Repository navigation
Expand file tree
/
Copy pathdeep-learning.html
More file actions
1943 lines (1507 loc) · 80.8 KB
/
Copy pathdeep-learning.html
File metadata and controls
1943 lines (1507 loc) · 80.8 KB
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 lang="en" data-theme="light">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>SMILE - Deep Learning</title>
<meta name="description" content="Statistical Machine Intelligence and Learning Engine">
<script>
(function () {
try {
var stored = localStorage.getItem('smile-theme');
var theme = (stored === 'dark' || stored === 'light') ? stored : 'light';
document.documentElement.setAttribute('data-theme', theme);
} catch (e) {}
})();
</script>
<link rel="stylesheet" href="css/site.css">
<link rel="stylesheet" href="css/prism.min.css" id="prism-light">
<link rel="stylesheet" href="css/prism-tomorrow.min.css" id="prism-dark" disabled>
<script>
(function () {
var dark = document.documentElement.getAttribute('data-theme') === 'dark';
var lightSheet = document.getElementById('prism-light');
var darkSheet = document.getElementById('prism-dark');
if (lightSheet && darkSheet) {
lightSheet.disabled = dark;
darkSheet.disabled = !dark;
}
})();
</script>
<script>
(function () {
var queued = [];
function stub() { queued.push(arguments); }
stub._smileQueue = true;
window.vegaEmbed = stub;
document.addEventListener('DOMContentLoaded', function () {
var embed = window.vegaEmbed;
if (typeof embed !== 'function' || embed._smileQueue) return;
queued.forEach(function (args) { embed.apply(window, args); });
});
})();
</script>
<script defer src="https://cdn.jsdelivr.net/npm/vega@5"></script>
<script defer src="https://cdn.jsdelivr.net/npm/vega-lite@5"></script>
<script defer src="https://cdn.jsdelivr.net/npm/vega-embed@6"></script>
<script async src="https://www.googletagmanager.com/gtag/js?id=G-57GD08QCML"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-57GD08QCML');
</script>
</head>
<body>
<a class="skip-link" href="#content">Skip to content</a>
<header class="site-header" id="site-header">
<nav class="nav-bar" aria-label="Primary">
<a class="nav-word" href="index.html">SMILE</a>
<ul class="nav-links">
<li>
<button type="button" aria-expanded="false" aria-haspopup="true">Products</button>
<ul class="nav-dropdown">
<li><a href="core-engine.html">Core Engine</a></li>
<li><a href="serve-engine.html">Serve Engine</a></li>
<li><a href="agent-studio.html">Agent Studio</a></li>
</ul>
</li>
<li class="nav-resources">
<button type="button" aria-expanded="false" aria-haspopup="true">Resources</button>
<ul class="nav-dropdown">
<li class="nav-docs">
<button type="button" aria-expanded="false" aria-haspopup="true">Docs</button>
<div class="nav-panel">
<div>
<p>Start</p>
<a href="quickstart.html">Quick Start</a>
<a href="studio.html">SMILE Studio</a>
<a href="tutorial.html">Tutorial</a>
<a href="data.html">Data Processing</a>
<a href="data-io.html">Data Import and Export</a>
<a href="visualization.html">Data Visualization</a>
<a href="vegalite.html">Declarative Visualization</a>
</div>
<div>
<p>Supervised</p>
<a href="classification.html">Classification</a>
<a href="regression.html">Regression</a>
<a href="deep-learning.html">Deep Learning</a>
<a href="time-series.html">Time Series</a>
<a href="sequence-labeling.html">Sequence Labeling</a>
<a href="nlp.html">Natural Language Processing</a>
<a href="feature-engineering.html">Feature Engineering</a>
<a href="model-validation.html">Model Validation</a>
<a href="hpo.html">Hyperparameter Optimization</a>
<a href="training.html">Unified Training Interface</a>
</div>
<div>
<p>Unsupervised</p>
<a href="clustering.html">Clustering</a>
<a href="vector-quantization.html">Vector Quantization</a>
<a href="anomaly-detection.html">Anomaly Detection</a>
<a href="association-rule.html">Association Rule Mining</a>
<a href="mds.html">Multi-Dimensional Scaling</a>
<a href="manifold.html">Manifold Learning</a>
</div>
<div>
<p>GenAI</p>
<a href="llm.html">Large Language Models</a>
<a href="computer-vision.html">Computer Vision</a>
<a href="onnx.html">ONNX</a>
<a href="onnx-genai.html">ONNX GenAI</a>
<a href="serve.html">SMILE Serve</a>
<p>Math</p>
<a href="linear-algebra.html">Linear Algebra</a>
<a href="statistics.html">Statistics</a>
<a href="interpolation.html">Interpolation</a>
<a href="wavelet.html">Wavelet</a>
<a href="compressed-sensing.html">Compressed Sensing</a>
<a href="graph.html">Graph Data Structure</a>
</div>
</div>
</li>
<li class="nav-api">
<button type="button" aria-expanded="false" aria-haspopup="true">API</button>
<ul class="nav-dropdown nav-flyout">
<li><a href="api/java/index.html" target="_blank" rel="noopener">Java</a></li>
<li><a href="api/scala/index.html" target="_blank" rel="noopener">Scala</a></li>
<li><a href="api/kotlin/index.html" target="_blank" rel="noopener">Kotlin</a></li>
<li><a href="api/clojure/index.html" target="_blank" rel="noopener">Clojure</a></li>
</ul>
</li>
<li><a href="faq.html">FAQ</a></li>
<li><a href="about.html">About</a></li>
</ul>
</li>
<li><a href="enterprise.html">Enterprise</a></li>
<li><a href="licensing.html">Licensing</a></li>
</ul>
<div class="nav-actions">
<a class="btn btn-quiet nav-start" href="quickstart.html">Start</a>
<button type="button" class="btn btn-accent nav-demo" data-contact-open data-contact-mode="demo">Get a demo</button>
<button type="button" class="icon-btn" id="search-toggle" aria-label="Open search" title="Search (⌘K)">
<svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" aria-hidden="true"><circle cx="11" cy="11" r="7"/><path d="M20 20l-3-3"/></svg>
</button>
<button type="button" class="icon-btn" id="theme-toggle" aria-label="Toggle dark mode" title="Toggle theme">
<svg class="icon-sun" width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" aria-hidden="true">
<circle cx="12" cy="12" r="4"/><path d="M12 2v2M12 20v2M4.93 4.93l1.41 1.41M17.66 17.66l1.41 1.41M2 12h2M20 12h2M4.93 19.07l1.41-1.41M17.66 6.34l1.41-1.41"/>
</svg>
<svg class="icon-moon" width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" aria-hidden="true" hidden>
<path d="M21 14.5A8.5 8.5 0 1 1 9.5 3 7 7 0 0 0 21 14.5z"/>
</svg>
</button>
<button type="button" class="icon-btn menu-toggle" id="menu-toggle" aria-expanded="false" aria-controls="mobile-drawer" aria-label="Open menu">
<svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" aria-hidden="true"><path d="M4 7h16M4 12h16M4 17h16"/></svg>
</button>
</div>
</nav>
<div class="mobile-drawer" id="mobile-drawer">
<button type="button" class="btn btn-accent mobile-contact" data-contact-open data-contact-mode="demo">Get a demo</button>
<a class="btn btn-quiet mobile-start" href="quickstart.html">Start</a>
<details><summary>Products</summary>
<a href="core-engine.html">Core Engine</a>
<a href="serve-engine.html">Serve Engine</a>
<a href="agent-studio.html">Agent Studio</a>
</details>
<details><summary>Resources</summary>
<details><summary>Docs</summary>
<p class="drawer-label">Start</p>
<a href="quickstart.html">Quick Start</a>
<a href="studio.html">SMILE Studio</a>
<a href="tutorial.html">Tutorial</a>
<a href="data.html">Data Processing</a>
<a href="data-io.html">Data Import and Export</a>
<a href="visualization.html">Data Visualization</a>
<a href="vegalite.html">Declarative Visualization</a>
<p class="drawer-label">Supervised</p>
<a href="classification.html">Classification</a>
<a href="regression.html">Regression</a>
<a href="deep-learning.html">Deep Learning</a>
<a href="time-series.html">Time Series</a>
<a href="sequence-labeling.html">Sequence Labeling</a>
<a href="nlp.html">Natural Language Processing</a>
<a href="feature-engineering.html">Feature Engineering</a>
<a href="model-validation.html">Model Validation</a>
<a href="hpo.html">Hyperparameter Optimization</a>
<a href="training.html">Unified Training Interface</a>
<p class="drawer-label">Unsupervised</p>
<a href="clustering.html">Clustering</a>
<a href="vector-quantization.html">Vector Quantization</a>
<a href="anomaly-detection.html">Anomaly Detection</a>
<a href="association-rule.html">Association Rule Mining</a>
<a href="mds.html">Multi-Dimensional Scaling</a>
<a href="manifold.html">Manifold Learning</a>
<p class="drawer-label">GenAI</p>
<a href="llm.html">Large Language Models</a>
<a href="computer-vision.html">Computer Vision</a>
<a href="onnx.html">ONNX</a>
<a href="onnx-genai.html">ONNX GenAI</a>
<a href="serve.html">SMILE Serve</a>
<p class="drawer-label">Math</p>
<a href="linear-algebra.html">Linear Algebra</a>
<a href="statistics.html">Statistics</a>
<a href="interpolation.html">Interpolation</a>
<a href="wavelet.html">Wavelet</a>
<a href="compressed-sensing.html">Compressed Sensing</a>
<a href="graph.html">Graph Data Structure</a>
</details>
<details><summary>API</summary>
<a href="api/java/index.html" target="_blank" rel="noopener">Java</a>
<a href="api/scala/index.html" target="_blank" rel="noopener">Scala</a>
<a href="api/kotlin/index.html" target="_blank" rel="noopener">Kotlin</a>
<a href="api/clojure/index.html" target="_blank" rel="noopener">Clojure</a>
</details>
<a href="faq.html">FAQ</a>
<a href="about.html">About</a>
</details>
<a href="enterprise.html">Enterprise</a>
<a href="licensing.html">Licensing</a>
</div>
</header>
<div class="site-shell" id="content">
<div class="col-md-3 order-md-last d-none d-md-block">
<aside class="sidebar-toc glass" aria-label="Table of contents">
<p class="toc-title">Contents</p>
<div id="toc"></div>
</aside>
</div>
<div class="col-md-9 order-md-first">
<h1 id="dl-top" class="title">Deep Learning</h1>
<p>Deep learning is based on artificial neural networks with representation learning. The adjective “deep” refers to stacking many layers so that each stage transforms the input into a slightly more abstract representation. Unlike classical pipelines that rely on hand-crafted features, deep models discover useful features from data — though architecture choices such as depth, width, and regularization still matter.</p>
<p>SMILE’s <code>smile-core</code> module already ships a tabular <code>MLP</code> for classification and regression. The <code>smile-deep</code> module goes further: an idiomatic Java API over LibTorch (CPU, CUDA, and MPS), computer-vision models and transforms (<code>smile.vision</code>), and ONNX Runtime inference (<code>smile.onnx</code>). Large language models and ONNX GenAI are covered separately on <a href="llm.html">llm.html</a>.</p>
<h2 id="prerequisites" class="title">Prerequisites</h2>
<p>Runtime requirements for <code>smile-deep</code>:</p>
<dl class="options-dl">
<dt><code>Java 25+</code></dt>
<dd>Panama FFM is used for native interop; use a JDK that matches the project’s Java toolchain.</dd>
<dt><code>smile_torch / LibTorch</code></dt>
<dd>The native <code>smile_torch</code> shared library and its LibTorch dependencies must be on the platform loader path: <code>PATH</code> (Windows), <code>LD_LIBRARY_PATH</code> (Linux), or <code>DYLD_LIBRARY_PATH</code> (macOS).</dd>
<dt><code>ONNX Runtime</code></dt>
<dd>For <code>smile.onnx</code>, place <code>onnxruntime.dll</code> / <code>libonnxruntime.so</code> / <code>libonnxruntime.dylib</code> on the same search path (<a href="https://github.com/microsoft/onnxruntime/releases">ORT releases</a>).</dd>
<dt><code>--enable-native-access=ALL-UNNAMED</code></dt>
<dd>Required when launching outside Gradle or SMILE Studio so FFM can load native code.</dd>
<dt><code>--add-opens=java.base/java.nio=ALL-UNNAMED</code></dt>
<dd>Optional but recommended if you hit buffer-interop errors; SMILE’s Gradle tests set this flag.</dd>
</dl>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">// Example JVM flags for a custom launcher
// java --enable-native-access=ALL-UNNAMED \
// --add-opens=java.base/java.nio=ALL-UNNAMED \
// -jar myapp.jar</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="module-structure" class="title">Module Structure</h2>
<p>The deep-learning surface (excluding LLMs and GenAI) is organized as follows:</p>
<pre>
smile.deep
├── tensor/ Tensor, Index, Device, DeviceType, ScalarType, Layout
├── layer/ Layer interface and built-in layers
├── activation/ ActivationFunction and activation modules
├── metric/ Accuracy, Precision, Recall, F1Score, Averaging
├── Loss.java Static factories for standard losses
├── Optimizer.java SGD, Adam, AdamW, RMSprop
├── Model.java Trainable model wrapper over a LayerBlock
├── Dataset.java Dataset interface + factories
├── DatasetImpl.java, SampleBatch.java
└── CUDA.java GPU info helpers
smile.torch
├── smile_torch_h.java FFM downcalls for the C ABI
└── Native.java Cleaner / error-handling over raw FFM bindings
smile.vision
├── transform/ Transform, ImageClassification pipeline
├── layer/ MBConv, FusedMBConv, Conv2dNormActivation, …
├── EfficientNet.java EfficientNet-V2 + pretrained factories
├── VisionModel.java LayerBlock coupled with a Transform
├── ImageDataset.java Folder-per-class dataset with prefetch
└── ImageNet.java 1000-class labels / folders / target maps
smile.onnx
├── foreign/ Panama FFM bindings to the ONNX Runtime C API
├── InferenceSession.java Load and run ONNX models
├── OrtValue.java Tensor / sequence / map containers
├── SessionOptions.java Threads, graph opts, execution providers
├── RunOptions.java Per-run logging and cancellation
├── Environment.java Shared OrtEnv / thread pools
└── ModelMetadata.java, NodeInfo.java, TensorInfo.java, …
</pre>
<p>The native bridge lives under <code>deep/src/main/cpp</code> and exposes a compact C ABI (<code>smile_torch</code>) over LibTorch. ONNX Runtime is linked separately through <code>smile.onnx.foreign</code>.</p>
<h2 id="tensors" class="title">Tensors</h2>
<p><code>Tensor</code> is a multidimensional array backed by a native LibTorch tensor. It implements <code>AutoCloseable</code> — always close tensors (or use a scope) when they are no longer needed to avoid native memory leaks.</p>
<h3 id="tensor-factories">Factory Methods</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.tensor.Tensor;
// Zeros / ones
Tensor z = Tensor.zeros(3, 4); // shape [3,4], float32
Tensor o = Tensor.ones(2, 3);
// Random
Tensor r = Tensor.rand(5, 5); // uniform [0,1)
Tensor rn = Tensor.randn(5, 5); // standard normal
// From Java arrays
float[] data = {1f, 2f, 3f, 4f};
Tensor t = Tensor.of(data, 2, 2); // shape [2,2]
long[] ldata = {0L, 1L, 2L};
Tensor li = Tensor.of(ldata, 3); // Int64 tensor
// Arange and eye
Tensor ar = Tensor.arange(0, 10, 1); // [0,1,...,9]
Tensor eye = Tensor.eye(4);</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="tensor-indexing">Indexing</h3>
<p><code>smile.deep.tensor.Index</code> provides Python-style index objects for slicing and advanced indexing.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.tensor.Index;
import smile.deep.tensor.Tensor;
Tensor t = Tensor.rand(4, 4);
Tensor col1 = t.get(Index.Colon, Index.of(1)); // all rows, column 1
Tensor row2 = t.get(Index.of(2)); // row 2
Tensor sub = t.get(Index.Slice(1, 3)); // rows 1–2
Tensor last = t.get(Index.Ellipsis, Index.of(3));
Tensor newDim = t.get(Index.None, Index.of(0)); // insert batch dim
int[] rows = {0, 2};
Tensor rowIdx = Tensor.of(rows, 2);
Tensor subset = t.get(rowIdx); // rows 0 and 2</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="tensor-arithmetic">Arithmetic & Math</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.tensor.Device;
import smile.deep.tensor.ScalarType;
import smile.deep.tensor.Tensor;
Tensor a = Tensor.ones(3);
Tensor b = Tensor.ones(3).mul(2.0);
// Non-mutating
Tensor sum = a.add(b);
Tensor diff = a.sub(1.0f);
Tensor prod = a.mul(3.0);
Tensor quot = a.div(2.0);
// In-place (trailing underscore)
a.add_(1.0);
a.sub_(0.5f);
a.mul_(2.0);
a.exp_();
a.fill_(0.0f);
// Reductions and shape
double s = a.sum().doubleValue();
Tensor argmax = a.argmax(0, false);
Tensor topk2 = a.topk(2, 0, true, true)._1();
long[] shape = a.shape();
int rank = a.dim();
Tensor flat = a.view(-1);
Tensor t2d = flat.reshape(3, 1);
Tensor tr = t2d.t().contiguous();
// Casting / device move
Tensor fp16 = a.to(ScalarType.Float16);
Tensor onCuda = a.to(Device.CUDA());</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="tensor-scope">AutoScope & noGradGuard</h3>
<p>Use <code>AutoScope</code> with <code>Tensor.push</code> / <code>Tensor.pop</code> to batch-free many tensors. Call <code>promoteToParent()</code> on values that must outlive the current scope. For inference, <code>Tensor.noGradGuard()</code> disables gradient tracking and reduces memory use.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.tensor.Tensor;
import smile.util.AutoScope;
try (var scope = new AutoScope()) {
Tensor.push(scope);
Tensor result = Tensor.randn(8, 8).mul(2.0);
result.promoteToParent(); // survive pop of this scope
Tensor.pop(); // closes other tracked tensors
}
try (var guard = Tensor.noGradGuard()) {
// Tensor output = model.forward(input);
// no gradient graph is built
}</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="tensor-options">Options: dtype & device</h3>
<p>Set global defaults with <code>Tensor.setDefaultOptions</code>, or pass a per-call <code>Tensor.Options</code>. There is no <code>Device.setDefaultDevice()</code> — prefer <code>Device.preferredDevice()</code> and apply it via options or <code>model.to(device)</code>.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.tensor.Device;
import smile.deep.tensor.ScalarType;
import smile.deep.tensor.Tensor;
import smile.deep.tensor.Tensor.Options;
Device device = Device.preferredDevice(); // CUDA if available, else CPU
Tensor.setDefaultOptions(new Options()
.dtype(ScalarType.Float32)
.device(device));
// Per-tensor override
Tensor t = Tensor.ones(new Options().dtype(ScalarType.Float64), 3, 3);
// BF16 performance check (Ampere or newer)
boolean bf16 = Tensor.isBF16Supported();</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="layers" class="title">Layers</h2>
<p>All layers implement <code>Layer</code>, which extends <code>Function<Tensor, Tensor></code> and exposes <code>forward</code>, a native module handle, a name, and <code>to(Device)</code>.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">// Layer interface (sketch)
// public interface Layer extends Function<Tensor, Tensor> {
// Tensor forward(Tensor input);
// MemorySegment module();
// String name();
// Layer to(Device device);
// }</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="layers-dense">Dense Shortcuts</h3>
<p><code>Layer</code> factories combine a linear layer with an activation (and optional dropout) into a <code>SequentialBlock</code>.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.layer.Layer;
import smile.deep.layer.LinearLayer;
import smile.deep.layer.SequentialBlock;
LinearLayer fc = Layer.linear(128, 64); // no activation
SequentialBlock r = Layer.relu(128, 64); // Linear + ReLU
SequentialBlock rd = Layer.relu(128, 64, 0.2); // + Dropout(0.2)
SequentialBlock g = Layer.gelu(128, 64);
SequentialBlock s = Layer.silu(128, 64);
SequentialBlock t = Layer.tanh(128, 64);
SequentialBlock sg = Layer.sigmoid(128, 64);
SequentialBlock ls = Layer.logSoftmax(128, 64);
SequentialBlock lk = Layer.leaky(128, 64, 0.01);</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="layers-conv">Convolution & Pooling</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.layer.Layer;
// Simple conv: in, out, kernel
var c1 = Layer.conv2d(3, 32, 3);
// Full control: kernel, stride, padding, dilation, groups, bias, mode
var c2 = Layer.conv2d(3, 32, 3, 1, 1, 1, 1, true, "zeros");
// String padding ("valid" or "same")
var c3 = Layer.conv2d(3, 32, 3, 1, "same", 1, 1, true, "zeros");
var mp = Layer.maxPool2d(2);
var ap = Layer.avgPool2d(2);
var aa = Layer.adaptiveAvgPool2d(1); // global 1×1</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="layers-norm-drop">Norms, Dropout & Embedding</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.layer.Layer;
var bn1 = Layer.batchNorm1d(64);
var bn2 = Layer.batchNorm2d(32);
var gn = Layer.groupNorm(4, 32);
var rms = Layer.rmsNorm(64);
var drop = Layer.dropout(0.3);
var emb = Layer.embedding(50000, 256);
var emb2 = Layer.embedding(50000, 256, 1.0); // scale alpha</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="layers-sequential">SequentialBlock</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.layer.Layer;
import smile.deep.layer.SequentialBlock;
import smile.deep.tensor.Tensor;
SequentialBlock mlp = new SequentialBlock(
Layer.relu(784, 256),
Layer.relu(256, 128),
Layer.logSoftmax(128, 10)
);
// Tensor output = mlp.forward(input);
SequentialBlock seq = new SequentialBlock();
seq.add(Layer.linear(64, 32));
seq.add(Layer.relu(32, 10));</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="activations" class="title">Activations</h2>
<p>Activations implement <code>ActivationFunction</code> (which extends <code>Layer</code>). Use them standalone or inside a <code>SequentialBlock</code>.</p>
<div class="table-responsive">
<table class="table table-striped table-bordered">
<thead><tr><th>Class</th><th>Activation</th></tr></thead>
<tbody>
<tr><td><code>ReLU</code></td><td>max(0, x)</td></tr>
<tr><td><code>LeakyReLU</code></td><td>max(αx, x)</td></tr>
<tr><td><code>GELU</code></td><td>Gaussian-error linear unit</td></tr>
<tr><td><code>SiLU</code></td><td>x·σ(x) (Swish)</td></tr>
<tr><td><code>Tanh</code></td><td>tanh(x)</td></tr>
<tr><td><code>Sigmoid</code></td><td>σ(x)</td></tr>
<tr><td><code>Softmax</code></td><td>softmax along last dim</td></tr>
<tr><td><code>LogSoftmax</code></td><td>log-softmax</td></tr>
<tr><td><code>LogSigmoid</code></td><td>log(σ(x))</td></tr>
<tr><td><code>GLU</code></td><td>Gated linear unit</td></tr>
<tr><td><code>HardShrink</code></td><td>x if |x| > λ else 0</td></tr>
<tr><td><code>SoftShrink</code></td><td>sign(x)·max(0,|x|−λ)</td></tr>
<tr><td><code>TanhShrink</code></td><td>x − tanh(x)</td></tr>
</tbody>
</table>
</div>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.activation.GELU;
import smile.deep.activation.ReLU;
import smile.deep.tensor.Tensor;
Tensor x = Tensor.randn(8, 16);
ReLU relu = new ReLU(true); // inplace
Tensor y = relu.forward(x);
GELU gelu = new GELU();
Tensor z = gelu.forward(x);</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="loss-optimizers" class="title">Loss & Optimizers</h2>
<h3 id="loss">Loss Factories</h3>
<p><code>Loss</code> is a <code>BiFunction<Tensor, Tensor, Tensor></code>. Standard PyTorch-style losses are available as static factories.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.Loss;
import smile.deep.tensor.Tensor;
Loss l1 = Loss.l1(); // MAE
Loss mse = Loss.mse();
Loss bce = Loss.bce(); // expects sigmoid input
Loss bceL = Loss.bceWithLogits();
Loss ce = Loss.crossEntropy(); // classification logits
Loss nll = Loss.nll(); // expects log-softmax
Loss sl1 = Loss.smoothL1();
Loss hub = Loss.huber(0.5);
Loss kl = Loss.kl();
Loss hinge = Loss.hingeEmbedding();
// Tensor lossTensor = ce.apply(logits, labels);
// double lossVal = lossTensor.doubleValue();
// Three-argument helpers
// Tensor mr = Loss.marginRanking(input1, input2, target);
// Tensor tm = Loss.tripleMarginRanking(anchor, positive, negative);</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="optimizers">Optimizers</h3>
<p>Create an optimizer from a <code>Model</code> and learning rate. Each step is <code>reset()</code> → <code>loss.backward()</code> → <code>step()</code>. Note the capital <code>A</code> in <code>Optimizer.Adam</code>.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.Model;
import smile.deep.Optimizer;
import smile.deep.layer.Layer;
import smile.deep.layer.SequentialBlock;
Model model = new Model(new SequentialBlock(
Layer.relu(784, 128),
Layer.logSoftmax(128, 10)));
Optimizer sgd = Optimizer.SGD(model, 0.01);
Optimizer adam = Optimizer.Adam(model, 1e-3);
Optimizer adamW = Optimizer.AdamW(model, 1e-3);
Optimizer rms = Optimizer.RMSprop(model, 1e-3);
// Per step inside a training loop:
// optimizer.reset();
// loss.backward();
// optimizer.step();</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="model-api" class="title">Model API</h2>
<p>Compose a <code>Model</code> from a custom <code>LayerBlock</code> when you need architectures beyond a plain sequential stack. Register submodules with <code>add(name, layer)</code> so parameters are tracked for the optimizer.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.Model;
import smile.deep.layer.Conv2dLayer;
import smile.deep.layer.Layer;
import smile.deep.layer.LayerBlock;
import smile.deep.layer.LinearLayer;
import smile.deep.tensor.Tensor;
LayerBlock net = new LayerBlock("MyCNN") {
private final Conv2dLayer conv1 = Layer.conv2d(1, 32, 3);
private final LinearLayer fc = Layer.linear(32 * 13 * 13, 10);
{
add("conv1", conv1);
add("fc", fc);
}
@Override
public Tensor forward(Tensor input) {
Tensor h = conv1.forward(input);
h = h.relu_();
h = Layer.maxPool2d(2).forward(h);
h = h.view(h.size(0), -1);
return fc.forward(h);
}
};
Model model = new Model(net);</code></pre>
<div class="playground-editor" hidden></div>
</div>
<p><code>Model.train</code> runs the full epoch loop: shuffle via the training <code>Dataset</code> iterator (each pass reshuffles), optimizer reset / forward / backward / step, optional validation metrics, and optional checkpoint path.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.Loss;
import smile.deep.Model;
import smile.deep.Optimizer;
import smile.deep.metric.Accuracy;
// model.train(
// 10, // epochs
// optimizer,
// Loss.crossEntropy(),
// trainDataset,
// testDataset, // may be null
// null, // checkpoint path
// new Accuracy() // Metric...
// );</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="metrics-data" class="title">Metrics & Data</h2>
<h3 id="metrics">Metrics</h3>
<p>Metrics implement <code>update</code>, <code>compute</code>, <code>reset</code>, and <code>name</code>. For multi-class scores, pass an <code>Averaging</code> strategy (<code>Macro</code>, <code>Micro</code>, <code>Weighted</code>).</p>
<div class="table-responsive">
<table class="table table-striped table-bordered">
<thead><tr><th>Class</th><th>Description</th></tr></thead>
<tbody>
<tr><td><code>Accuracy</code></td><td># correct / total</td></tr>
<tr><td><code>Precision</code></td><td>TP / (TP + FP)</td></tr>
<tr><td><code>Recall</code></td><td>TP / (TP + FN)</td></tr>
<tr><td><code>F1Score</code></td><td>Harmonic mean of precision and recall</td></tr>
</tbody>
</table>
</div>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.metric.Accuracy;
import smile.deep.metric.Averaging;
import smile.deep.metric.F1Score;
import smile.deep.metric.Precision;
import smile.deep.metric.Recall;
Accuracy acc = new Accuracy();
Precision mp = new Precision(Averaging.Macro);
Recall mr = new Recall(Averaging.Micro);
F1Score wf1 = new F1Score(Averaging.Weighted);
F1Score binF1 = new F1Score(); // binary, threshold 0.5
// acc.update(output, target);
// double result = acc.compute();
// acc.reset();</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h3 id="datasets">Datasets & Sampling</h3>
<p>Build mini-batch datasets from arrays with <code>Dataset.of</code>, or load IDX MNIST with <code>Dataset.mnist</code>. <code>Dataset</code> itself is <code>Iterable<SampleBatch></code> — each iterator reshuffles. Every <code>SampleBatch</code> exposes <code>data()</code> and <code>target()</code> tensors.</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.Dataset;
import smile.deep.SampleBatch;
float[][] features = /* ... */;
int[] labels = /* ... */;
Dataset ds = Dataset.of(features, labels, 64);
// also: Dataset.of(double[][], int[], batch)
// MNIST IDX: Dataset.mnist("data/mnist", true, 64);
for (SampleBatch batch : ds) {
var x = batch.data();
var y = batch.target();
// train step...
batch.close();
}</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="cuda" class="title">CUDA</h2>
<p><code>smile.deep.CUDA</code> reports device availability and memory. Use <code>Device.CUDA()</code> or <code>Device.CUDA(index)</code> to place tensors and models. <code>Tensor.isBF16Supported()</code> reports whether BF16 is performant on the current GPU (Ampere or newer).</p>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import smile.deep.CUDA;
import smile.deep.tensor.Device;
import smile.deep.tensor.Tensor;
boolean available = CUDA.isAvailable();
long count = CUDA.deviceCount();
long total = CUDA.totalMemory(0);
long free = CUDA.freeMemory(0);
long[] stats = CUDA.allocatorStats(0); // {allocated, reserved}
Device gpu = CUDA.device(); // Device.CUDA()
Device gpu1 = CUDA.device((byte) 1);
boolean bf16 = Tensor.isBF16Supported();</code></pre>
<div class="playground-editor" hidden></div>
</div>
<h2 id="vision" class="title">Computer Vision</h2>
<p>The <code>smile.vision</code> package builds image-classification pipelines on the deep layer stack: transforms, folder datasets, ImageNet label maps, and EfficientNet-V2.</p>
<h3 id="vision-transforms">Transforms</h3>
<div class="code-playground glass" data-lang="java">
<div class="playground-toolbar">
<span class="lang-tab is-active">Java</span>
<span class="toolbar-spacer"></span>
<button type="button" class="btn btn-ghost btn-sm playground-copy">Copy</button>
</div>
<pre class="playground-source language-java"><code class="language-java">import java.awt.image.BufferedImage;
import java.io.File;
import javax.imageio.ImageIO;
import smile.deep.tensor.Tensor;
import smile.vision.transform.ImageClassification;
import smile.vision.transform.Transform;
// ImageNet-style: resize shorter side → 384, center-crop 384×384, normalize
Transform t = Transform.classification(384, 384);
Transform custom = new ImageClassification(
224, 256,
new float[]{0.5f, 0.5f, 0.5f},
new float[]{0.5f, 0.5f, 0.5f},
java.awt.Image.SCALE_SMOOTH);
BufferedImage img = ImageIO.read(new File("cat.jpg"));
try (Tensor batch = t.forward(img)) {
// batch shape: [1, 3, 384, 384]
}