <cite>tf.keras</cite> has <cite>compute_mask()</cite>
method to support masking.</p>
<dl class="simple">
<dt>Args:</dt><dd><p>inputs: Input tensor, or list/tuple of input tensors.
<a href="#id3"><span class="problematic" id="id4">*</span></a>args: Additional positional arguments. Currently unused.
<a href="#id5"><span class="problematic" id="id6">**</span></a>kwargs: Additional keyword arguments. Currently unused.</p>
</dd>
<dt>Returns:</dt><dd><p>A tensor or list/tuple of tensors.</p>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.u_net.MultiHeadAttention.get_config">
<span class="sig-name descname"><span class="pre">get_config</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.MultiHeadAttention.get_config" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the config of the layer.</p>
<p>A layer config is a Python dictionary (serializable)
containing the configuration of a layer.
The same layer can be reinstantiated later
(without its trained weights) from this configuration.</p>
<p>The config of a layer does not include connectivity
information, nor the layer class name. These are handled
by <cite>Network</cite> (one layer of abstraction above).</p>
<p>Note that <cite>get_config()</cite> does not guarantee to return a fresh copy of dict
every time it is called. The callers should make a copy of the returned dict
if they want to modify it.</p>
<dl class="simple">
<dt>Returns:</dt><dd><p>Python dictionary.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.u_net.conv_block">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.u_net.</span></span><span class="sig-name descname"><span class="pre">conv_block</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input_tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">channel</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">kernel_size</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">strides</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">(2,</span> <span class="pre">2)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dilation_rate</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dropout_rate</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.4</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.conv_block" title="Permalink to this definition">¶</a></dt>
<dd><p>Convolutional encoder block of U-net.</p>
<p>The block is a fully convolutional block. The encoder block does not downsample the input feature,
and thus the output will have the same dimension as the input.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.u_net.semantic_segmentation">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.u_net.</span></span><span class="sig-name descname"><span class="pre">semantic_segmentation</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">feature_num</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">352</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">timesteps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">256</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">multi_grid_layer_n</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">multi_grid_n</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ch_num</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">out_class</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">2</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dropout</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.4</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.semantic_segmentation" title="Permalink to this definition">¶</a></dt>
<dd><p>Improved U-net model with Atrous Spatial Pyramid Pooling (ASPP) block.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.u_net.semantic_segmentation_attn">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.u_net.</span></span><span class="sig-name descname"><span class="pre">semantic_segmentation_attn</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">feature_num</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">352</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">timesteps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">256</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ch_num</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">out_class</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">2</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.semantic_segmentation_attn" title="Permalink to this definition">¶</a></dt>
<dd><p>Customized attention U-net model.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.u_net.transpose_conv_block">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.u_net.</span></span><span class="sig-name descname"><span class="pre">transpose_conv_block</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input_tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">channel</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">kernel_size</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">strides</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">(2,</span> <span class="pre">2)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dropout_rate</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.4</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.u_net.transpose_conv_block" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
</section>
<section id="module-omnizart.models.t2t">
<span id="tensor2tensor"></span><h2>Tensor2Tensor<a class="headerlink" href="#module-omnizart.models.t2t" title="Permalink to this headline">¶</a></h2>
<p>Implementation of memory efficient attention.</p>
<p>Original implemetation are from <a class="reference external" href="https://github.com/tensorflow/tensor2tensor">tensor2tensor</a>.
Rewrite in tensorflow 2.0.</p>
<dl class="py class">
<dt class="sig sig-object py" id="omnizart.models.t2t.MultiHeadAttention">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">MultiHeadAttention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.MultiHeadAttention" title="Permalink to this definition">¶</a></dt>
<dd><p>Multi-head attention keras layer wrapper</p>
<p class="rubric">Methods</p>
<table class="longtable docutils align-default">
<colgroup>
<col style="width: 10%" />
<col style="width: 90%" />
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="#omnizart.models.t2t.MultiHeadAttention.call" title="omnizart.models.t2t.MultiHeadAttention.call"><code class="xref py py-obj docutils literal notranslate"><span class="pre">call</span></code></a>(q, k, v)</p></td>
<td><p>This is where the layer's logic lives.</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="#omnizart.models.t2t.MultiHeadAttention.get_config" title="omnizart.models.t2t.MultiHeadAttention.get_config"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_config</span></code></a>()</p></td>
<td><p>Returns the config of the layer.</p></td>
</tr>
</tbody>
</table>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.t2t.MultiHeadAttention.call">
<span class="sig-name descname"><span class="pre">call</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">q</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">k</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">v</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.MultiHeadAttention.call" title="Permalink to this definition">¶</a></dt>
<dd><p>This is where the layer’s logic lives.</p>
<p>Note here that <cite>call()</cite> method in <cite>tf.keras</cite> is little bit different
from <cite>keras</cite> API. In <cite>keras</cite> API, you can pass support masking for
layers as additional arguments. Whereas <cite>tf.keras</cite> has <cite>compute_mask()</cite>
method to support masking.</p>
<dl class="simple">
<dt>Args:</dt><dd><p>inputs: Input tensor, or list/tuple of input tensors.
<a href="#id8"><span class="problematic" id="id9">*</span></a>args: Additional positional arguments. Currently unused.
<a href="#id10"><span class="problematic" id="id11">**</span></a>kwargs: Additional keyword arguments. Currently unused.</p>
</dd>
<dt>Returns:</dt><dd><p>A tensor or list/tuple of tensors.</p>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.t2t.MultiHeadAttention.get_config">
<span class="sig-name descname"><span class="pre">get_config</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.MultiHeadAttention.get_config" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the config of the layer.</p>
<p>A layer config is a Python dictionary (serializable)
containing the configuration of a layer.
The same layer can be reinstantiated later
(without its trained weights) from this configuration.</p>
<p>The config of a layer does not include connectivity
information, nor the layer class name. These are handled
by <cite>Network</cite> (one layer of abstraction above).</p>
<p>Note that <cite>get_config()</cite> does not guarantee to return a fresh copy of dict
every time it is called. The callers should make a copy of the returned dict
if they want to modify it.</p>
<dl class="simple">
<dt>Returns:</dt><dd><p>Python dictionary.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.cast_like">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">cast_like</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">y</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.cast_like" title="Permalink to this definition">¶</a></dt>
<dd><p>Cast x to y’s dtype, if necessary.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.combine_heads_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">combine_heads_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.combine_heads_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Inverse of split_heads_2d.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x</strong></dt><dd><p>A Tensor with shape [batch, num_heads, height, width, channels / num_heads]</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A Tensor with shape [batch, height, width, channels]</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.combine_last_two_dimensions">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">combine_last_two_dimensions</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.combine_last_two_dimensions" title="Permalink to this definition">¶</a></dt>
<dd><p>Reshape x so that the last two dimension become one.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x</strong></dt><dd><p>A Tensor with shape […, a, b]</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A Tensor with shape […, ab]</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.dot_product_attention">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">dot_product_attention</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">q</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">k</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">v</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">bias</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dropout_rate</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.0</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">name</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">save_weights_to</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dropout_broadcast_dims</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">activation_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.dot_product_attention" title="Permalink to this definition">¶</a></dt>
<dd><p>Dot-product attention.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>q</strong></dt><dd><p>Tensor with shape […, length_q, depth_k].</p>
</dd>
<dt><strong>k</strong></dt><dd><p>Tensor with shape […, length_kv, depth_k]. Leading dimensions must
match with q.</p>
</dd>
<dt><strong>v</strong></dt><dd><p>Tensor with shape […, length_kv, depth_v] Leading dimensions must
match with q.</p>
</dd>
<dt><strong>bias</strong></dt><dd><p>Bias Tensor (see attention_bias())</p>
</dd>
<dt><strong>dropout_rate: float</strong></dt><dd><p>Dropout rate of layers.</p>
</dd>
<dt><strong>image_shapes: tuple</strong></dt><dd><p>Optional tuple of integer scalars.</p>
</dd>
<dt><strong>name: str</strong></dt><dd><p>An optional string</p>
</dd>
<dt><strong>save_weights_to: dict</strong></dt><dd><p>An optional dictionary to capture attention weights
for visualization; the weights tensor will be appended there under
a string key created from the variable scope (including name).</p>
</dd>
<dt><strong>dropout_broadcast_dims: list</strong></dt><dd><p>An optional list of integers less than rank of q.
Specifies in which dimensions to broadcast the dropout decisions.</p>
</dd>
<dt><strong>activation_dtype:</strong></dt><dd><p>Used to define function activation dtype when using
mixed precision.</p>
</dd>
<dt><strong>weight_dtype:</strong></dt><dd><p>The dtype weights are stored in when using mixed precision</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>Tensor with shape […, length_q, depth_v].</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.dropout_with_broadcast_dims">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">dropout_with_broadcast_dims</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">keep_prob</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">broadcast_dims</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.dropout_with_broadcast_dims" title="Permalink to this definition">¶</a></dt>
<dd><p>Like tf.nn.dropout but takes broadcast_dims instead of noise_shape.</p>
<p>Instead of specifying noise_shape, this function takes broadcast_dims -
a list of dimension numbers in which noise_shape should be 1. The random
keep/drop tensor has dimensionality 1 along these dimensions.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x: float</strong></dt><dd><p>A floating point tensor.</p>
</dd>
<dt><strong>keep_prob</strong></dt><dd><p>A scalar Tensor with the same type as x.
The probability that each element is kept.</p>
</dd>
<dt><strong>broadcast_dims: int</strong></dt><dd><p>An optional list of integers
the dimensions along which to broadcast the keep/drop flags.</p>
</dd>
<dt><strong>**kwargs</strong></dt><dd><p>keyword arguments to tf.nn.dropout other than “noise_shape”.</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>Tensor of the same shape as x.</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.embedding_to_padding">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">embedding_to_padding</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">emb</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.embedding_to_padding" title="Permalink to this definition">¶</a></dt>
<dd><p>Calculates the padding mask based on which embeddings are all zero.</p>
<p>We have hacked symbol_modality to return all-zero embeddings for padding.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>emb:</strong></dt><dd><p>A Tensor with shape […, depth].</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A float Tensor with shape […]. Each element is 1 if its corresponding
embedding vector is all zero, and is 0 otherwise.</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.gather_blocks_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">gather_blocks_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">indices</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.gather_blocks_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Gathers flattened blocks from x.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.gather_indices_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">gather_indices_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">block_shape</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">block_stride</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.gather_indices_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Getting gather indices.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.local_attention_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">local_attention_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">q</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">k</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">v</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">query_shape</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">(8,</span> <span class="pre">16)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">memory_flange</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">(8,</span> <span class="pre">16)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">name</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.local_attention_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Strided block local self-attention.</p>
<p>The 2-D sequence is divided into 2-D blocks of shape query_shape. Attention
for a given query position can only see memory positions less than or equal to
the query position. The memory positions are the corresponding block with
memory_flange many positions to add to the height and width of the block
(namely, left, top, and right).</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>q</strong></dt><dd><p>A tensor with shape [batch, heads, h, w, depth_k]</p>
</dd>
<dt><strong>k</strong></dt><dd><p>A tensor with shape [batch, heads, h, w, depth_k]</p>
</dd>
<dt><strong>v</strong></dt><dd><p>A tensor with shape [batch, heads, h, w, depth_v]. In the current
implementation, depth_v must be equal to depth_k.</p>
</dd>
<dt><strong>query_shape: tuple</strong></dt><dd><p>An tuple indicating the height and width of each query block.</p>
</dd>
<dt><strong>memory_flange: tuple</strong></dt><dd><p>An integer indicating how much to look in height and width
from each query block.</p>
</dd>
<dt><strong>name: str</strong></dt><dd><p>An optional string</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A Tensor of shape [batch, heads, h, w, depth_v]</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.maybe_upcast">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">maybe_upcast</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">logits</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">activation_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">hparams</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.maybe_upcast" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.mixed_precision_is_enabled">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">mixed_precision_is_enabled</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">activation_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight_dtype</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">hparams</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.mixed_precision_is_enabled" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.pad_to_multiple_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">pad_to_multiple_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">block_shape</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.pad_to_multiple_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Making sure x is a multiple of shape.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x</strong></dt><dd><p>A [batch, heads, h, w, depth] or [batch, h, w, depth] tensor</p>
</dd>
<dt><strong>block_shape</strong></dt><dd><p>A 2D list of integer shapes</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>padded_x</dt><dd><p>A [batch, heads, h, w, depth] or [batch, h, w, depth] tensor</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.positional_encoding">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">positional_encoding</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">batch_size</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">timesteps</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">n_units</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">512</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">zero_pad</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">scale</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.positional_encoding" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.relative_positional_encoding">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">relative_positional_encoding</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">n_steps</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">n_units</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">512</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">max_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">2</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.relative_positional_encoding" title="Permalink to this definition">¶</a></dt>
<dd></dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.reshape_range">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">reshape_range</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">i</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">j</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">shape</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.reshape_range" title="Permalink to this definition">¶</a></dt>
<dd><p>Reshapes a tensor between dimensions i and j.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.scatter_blocks_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">scatter_blocks_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">indices</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">shape</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.scatter_blocks_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>scatters blocks from x into shape with indices.</p>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.split_heads_2d">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">split_heads_2d</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_heads</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.split_heads_2d" title="Permalink to this definition">¶</a></dt>
<dd><p>Split channels (dimension 3) into multiple heads (becomes dimension 1).</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x</strong></dt><dd><p>A tensor with shape [batch, height, width, channels]</p>
</dd>
<dt><strong>num_heads: int</strong></dt><dd><p>Number of heads in attention’s computation.</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A tensor with shape [batch, num_heads, height, width, channels / num_heads]</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py function">
<dt class="sig sig-object py" id="omnizart.models.t2t.split_last_dimension">
<span class="sig-prename descclassname"><span class="pre">omnizart.models.t2t.</span></span><span class="sig-name descname"><span class="pre">split_last_dimension</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">x</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">n</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.t2t.split_last_dimension" title="Permalink to this definition">¶</a></dt>
<dd><p>Reshape x so that the last dimension becomes two dimensions.</p>
<p>The first of these two dimensions is n.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><dl class="simple">
<dt><strong>x</strong></dt><dd><p>A Tensor with shape […, m]</p>
</dd>
<dt><strong>n: int</strong></dt><dd><p>An integer.</p>
</dd>
</dl>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><dl class="simple">
<dt>y</dt><dd><p>A Tensor with shape […, n, m/n]</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
</section>
<section id="module-omnizart.models.spectral_norm_net">
<span id="spectral-normalization-model"></span><h2>Spectral Normalization Model<a class="headerlink" href="#module-omnizart.models.spectral_norm_net" title="Permalink to this headline">¶</a></h2>
<p>Transcription model of drum leveraging spectral normalization.</p>
<p>The model was originally developed with tensorflow 1.12.
We rewrite the model with tensorflow 2.3 module and uses keras to implement most of
the functionalities for better readability.</p>
<p>Original Author: I-Chieh, Wei
Rewrite by: BreezeWhite</p>
<dl class="py class">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.ConvSN2D">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">omnizart.models.spectral_norm_net.</span></span><span class="sig-name descname"><span class="pre">ConvSN2D</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.ConvSN2D" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">tensorflow.python.keras.engine.base_layer.Layer</span></code></p>
<p>Just a wrapper layer for using spectral normalization.</p>
<p>Original implementation referes to <a class="reference external" href="https://github.com/thisisiron/spectral_normalization-tf2">here</a>.</p>
<p class="rubric">Methods</p>
<table class="longtable docutils align-default">
<colgroup>
<col style="width: 10%" />
<col style="width: 90%" />
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="#omnizart.models.spectral_norm_net.ConvSN2D.call" title="omnizart.models.spectral_norm_net.ConvSN2D.call"><code class="xref py py-obj docutils literal notranslate"><span class="pre">call</span></code></a>(inputs)</p></td>
<td><p>This is where the layer's logic lives.</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="#omnizart.models.spectral_norm_net.ConvSN2D.get_config" title="omnizart.models.spectral_norm_net.ConvSN2D.get_config"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_config</span></code></a>()</p></td>
<td><p>This is neccessary to save the model architecture.</p></td>
</tr>
</tbody>
</table>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.ConvSN2D.call">
<span class="sig-name descname"><span class="pre">call</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.ConvSN2D.call" title="Permalink to this definition">¶</a></dt>
<dd><p>This is where the layer’s logic lives.</p>
<p>Note here that <cite>call()</cite> method in <cite>tf.keras</cite> is little bit different
from <cite>keras</cite> API. In <cite>keras</cite> API, you can pass support masking for
layers as additional arguments. Whereas <cite>tf.keras</cite> has <cite>compute_mask()</cite>
method to support masking.</p>
<dl class="simple">
<dt>Args:</dt><dd><p>inputs: Input tensor, or list/tuple of input tensors.
<a href="#id12"><span class="problematic" id="id13">*</span></a>args: Additional positional arguments. Currently unused.
<a href="#id14"><span class="problematic" id="id15">**</span></a>kwargs: Additional keyword arguments. Currently unused.</p>
</dd>
<dt>Returns:</dt><dd><p>A tensor or list/tuple of tensors.</p>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.ConvSN2D.get_config">
<span class="sig-name descname"><span class="pre">get_config</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.ConvSN2D.get_config" title="Permalink to this definition">¶</a></dt>
<dd><p>This is neccessary to save the model architecture.</p>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.SpectralNormalization">
<em class="property"><span class="pre">class</span><span class="w"> </span></em><span class="sig-prename descclassname"><span class="pre">omnizart.models.spectral_norm_net.</span></span><span class="sig-name descname"><span class="pre">SpectralNormalization</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.SpectralNormalization" title="Permalink to this definition">¶</a></dt>
<dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">tensorflow.python.keras.layers.wrappers.Wrapper</span></code></p>
<p>Spectral normalization layer.</p>
<p>Original implementation referes to <a class="reference external" href="https://github.com/thisisiron/spectral_normalization-tf2">here</a>.</p>
<p class="rubric">Methods</p>
<table class="longtable docutils align-default">
<colgroup>
<col style="width: 10%" />
<col style="width: 90%" />
</colgroup>
<tbody>
<tr class="row-odd"><td><p><a class="reference internal" href="#omnizart.models.spectral_norm_net.SpectralNormalization.build" title="omnizart.models.spectral_norm_net.SpectralNormalization.build"><code class="xref py py-obj docutils literal notranslate"><span class="pre">build</span></code></a>(input_shape)</p></td>
<td><p>Creates the variables of the layer (optional, for subclass implementers).</p></td>
</tr>
<tr class="row-even"><td><p><a class="reference internal" href="#omnizart.models.spectral_norm_net.SpectralNormalization.call" title="omnizart.models.spectral_norm_net.SpectralNormalization.call"><code class="xref py py-obj docutils literal notranslate"><span class="pre">call</span></code></a>(inputs)</p></td>
<td><p>This is where the layer's logic lives.</p></td>
</tr>
</tbody>
</table>
<table class="docutils align-default">
<colgroup>
<col style="width: 66%" />
<col style="width: 34%" />
</colgroup>
<tbody>
<tr class="row-odd"><td><p><strong>restore_weights</strong></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><strong>update_weights</strong></p></td>
<td></td>
</tr>
</tbody>
</table>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.SpectralNormalization.build">
<span class="sig-name descname"><span class="pre">build</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input_shape</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.SpectralNormalization.build" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates the variables of the layer (optional, for subclass implementers).</p>
<p>This is a method that implementers of subclasses of <cite>Layer</cite> or <cite>Model</cite>
can override if they need a state-creation step in-between
layer instantiation and layer call.</p>
<p>This is typically used to create the weights of <cite>Layer</cite> subclasses.</p>
<dl class="simple">
<dt>Args:</dt><dd><dl class="simple">
<dt>input_shape: Instance of <cite>TensorShape</cite>, or list of instances of</dt><dd><p><cite>TensorShape</cite> if the layer expects a list of inputs
(one instance per input).</p>
</dd>
</dl>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt class="sig sig-object py" id="omnizart.models.spectral_norm_net.SpectralNormalization.call">
<span class="sig-name descname"><span class="pre">call</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em><span class="sig-paren">)</span><a class="headerlink" href="#omnizart.models.spectral_norm_net.SpectralNormalization.call" title="Permalink to this definition">¶</a></dt>
<dd><p>This is where the layer’s logic lives.</p>
<p>Note here that <cite>call()</cite> method in <cite>tf.keras</cite> is little bit different
from <cite>keras</cite> API. In <cite>keras</cite> API, you can pass support masking for
layers as additional arguments. Whereas <cite>tf.keras</cite> has <cite>compute_mask()</cite>
method to support masking.</p>
<dl class="simple">
<dt>Args:</dt><dd><p>inputs: Input tensor, or list/tuple of input tensors.
<a href="#id17"><span class="problematic" id="id18">*</span></a>args: Additional positional arguments. Currently unused.
<a href="#id19"><span class="problematic" id="id20">**</span></a>kwargs: Additional keyword arguments. Currently unused.</p>
</dd>
<dt>Returns:</dt><dd><p>A tensor or list/tuple of tensors.</p>
</dd>