-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathsvm.html
1042 lines (1007 loc) · 103 KB
/
svm.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta name="generator" content="Docutils 0.19: https://docutils.sourceforge.io/" />
<meta property="og:title" content="1.4. Support Vector Machines" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/modules/svm.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="Support vector machines (SVMs) are a set of supervised learning methods used for classification, regression and outliers detection. The advantages of support vector machines are: Effective in high ..." />
<meta property="og:image" content="https://scikit-learn/stable/_images/sphx_glr_plot_iris_svc_001.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="Support vector machines (SVMs) are a set of supervised learning methods used for classification, regression and outliers detection. The advantages of support vector machines are: Effective in high ..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>1.4. Support Vector Machines — scikit-learn 1.3.2 documentation</title>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/svm.html" />
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<link rel="stylesheet" href="../_static/css/vendor/bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="../_static/plot_directive.css" type="text/css" />
<link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Vibur" type="text/css" />
<link rel="stylesheet" href="../_static/jupyterlite_sphinx.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<script id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
<script src="../_static/js/vendor/jquery-3.6.3.slim.min.js"></script>
</head>
<body>
<nav id="navbar" class="sk-docs-navbar navbar navbar-expand-md navbar-light bg-light py-0">
<div class="container-fluid sk-docs-container px-0">
<a class="navbar-brand py-0" href="../index.html">
<img
class="sk-brand-img"
src="../_static/scikit-learn-logo-small.png"
alt="logo"/>
</a>
<button
id="sk-navbar-toggler"
class="navbar-toggler"
type="button"
data-toggle="collapse"
data-target="#navbarSupportedContent"
aria-controls="navbarSupportedContent"
aria-expanded="false"
aria-label="Toggle navigation"
>
<span class="navbar-toggler-icon"></span>
</button>
<div class="sk-navbar-collapse collapse navbar-collapse" id="navbarSupportedContent">
<ul class="navbar-nav mr-auto">
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../install.html">Install</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../user_guide.html">User Guide</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="classes.html">API</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../auto_examples/index.html">Examples</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" target="_blank" rel="noopener noreferrer" href="https://blog.scikit-learn.org/">Community</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../getting_started.html" >Getting Started</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../tutorial/index.html" >Tutorial</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../whats_new/v1.3.html" >What's new</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../glossary.html" >Glossary</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://scikit-learn.org/dev/developers/index.html" target="_blank" rel="noopener noreferrer">Development</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../faq.html" >FAQ</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../support.html" >Support</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../related_projects.html" >Related packages</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../roadmap.html" >Roadmap</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../governance.html" >Governance</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../about.html" >About us</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</li>
<li class="nav-item dropdown nav-more-item-dropdown">
<a class="sk-nav-link nav-link dropdown-toggle" href="#" id="navbarDropdown" role="button" data-toggle="dropdown" aria-haspopup="true" aria-expanded="false">More</a>
<div class="dropdown-menu" aria-labelledby="navbarDropdown">
<a class="sk-nav-dropdown-item dropdown-item" href="../getting_started.html" >Getting Started</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../tutorial/index.html" >Tutorial</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../whats_new/v1.3.html" >What's new</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../glossary.html" >Glossary</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://scikit-learn.org/dev/developers/index.html" target="_blank" rel="noopener noreferrer">Development</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../faq.html" >FAQ</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../support.html" >Support</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../related_projects.html" >Related packages</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../roadmap.html" >Roadmap</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../governance.html" >Governance</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../about.html" >About us</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</div>
</li>
</ul>
<div id="searchbox" role="search">
<div class="searchformwrapper">
<form class="search" action="../search.html" method="get">
<input class="sk-search-text-input" type="text" name="q" aria-labelledby="searchlabel" />
<input class="sk-search-text-btn" type="submit" value="Go" />
</form>
</div>
</div>
</div>
</div>
</nav>
<div class="d-flex" id="sk-doc-wrapper">
<input type="checkbox" name="sk-toggle-checkbox" id="sk-toggle-checkbox">
<label id="sk-sidemenu-toggle" class="sk-btn-toggle-toc btn sk-btn-primary" for="sk-toggle-checkbox">Toggle Menu</label>
<div id="sk-sidebar-wrapper" class="border-right">
<div class="sk-sidebar-toc-wrapper">
<div class="btn-group w-100 mb-2" role="group" aria-label="rellinks">
<a href="kernel_ridge.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="1.3. Kernel ridge regression">Prev</a><a href="../supervised_learning.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="1. Supervised learning">Up</a>
<a href="sgd.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="1.5. Stochastic Gradient Descent">Next</a>
</div>
<div class="alert alert-danger p-1 mb-2" role="alert">
<p class="text-center mb-0">
<strong>scikit-learn 1.3.2</strong><br/>
<a href="http://scikit-learn.org/dev/versions.html">Other versions</a>
</p>
</div>
<div class="alert alert-warning p-1 mb-2" role="alert">
<p class="text-center mb-0">
Please <a class="font-weight-bold" href="../about.html#citing-scikit-learn"><string>cite us</string></a> if you use the software.
</p>
</div>
<div class="sk-sidebar-toc">
<ul>
<li><a class="reference internal" href="#">1.4. Support Vector Machines</a><ul>
<li><a class="reference internal" href="#classification">1.4.1. Classification</a><ul>
<li><a class="reference internal" href="#multi-class-classification">1.4.1.1. Multi-class classification</a></li>
<li><a class="reference internal" href="#scores-and-probabilities">1.4.1.2. Scores and probabilities</a></li>
<li><a class="reference internal" href="#unbalanced-problems">1.4.1.3. Unbalanced problems</a></li>
</ul>
</li>
<li><a class="reference internal" href="#regression">1.4.2. Regression</a></li>
<li><a class="reference internal" href="#density-estimation-novelty-detection">1.4.3. Density estimation, novelty detection</a></li>
<li><a class="reference internal" href="#complexity">1.4.4. Complexity</a></li>
<li><a class="reference internal" href="#tips-on-practical-use">1.4.5. Tips on Practical Use</a></li>
<li><a class="reference internal" href="#kernel-functions">1.4.6. Kernel functions</a><ul>
<li><a class="reference internal" href="#parameters-of-the-rbf-kernel">1.4.6.1. Parameters of the RBF Kernel</a></li>
<li><a class="reference internal" href="#custom-kernels">1.4.6.2. Custom Kernels</a></li>
</ul>
</li>
<li><a class="reference internal" href="#mathematical-formulation">1.4.7. Mathematical formulation</a><ul>
<li><a class="reference internal" href="#svc">1.4.7.1. SVC</a></li>
<li><a class="reference internal" href="#svr">1.4.7.2. SVR</a></li>
</ul>
</li>
<li><a class="reference internal" href="#implementation-details">1.4.8. Implementation details</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
<div id="sk-page-content-wrapper">
<div class="sk-page-content container-fluid body px-md-3" role="main">
<section id="support-vector-machines">
<span id="svm"></span><h1><span class="section-number">1.4. </span>Support Vector Machines<a class="headerlink" href="#support-vector-machines" title="Permalink to this heading">¶</a></h1>
<p><strong>Support vector machines (SVMs)</strong> are a set of supervised learning
methods used for <a class="reference internal" href="#svm-classification"><span class="std std-ref">classification</span></a>,
<a class="reference internal" href="#svm-regression"><span class="std std-ref">regression</span></a> and <a class="reference internal" href="#svm-outlier-detection"><span class="std std-ref">outliers detection</span></a>.</p>
<p>The advantages of support vector machines are:</p>
<blockquote>
<div><ul class="simple">
<li><p>Effective in high dimensional spaces.</p></li>
<li><p>Still effective in cases where number of dimensions is greater
than the number of samples.</p></li>
<li><p>Uses a subset of training points in the decision function (called
support vectors), so it is also memory efficient.</p></li>
<li><p>Versatile: different <a class="reference internal" href="#svm-kernels"><span class="std std-ref">Kernel functions</span></a> can be
specified for the decision function. Common kernels are
provided, but it is also possible to specify custom kernels.</p></li>
</ul>
</div></blockquote>
<p>The disadvantages of support vector machines include:</p>
<blockquote>
<div><ul class="simple">
<li><p>If the number of features is much greater than the number of
samples, avoid over-fitting in choosing <a class="reference internal" href="#svm-kernels"><span class="std std-ref">Kernel functions</span></a> and regularization
term is crucial.</p></li>
<li><p>SVMs do not directly provide probability estimates, these are
calculated using an expensive five-fold cross-validation
(see <a class="reference internal" href="#scores-probabilities"><span class="std std-ref">Scores and probabilities</span></a>, below).</p></li>
</ul>
</div></blockquote>
<p>The support vector machines in scikit-learn support both dense
(<code class="docutils literal notranslate"><span class="pre">numpy.ndarray</span></code> and convertible to that by <code class="docutils literal notranslate"><span class="pre">numpy.asarray</span></code>) and
sparse (any <code class="docutils literal notranslate"><span class="pre">scipy.sparse</span></code>) sample vectors as input. However, to use
an SVM to make predictions for sparse data, it must have been fit on such
data. For optimal performance, use C-ordered <code class="docutils literal notranslate"><span class="pre">numpy.ndarray</span></code> (dense) or
<code class="docutils literal notranslate"><span class="pre">scipy.sparse.csr_matrix</span></code> (sparse) with <code class="docutils literal notranslate"><span class="pre">dtype=float64</span></code>.</p>
<section id="classification">
<span id="svm-classification"></span><h2><span class="section-number">1.4.1. </span>Classification<a class="headerlink" href="#classification" title="Permalink to this heading">¶</a></h2>
<p><a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> are classes
capable of performing binary and multi-class classification on a dataset.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/svm/plot_iris_svc.html"><img alt="../_images/sphx_glr_plot_iris_svc_001.png" src="../_images/sphx_glr_plot_iris_svc_001.png" /></a>
</figure>
<p><a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> are similar methods, but accept slightly
different sets of parameters and have different mathematical formulations (see
section <a class="reference internal" href="#svm-mathematical-formulation"><span class="std std-ref">Mathematical formulation</span></a>). On the other hand,
<a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> is another (faster) implementation of Support Vector
Classification for the case of a linear kernel. It also
lacks some of the attributes of <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>, like
<code class="docutils literal notranslate"><span class="pre">support_</span></code>. <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> uses <code class="docutils literal notranslate"><span class="pre">squared_hinge</span></code> loss and due to its
implementation in <code class="docutils literal notranslate"><span class="pre">liblinear</span></code> it also regularizes the intercept, if considered.
This effect can however be reduced by carefully fine tuning its
<code class="docutils literal notranslate"><span class="pre">intercept_scaling</span></code> parameter, which allows the intercept term to have a
different regularization behavior compared to the other features. The
classification results and score can therefore differ from the other two
classifiers.</p>
<p>As other classifiers, <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> and
<a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> take as input two arrays: an array <code class="docutils literal notranslate"><span class="pre">X</span></code> of shape
<code class="docutils literal notranslate"><span class="pre">(n_samples,</span> <span class="pre">n_features)</span></code> holding the training samples, and an array <code class="docutils literal notranslate"><span class="pre">y</span></code> of
class labels (strings or integers), of shape <code class="docutils literal notranslate"><span class="pre">(n_samples)</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">svm</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">0</span><span class="p">],</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">SVC()</span>
</pre></div>
</div>
<p>After being fitted, the model can then be used to predict new values:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mf">2.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([1])</span>
</pre></div>
</div>
<p>SVMs decision function (detailed in the <a class="reference internal" href="#svm-mathematical-formulation"><span class="std std-ref">Mathematical formulation</span></a>)
depends on some subset of the training data, called the support vectors. Some
properties of these support vectors can be found in attributes
<code class="docutils literal notranslate"><span class="pre">support_vectors_</span></code>, <code class="docutils literal notranslate"><span class="pre">support_</span></code> and <code class="docutils literal notranslate"><span class="pre">n_support_</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="c1"># get support vectors</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">support_vectors_</span>
<span class="go">array([[0., 0.],</span>
<span class="go"> [1., 1.]])</span>
<span class="gp">>>> </span><span class="c1"># get indices of support vectors</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">support_</span>
<span class="go">array([0, 1]...)</span>
<span class="gp">>>> </span><span class="c1"># get number of support vectors for each class</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">n_support_</span>
<span class="go">array([1, 1]...)</span>
</pre></div>
</div>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_separating_hyperplane.html#sphx-glr-auto-examples-svm-plot-separating-hyperplane-py"><span class="std std-ref">SVM: Maximum margin separating hyperplane</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/svm/plot_svm_nonlinear.html#sphx-glr-auto-examples-svm-plot-svm-nonlinear-py"><span class="std std-ref">Non-linear SVM</span></a></p></li>
<li><p><a class="reference internal" href="../auto_examples/svm/plot_svm_anova.html#sphx-glr-auto-examples-svm-plot-svm-anova-py"><span class="std std-ref">SVM-Anova: SVM with univariate feature selection</span></a>,</p></li>
</ul>
</aside>
<section id="multi-class-classification">
<span id="svm-multi-class"></span><h3><span class="section-number">1.4.1.1. </span>Multi-class classification<a class="headerlink" href="#multi-class-classification" title="Permalink to this heading">¶</a></h3>
<p><a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> implement the “one-versus-one”
approach for multi-class classification. In total,
<code class="docutils literal notranslate"><span class="pre">n_classes</span> <span class="pre">*</span> <span class="pre">(n_classes</span> <span class="pre">-</span> <span class="pre">1)</span> <span class="pre">/</span> <span class="pre">2</span></code>
classifiers are constructed and each one trains data from two classes.
To provide a consistent interface with other classifiers, the
<code class="docutils literal notranslate"><span class="pre">decision_function_shape</span></code> option allows to monotonically transform the
results of the “one-versus-one” classifiers to a “one-vs-rest” decision
function of shape <code class="docutils literal notranslate"><span class="pre">(n_samples,</span> <span class="pre">n_classes)</span></code>.</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">],</span> <span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="p">[</span><span class="mi">2</span><span class="p">],</span> <span class="p">[</span><span class="mi">3</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">Y</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">decision_function_shape</span><span class="o">=</span><span class="s1">'ovo'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">Y</span><span class="p">)</span>
<span class="go">SVC(decision_function_shape='ovo')</span>
<span class="gp">>>> </span><span class="n">dec</span> <span class="o">=</span> <span class="n">clf</span><span class="o">.</span><span class="n">decision_function</span><span class="p">([[</span><span class="mi">1</span><span class="p">]])</span>
<span class="gp">>>> </span><span class="n">dec</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="c1"># 4 classes: 4*3/2 = 6</span>
<span class="go">6</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">decision_function_shape</span> <span class="o">=</span> <span class="s2">"ovr"</span>
<span class="gp">>>> </span><span class="n">dec</span> <span class="o">=</span> <span class="n">clf</span><span class="o">.</span><span class="n">decision_function</span><span class="p">([[</span><span class="mi">1</span><span class="p">]])</span>
<span class="gp">>>> </span><span class="n">dec</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="c1"># 4 classes</span>
<span class="go">4</span>
</pre></div>
</div>
<p>On the other hand, <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> implements “one-vs-the-rest”
multi-class strategy, thus training <code class="docutils literal notranslate"><span class="pre">n_classes</span></code> models.</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">lin_clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">LinearSVC</span><span class="p">(</span><span class="n">dual</span><span class="o">=</span><span class="s2">"auto"</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">lin_clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">Y</span><span class="p">)</span>
<span class="go">LinearSVC(dual='auto')</span>
<span class="gp">>>> </span><span class="n">dec</span> <span class="o">=</span> <span class="n">lin_clf</span><span class="o">.</span><span class="n">decision_function</span><span class="p">([[</span><span class="mi">1</span><span class="p">]])</span>
<span class="gp">>>> </span><span class="n">dec</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="go">4</span>
</pre></div>
</div>
<p>See <a class="reference internal" href="#svm-mathematical-formulation"><span class="std std-ref">Mathematical formulation</span></a> for a complete description of
the decision function.</p>
<p><details>
<summary class="btn btn-light">
<strong>Details on multi-class strategies</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>Note that the <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> also implements an alternative multi-class
strategy, the so-called multi-class SVM formulated by Crammer and Singer
<a class="footnote-reference brackets" href="#id18" id="id1" role="doc-noteref"><span class="fn-bracket">[</span>16<span class="fn-bracket">]</span></a>, by using the option <code class="docutils literal notranslate"><span class="pre">multi_class='crammer_singer'</span></code>. In practice,
one-vs-rest classification is usually preferred, since the results are mostly
similar, but the runtime is significantly less.</p>
<p>For “one-vs-rest” <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> the attributes <code class="docutils literal notranslate"><span class="pre">coef_</span></code> and <code class="docutils literal notranslate"><span class="pre">intercept_</span></code>
have the shape <code class="docutils literal notranslate"><span class="pre">(n_classes,</span> <span class="pre">n_features)</span></code> and <code class="docutils literal notranslate"><span class="pre">(n_classes,)</span></code> respectively.
Each row of the coefficients corresponds to one of the <code class="docutils literal notranslate"><span class="pre">n_classes</span></code>
“one-vs-rest” classifiers and similar for the intercepts, in the
order of the “one” class.</p>
<p>In the case of “one-vs-one” <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>, the layout of
the attributes is a little more involved. In the case of a linear
kernel, the attributes <code class="docutils literal notranslate"><span class="pre">coef_</span></code> and <code class="docutils literal notranslate"><span class="pre">intercept_</span></code> have the shape
<code class="docutils literal notranslate"><span class="pre">(n_classes</span> <span class="pre">*</span> <span class="pre">(n_classes</span> <span class="pre">-</span> <span class="pre">1)</span> <span class="pre">/</span> <span class="pre">2,</span> <span class="pre">n_features)</span></code> and <code class="docutils literal notranslate"><span class="pre">(n_classes</span> <span class="pre">*</span>
<span class="pre">(n_classes</span> <span class="pre">-</span> <span class="pre">1)</span> <span class="pre">/</span> <span class="pre">2)</span></code> respectively. This is similar to the layout for
<a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> described above, with each row now corresponding
to a binary classifier. The order for classes
0 to n is “0 vs 1”, “0 vs 2” , … “0 vs n”, “1 vs 2”, “1 vs 3”, “1 vs n”, . .
. “n-1 vs n”.</p>
<p>The shape of <code class="docutils literal notranslate"><span class="pre">dual_coef_</span></code> is <code class="docutils literal notranslate"><span class="pre">(n_classes-1,</span> <span class="pre">n_SV)</span></code> with
a somewhat hard to grasp layout.
The columns correspond to the support vectors involved in any
of the <code class="docutils literal notranslate"><span class="pre">n_classes</span> <span class="pre">*</span> <span class="pre">(n_classes</span> <span class="pre">-</span> <span class="pre">1)</span> <span class="pre">/</span> <span class="pre">2</span></code> “one-vs-one” classifiers.
Each support vector <code class="docutils literal notranslate"><span class="pre">v</span></code> has a dual coefficient in each of the
<code class="docutils literal notranslate"><span class="pre">n_classes</span> <span class="pre">-</span> <span class="pre">1</span></code> classifiers comparing the class of <code class="docutils literal notranslate"><span class="pre">v</span></code> against another class.
Note that some, but not all, of these dual coefficients, may be zero.
The <code class="docutils literal notranslate"><span class="pre">n_classes</span> <span class="pre">-</span> <span class="pre">1</span></code> entries in each column are these dual coefficients,
ordered by the opposing class.</p>
<p>This might be clearer with an example: consider a three class problem with
class 0 having three support vectors
<span class="math notranslate nohighlight">\(v^{0}_0, v^{1}_0, v^{2}_0\)</span> and class 1 and 2 having two support vectors
<span class="math notranslate nohighlight">\(v^{0}_1, v^{1}_1\)</span> and <span class="math notranslate nohighlight">\(v^{0}_2, v^{1}_2\)</span> respectively. For each
support vector <span class="math notranslate nohighlight">\(v^{j}_i\)</span>, there are two dual coefficients. Let’s call
the coefficient of support vector <span class="math notranslate nohighlight">\(v^{j}_i\)</span> in the classifier between
classes <span class="math notranslate nohighlight">\(i\)</span> and <span class="math notranslate nohighlight">\(k\)</span> <span class="math notranslate nohighlight">\(\alpha^{j}_{i,k}\)</span>.
Then <code class="docutils literal notranslate"><span class="pre">dual_coef_</span></code> looks like this:</p>
<table class="docutils align-default">
<tbody>
<tr class="row-odd"><td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{0,1}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{0,1}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{2}_{0,1}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{1,0}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{1,0}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{2,0}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{2,0}\)</span></p></td>
</tr>
<tr class="row-even"><td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{0,2}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{0,2}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{2}_{0,2}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{1,2}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{1,2}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{0}_{2,1}\)</span></p></td>
<td><p><span class="math notranslate nohighlight">\(\alpha^{1}_{2,1}\)</span></p></td>
</tr>
<tr class="row-odd"><td colspan="3"><p>Coefficients
for SVs of class 0</p></td>
<td colspan="2"><p>Coefficients
for SVs of class 1</p></td>
<td colspan="2"><p>Coefficients
for SVs of class 2</p></td>
</tr>
</tbody>
</table>
<p></div>
</details></p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_iris_svc.html#sphx-glr-auto-examples-svm-plot-iris-svc-py"><span class="std std-ref">Plot different SVM classifiers in the iris dataset</span></a>,</p></li>
</ul>
</aside>
</section>
<section id="scores-and-probabilities">
<span id="scores-probabilities"></span><h3><span class="section-number">1.4.1.2. </span>Scores and probabilities<a class="headerlink" href="#scores-and-probabilities" title="Permalink to this heading">¶</a></h3>
<p>The <code class="docutils literal notranslate"><span class="pre">decision_function</span></code> method of <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> gives
per-class scores for each sample (or a single score per sample in the binary
case). When the constructor option <code class="docutils literal notranslate"><span class="pre">probability</span></code> is set to <code class="docutils literal notranslate"><span class="pre">True</span></code>,
class membership probability estimates (from the methods <code class="docutils literal notranslate"><span class="pre">predict_proba</span></code> and
<code class="docutils literal notranslate"><span class="pre">predict_log_proba</span></code>) are enabled. In the binary case, the probabilities are
calibrated using Platt scaling <a class="footnote-reference brackets" href="#id11" id="id2" role="doc-noteref"><span class="fn-bracket">[</span>9<span class="fn-bracket">]</span></a>: logistic regression on the SVM’s scores,
fit by an additional cross-validation on the training data.
In the multiclass case, this is extended as per <a class="footnote-reference brackets" href="#id12" id="id3" role="doc-noteref"><span class="fn-bracket">[</span>10<span class="fn-bracket">]</span></a>.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The same probability calibration procedure is available for all estimators
via the <a class="reference internal" href="generated/sklearn.calibration.CalibratedClassifierCV.html#sklearn.calibration.CalibratedClassifierCV" title="sklearn.calibration.CalibratedClassifierCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">CalibratedClassifierCV</span></code></a> (see
<a class="reference internal" href="calibration.html#calibration"><span class="std std-ref">Probability calibration</span></a>). In the case of <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>, this
procedure is builtin in <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a> which is used under the hood, so it does
not rely on scikit-learn’s
<a class="reference internal" href="generated/sklearn.calibration.CalibratedClassifierCV.html#sklearn.calibration.CalibratedClassifierCV" title="sklearn.calibration.CalibratedClassifierCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">CalibratedClassifierCV</span></code></a>.</p>
</div>
<p>The cross-validation involved in Platt scaling
is an expensive operation for large datasets.
In addition, the probability estimates may be inconsistent with the scores:</p>
<ul class="simple">
<li><p>the “argmax” of the scores may not be the argmax of the probabilities</p></li>
<li><p>in binary classification, a sample may be labeled by <code class="docutils literal notranslate"><span class="pre">predict</span></code> as
belonging to the positive class even if the output of <code class="docutils literal notranslate"><span class="pre">predict_proba</span></code> is
less than 0.5; and similarly, it could be labeled as negative even if the
output of <code class="docutils literal notranslate"><span class="pre">predict_proba</span></code> is more than 0.5.</p></li>
</ul>
<p>Platt’s method is also known to have theoretical issues.
If confidence scores are required, but these do not have to be probabilities,
then it is advisable to set <code class="docutils literal notranslate"><span class="pre">probability=False</span></code>
and use <code class="docutils literal notranslate"><span class="pre">decision_function</span></code> instead of <code class="docutils literal notranslate"><span class="pre">predict_proba</span></code>.</p>
<p>Please note that when <code class="docutils literal notranslate"><span class="pre">decision_function_shape='ovr'</span></code> and <code class="docutils literal notranslate"><span class="pre">n_classes</span> <span class="pre">></span> <span class="pre">2</span></code>,
unlike <code class="docutils literal notranslate"><span class="pre">decision_function</span></code>, the <code class="docutils literal notranslate"><span class="pre">predict</span></code> method does not try to break ties
by default. You can set <code class="docutils literal notranslate"><span class="pre">break_ties=True</span></code> for the output of <code class="docutils literal notranslate"><span class="pre">predict</span></code> to be
the same as <code class="docutils literal notranslate"><span class="pre">np.argmax(clf.decision_function(...),</span> <span class="pre">axis=1)</span></code>, otherwise the
first class among the tied classes will always be returned; but have in mind
that it comes with a computational cost. See
<a class="reference internal" href="../auto_examples/svm/plot_svm_tie_breaking.html#sphx-glr-auto-examples-svm-plot-svm-tie-breaking-py"><span class="std std-ref">SVM Tie Breaking Example</span></a> for an example on
tie breaking.</p>
</section>
<section id="unbalanced-problems">
<h3><span class="section-number">1.4.1.3. </span>Unbalanced problems<a class="headerlink" href="#unbalanced-problems" title="Permalink to this heading">¶</a></h3>
<p>In problems where it is desired to give more importance to certain
classes or certain individual samples, the parameters <code class="docutils literal notranslate"><span class="pre">class_weight</span></code> and
<code class="docutils literal notranslate"><span class="pre">sample_weight</span></code> can be used.</p>
<p><a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> (but not <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>) implements the parameter
<code class="docutils literal notranslate"><span class="pre">class_weight</span></code> in the <code class="docutils literal notranslate"><span class="pre">fit</span></code> method. It’s a dictionary of the form
<code class="docutils literal notranslate"><span class="pre">{class_label</span> <span class="pre">:</span> <span class="pre">value}</span></code>, where value is a floating point number > 0
that sets the parameter <code class="docutils literal notranslate"><span class="pre">C</span></code> of class <code class="docutils literal notranslate"><span class="pre">class_label</span></code> to <code class="docutils literal notranslate"><span class="pre">C</span> <span class="pre">*</span> <span class="pre">value</span></code>.
The figure below illustrates the decision boundary of an unbalanced problem,
with and without weight correction.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/svm/plot_separating_hyperplane_unbalanced.html"><img alt="../_images/sphx_glr_plot_separating_hyperplane_unbalanced_001.png" src="../_images/sphx_glr_plot_separating_hyperplane_unbalanced_001.png" style="width: 480.0px; height: 360.0px;" /></a>
</figure>
<p><a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a>,
<a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM" title="sklearn.svm.OneClassSVM"><code class="xref py py-class docutils literal notranslate"><span class="pre">OneClassSVM</span></code></a> implement also weights for
individual samples in the <code class="docutils literal notranslate"><span class="pre">fit</span></code> method through the <code class="docutils literal notranslate"><span class="pre">sample_weight</span></code> parameter.
Similar to <code class="docutils literal notranslate"><span class="pre">class_weight</span></code>, this sets the parameter <code class="docutils literal notranslate"><span class="pre">C</span></code> for the i-th
example to <code class="docutils literal notranslate"><span class="pre">C</span> <span class="pre">*</span> <span class="pre">sample_weight[i]</span></code>, which will encourage the classifier to
get these samples right. The figure below illustrates the effect of sample
weighting on the decision boundary. The size of the circles is proportional
to the sample weights:</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/svm/plot_weighted_samples.html"><img alt="../_images/sphx_glr_plot_weighted_samples_001.png" src="../_images/sphx_glr_plot_weighted_samples_001.png" style="width: 1050.0px; height: 450.0px;" /></a>
</figure>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_separating_hyperplane_unbalanced.html#sphx-glr-auto-examples-svm-plot-separating-hyperplane-unbalanced-py"><span class="std std-ref">SVM: Separating hyperplane for unbalanced classes</span></a></p></li>
<li><p><a class="reference internal" href="../auto_examples/svm/plot_weighted_samples.html#sphx-glr-auto-examples-svm-plot-weighted-samples-py"><span class="std std-ref">SVM: Weighted samples</span></a>,</p></li>
</ul>
</aside>
</section>
</section>
<section id="regression">
<span id="svm-regression"></span><h2><span class="section-number">1.4.2. </span>Regression<a class="headerlink" href="#regression" title="Permalink to this heading">¶</a></h2>
<p>The method of Support Vector Classification can be extended to solve
regression problems. This method is called Support Vector Regression.</p>
<p>The model produced by support vector classification (as described
above) depends only on a subset of the training data, because the cost
function for building the model does not care about training points
that lie beyond the margin. Analogously, the model produced by Support
Vector Regression depends only on a subset of the training data,
because the cost function ignores samples whose prediction is close to their
target.</p>
<p>There are three different implementations of Support Vector Regression:
<a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a>. <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a>
provides a faster implementation than <a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a> but only considers the
linear kernel, while <a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a> implements a slightly different formulation
than <a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a>. Due to its implementation in
<code class="docutils literal notranslate"><span class="pre">liblinear</span></code> <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a> also regularizes the intercept, if considered.
This effect can however be reduced by carefully fine tuning its
<code class="docutils literal notranslate"><span class="pre">intercept_scaling</span></code> parameter, which allows the intercept term to have a
different regularization behavior compared to the other features. The
classification results and score can therefore differ from the other two
classifiers. See <a class="reference internal" href="#svm-implementation-details"><span class="std std-ref">Implementation details</span></a> for further details.</p>
<p>As with classification classes, the fit method will take as
argument vectors X, y, only that in this case y is expected to have
floating point values instead of integer values:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">svm</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">0</span><span class="p">],</span> <span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="mf">0.5</span><span class="p">,</span> <span class="mf">2.5</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">regr</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVR</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">regr</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">SVR()</span>
<span class="gp">>>> </span><span class="n">regr</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]])</span>
<span class="go">array([1.5])</span>
</pre></div>
</div>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_svm_regression.html#sphx-glr-auto-examples-svm-plot-svm-regression-py"><span class="std std-ref">Support Vector Regression (SVR) using linear and non-linear kernels</span></a></p></li>
</ul>
</aside>
</section>
<section id="density-estimation-novelty-detection">
<span id="svm-outlier-detection"></span><h2><span class="section-number">1.4.3. </span>Density estimation, novelty detection<a class="headerlink" href="#density-estimation-novelty-detection" title="Permalink to this heading">¶</a></h2>
<p>The class <a class="reference internal" href="generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM" title="sklearn.svm.OneClassSVM"><code class="xref py py-class docutils literal notranslate"><span class="pre">OneClassSVM</span></code></a> implements a One-Class SVM which is used in
outlier detection.</p>
<p>See <a class="reference internal" href="outlier_detection.html#outlier-detection"><span class="std std-ref">Novelty and Outlier Detection</span></a> for the description and usage of OneClassSVM.</p>
</section>
<section id="complexity">
<h2><span class="section-number">1.4.4. </span>Complexity<a class="headerlink" href="#complexity" title="Permalink to this heading">¶</a></h2>
<p>Support Vector Machines are powerful tools, but their compute and
storage requirements increase rapidly with the number of training
vectors. The core of an SVM is a quadratic programming problem (QP),
separating support vectors from the rest of the training data. The QP
solver used by the <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a>-based implementation scales between
<span class="math notranslate nohighlight">\(O(n_{features} \times n_{samples}^2)\)</span> and
<span class="math notranslate nohighlight">\(O(n_{features} \times n_{samples}^3)\)</span> depending on how efficiently
the <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a> cache is used in practice (dataset dependent). If the data
is very sparse <span class="math notranslate nohighlight">\(n_{features}\)</span> should be replaced by the average number
of non-zero features in a sample vector.</p>
<p>For the linear case, the algorithm used in
<a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> by the <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/liblinear/">liblinear</a> implementation is much more
efficient than its <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a>-based <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> counterpart and can
scale almost linearly to millions of samples and/or features.</p>
</section>
<section id="tips-on-practical-use">
<h2><span class="section-number">1.4.5. </span>Tips on Practical Use<a class="headerlink" href="#tips-on-practical-use" title="Permalink to this heading">¶</a></h2>
<blockquote>
<div><ul>
<li><p><strong>Avoiding data copy</strong>: For <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> and
<a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a>, if the data passed to certain methods is not C-ordered
contiguous and double precision, it will be copied before calling the
underlying C implementation. You can check whether a given numpy array is
C-contiguous by inspecting its <code class="docutils literal notranslate"><span class="pre">flags</span></code> attribute.</p>
<p>For <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> (and <a class="reference internal" href="generated/sklearn.linear_model.LogisticRegression.html#sklearn.linear_model.LogisticRegression" title="sklearn.linear_model.LogisticRegression"><code class="xref py py-class docutils literal notranslate"><span class="pre">LogisticRegression</span></code></a>) any input passed as a numpy
array will be copied and converted to the <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/liblinear/">liblinear</a> internal sparse data
representation (double precision floats and int32 indices of non-zero
components). If you want to fit a large-scale linear classifier without
copying a dense numpy C-contiguous double precision array as input, we
suggest to use the <a class="reference internal" href="generated/sklearn.linear_model.SGDClassifier.html#sklearn.linear_model.SGDClassifier" title="sklearn.linear_model.SGDClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">SGDClassifier</span></code></a> class instead. The objective
function can be configured to be almost the same as the <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a>
model.</p>
</li>
<li><p><strong>Kernel cache size</strong>: For <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.SVR.html#sklearn.svm.SVR" title="sklearn.svm.SVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVR</span></code></a>, <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> and
<a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a>, the size of the kernel cache has a strong impact on run
times for larger problems. If you have enough RAM available, it is
recommended to set <code class="docutils literal notranslate"><span class="pre">cache_size</span></code> to a higher value than the default of
200(MB), such as 500(MB) or 1000(MB).</p></li>
<li><p><strong>Setting C</strong>: <code class="docutils literal notranslate"><span class="pre">C</span></code> is <code class="docutils literal notranslate"><span class="pre">1</span></code> by default and it’s a reasonable default
choice. If you have a lot of noisy observations you should decrease it:
decreasing C corresponds to more regularization.</p>
<p><a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a> are less sensitive to <code class="docutils literal notranslate"><span class="pre">C</span></code> when
it becomes large, and prediction results stop improving after a certain
threshold. Meanwhile, larger <code class="docutils literal notranslate"><span class="pre">C</span></code> values will take more time to train,
sometimes up to 10 times longer, as shown in <a class="footnote-reference brackets" href="#id13" id="id4" role="doc-noteref"><span class="fn-bracket">[</span>11<span class="fn-bracket">]</span></a>.</p>
</li>
<li><p>Support Vector Machine algorithms are not scale invariant, so <strong>it
is highly recommended to scale your data</strong>. For example, scale each
attribute on the input vector X to [0,1] or [-1,+1], or standardize it
to have mean 0 and variance 1. Note that the <em>same</em> scaling must be
applied to the test vector to obtain meaningful results. This can be done
easily by using a <a class="reference internal" href="generated/sklearn.pipeline.Pipeline.html#sklearn.pipeline.Pipeline" title="sklearn.pipeline.Pipeline"><code class="xref py py-class docutils literal notranslate"><span class="pre">Pipeline</span></code></a>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.pipeline</span> <span class="kn">import</span> <span class="n">make_pipeline</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.preprocessing</span> <span class="kn">import</span> <span class="n">StandardScaler</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.svm</span> <span class="kn">import</span> <span class="n">SVC</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">make_pipeline</span><span class="p">(</span><span class="n">StandardScaler</span><span class="p">(),</span> <span class="n">SVC</span><span class="p">())</span>
</pre></div>
</div>
<p>See section <a class="reference internal" href="preprocessing.html#preprocessing"><span class="std std-ref">Preprocessing data</span></a> for more details on scaling and
normalization.</p>
</li>
</ul>
<ul id="shrinking-svm">
<li><p>Regarding the <code class="docutils literal notranslate"><span class="pre">shrinking</span></code> parameter, quoting <a class="footnote-reference brackets" href="#id14" id="id5" role="doc-noteref"><span class="fn-bracket">[</span>12<span class="fn-bracket">]</span></a>: <em>We found that if the
number of iterations is large, then shrinking can shorten the training
time. However, if we loosely solve the optimization problem (e.g., by
using a large stopping tolerance), the code without using shrinking may
be much faster</em></p></li>
<li><p>Parameter <code class="docutils literal notranslate"><span class="pre">nu</span></code> in <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>/<a class="reference internal" href="generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM" title="sklearn.svm.OneClassSVM"><code class="xref py py-class docutils literal notranslate"><span class="pre">OneClassSVM</span></code></a>/<a class="reference internal" href="generated/sklearn.svm.NuSVR.html#sklearn.svm.NuSVR" title="sklearn.svm.NuSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVR</span></code></a>
approximates the fraction of training errors and support vectors.</p></li>
<li><p>In <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a>, if the data is unbalanced (e.g. many
positive and few negative), set <code class="docutils literal notranslate"><span class="pre">class_weight='balanced'</span></code> and/or try
different penalty parameters <code class="docutils literal notranslate"><span class="pre">C</span></code>.</p></li>
<li><p><strong>Randomness of the underlying implementations</strong>: The underlying
implementations of <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a> use a random number
generator only to shuffle the data for probability estimation (when
<code class="docutils literal notranslate"><span class="pre">probability</span></code> is set to <code class="docutils literal notranslate"><span class="pre">True</span></code>). This randomness can be controlled
with the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter. If <code class="docutils literal notranslate"><span class="pre">probability</span></code> is set to <code class="docutils literal notranslate"><span class="pre">False</span></code>
these estimators are not random and <code class="docutils literal notranslate"><span class="pre">random_state</span></code> has no effect on the
results. The underlying <a class="reference internal" href="generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM" title="sklearn.svm.OneClassSVM"><code class="xref py py-class docutils literal notranslate"><span class="pre">OneClassSVM</span></code></a> implementation is similar to
the ones of <a class="reference internal" href="generated/sklearn.svm.SVC.html#sklearn.svm.SVC" title="sklearn.svm.SVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">SVC</span></code></a> and <a class="reference internal" href="generated/sklearn.svm.NuSVC.html#sklearn.svm.NuSVC" title="sklearn.svm.NuSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">NuSVC</span></code></a>. As no probability estimation
is provided for <a class="reference internal" href="generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM" title="sklearn.svm.OneClassSVM"><code class="xref py py-class docutils literal notranslate"><span class="pre">OneClassSVM</span></code></a>, it is not random.</p>
<p>The underlying <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> implementation uses a random number
generator to select features when fitting the model with a dual coordinate
descent (i.e when <code class="docutils literal notranslate"><span class="pre">dual</span></code> is set to <code class="docutils literal notranslate"><span class="pre">True</span></code>). It is thus not uncommon
to have slightly different results for the same input data. If that
happens, try with a smaller <code class="docutils literal notranslate"><span class="pre">tol</span></code> parameter. This randomness can also be
controlled with the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter. When <code class="docutils literal notranslate"><span class="pre">dual</span></code> is
set to <code class="docutils literal notranslate"><span class="pre">False</span></code> the underlying implementation of <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> is
not random and <code class="docutils literal notranslate"><span class="pre">random_state</span></code> has no effect on the results.</p>
</li>
<li><p>Using L1 penalization as provided by <code class="docutils literal notranslate"><span class="pre">LinearSVC(penalty='l1',</span>
<span class="pre">dual=False)</span></code> yields a sparse solution, i.e. only a subset of feature
weights is different from zero and contribute to the decision function.
Increasing <code class="docutils literal notranslate"><span class="pre">C</span></code> yields a more complex model (more features are selected).
The <code class="docutils literal notranslate"><span class="pre">C</span></code> value that yields a “null” model (all weights equal to zero) can
be calculated using <a class="reference internal" href="generated/sklearn.svm.l1_min_c.html#sklearn.svm.l1_min_c" title="sklearn.svm.l1_min_c"><code class="xref py py-func docutils literal notranslate"><span class="pre">l1_min_c</span></code></a>.</p></li>
</ul>
</div></blockquote>
</section>
<section id="kernel-functions">
<span id="svm-kernels"></span><h2><span class="section-number">1.4.6. </span>Kernel functions<a class="headerlink" href="#kernel-functions" title="Permalink to this heading">¶</a></h2>
<p>The <em>kernel function</em> can be any of the following:</p>
<blockquote>
<div><ul class="simple">
<li><p>linear: <span class="math notranslate nohighlight">\(\langle x, x'\rangle\)</span>.</p></li>
<li><p>polynomial: <span class="math notranslate nohighlight">\((\gamma \langle x, x'\rangle + r)^d\)</span>, where
<span class="math notranslate nohighlight">\(d\)</span> is specified by parameter <code class="docutils literal notranslate"><span class="pre">degree</span></code>, <span class="math notranslate nohighlight">\(r\)</span> by <code class="docutils literal notranslate"><span class="pre">coef0</span></code>.</p></li>
<li><p>rbf: <span class="math notranslate nohighlight">\(\exp(-\gamma \|x-x'\|^2)\)</span>, where <span class="math notranslate nohighlight">\(\gamma\)</span> is
specified by parameter <code class="docutils literal notranslate"><span class="pre">gamma</span></code>, must be greater than 0.</p></li>
<li><p>sigmoid <span class="math notranslate nohighlight">\(\tanh(\gamma \langle x,x'\rangle + r)\)</span>,
where <span class="math notranslate nohighlight">\(r\)</span> is specified by <code class="docutils literal notranslate"><span class="pre">coef0</span></code>.</p></li>
</ul>
</div></blockquote>
<p>Different kernels are specified by the <code class="docutils literal notranslate"><span class="pre">kernel</span></code> parameter:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">linear_svc</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">kernel</span><span class="o">=</span><span class="s1">'linear'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">linear_svc</span><span class="o">.</span><span class="n">kernel</span>
<span class="go">'linear'</span>
<span class="gp">>>> </span><span class="n">rbf_svc</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">kernel</span><span class="o">=</span><span class="s1">'rbf'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">rbf_svc</span><span class="o">.</span><span class="n">kernel</span>
<span class="go">'rbf'</span>
</pre></div>
</div>
<p>See also <a class="reference internal" href="kernel_approximation.html#kernel-approximation"><span class="std std-ref">Kernel Approximation</span></a> for a solution to use RBF kernels that is much faster and more scalable.</p>
<section id="parameters-of-the-rbf-kernel">
<h3><span class="section-number">1.4.6.1. </span>Parameters of the RBF Kernel<a class="headerlink" href="#parameters-of-the-rbf-kernel" title="Permalink to this heading">¶</a></h3>
<p>When training an SVM with the <em>Radial Basis Function</em> (RBF) kernel, two
parameters must be considered: <code class="docutils literal notranslate"><span class="pre">C</span></code> and <code class="docutils literal notranslate"><span class="pre">gamma</span></code>. The parameter <code class="docutils literal notranslate"><span class="pre">C</span></code>,
common to all SVM kernels, trades off misclassification of training examples
against simplicity of the decision surface. A low <code class="docutils literal notranslate"><span class="pre">C</span></code> makes the decision
surface smooth, while a high <code class="docutils literal notranslate"><span class="pre">C</span></code> aims at classifying all training examples
correctly. <code class="docutils literal notranslate"><span class="pre">gamma</span></code> defines how much influence a single training example has.
The larger <code class="docutils literal notranslate"><span class="pre">gamma</span></code> is, the closer other examples must be to be affected.</p>
<p>Proper choice of <code class="docutils literal notranslate"><span class="pre">C</span></code> and <code class="docutils literal notranslate"><span class="pre">gamma</span></code> is critical to the SVM’s performance. One
is advised to use <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">GridSearchCV</span></code></a> with
<code class="docutils literal notranslate"><span class="pre">C</span></code> and <code class="docutils literal notranslate"><span class="pre">gamma</span></code> spaced exponentially far apart to choose good values.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_rbf_parameters.html#sphx-glr-auto-examples-svm-plot-rbf-parameters-py"><span class="std std-ref">RBF SVM parameters</span></a></p></li>
<li><p><a class="reference internal" href="../auto_examples/svm/plot_svm_nonlinear.html#sphx-glr-auto-examples-svm-plot-svm-nonlinear-py"><span class="std std-ref">Non-linear SVM</span></a></p></li>
</ul>
</aside>
</section>
<section id="custom-kernels">
<h3><span class="section-number">1.4.6.2. </span>Custom Kernels<a class="headerlink" href="#custom-kernels" title="Permalink to this heading">¶</a></h3>
<p>You can define your own kernels by either giving the kernel as a
python function or by precomputing the Gram matrix.</p>
<p>Classifiers with custom kernels behave the same way as any other
classifiers, except that:</p>
<blockquote>
<div><ul class="simple">
<li><p>Field <code class="docutils literal notranslate"><span class="pre">support_vectors_</span></code> is now empty, only indices of support
vectors are stored in <code class="docutils literal notranslate"><span class="pre">support_</span></code></p></li>
<li><p>A reference (and not a copy) of the first argument in the <code class="docutils literal notranslate"><span class="pre">fit()</span></code>
method is stored for future reference. If that array changes between the
use of <code class="docutils literal notranslate"><span class="pre">fit()</span></code> and <code class="docutils literal notranslate"><span class="pre">predict()</span></code> you will have unexpected results.</p></li>
</ul>
</div></blockquote>
<p><details>
<summary class="btn btn-light">
<strong>Using Python functions as kernels</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>You can use your own defined kernels by passing a function to the
<code class="docutils literal notranslate"><span class="pre">kernel</span></code> parameter.</p>
<p>Your kernel must take as arguments two matrices of shape
<code class="docutils literal notranslate"><span class="pre">(n_samples_1,</span> <span class="pre">n_features)</span></code>, <code class="docutils literal notranslate"><span class="pre">(n_samples_2,</span> <span class="pre">n_features)</span></code>
and return a kernel matrix of shape <code class="docutils literal notranslate"><span class="pre">(n_samples_1,</span> <span class="pre">n_samples_2)</span></code>.</p>
<p>The following code defines a linear kernel and creates a classifier
instance that will use that kernel:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">svm</span>
<span class="gp">>>> </span><span class="k">def</span> <span class="nf">my_kernel</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">Y</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">return</span> <span class="n">np</span><span class="o">.</span><span class="n">dot</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">Y</span><span class="o">.</span><span class="n">T</span><span class="p">)</span>
<span class="gp">...</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">kernel</span><span class="o">=</span><span class="n">my_kernel</span><span class="p">)</span>
</pre></div>
</div>
<p></div>
</details></p>
<p><details>
<summary class="btn btn-light">
<strong>Using the Gram matrix</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>You can pass pre-computed kernels by using the <code class="docutils literal notranslate"><span class="pre">kernel='precomputed'</span></code>
option. You should then pass Gram matrix instead of X to the <code class="docutils literal notranslate"><span class="pre">fit</span></code> and
<code class="docutils literal notranslate"><span class="pre">predict</span></code> methods. The kernel values between <em>all</em> training vectors and the
test vectors must be provided:</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="kn">import</span> <span class="n">make_classification</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.model_selection</span> <span class="kn">import</span> <span class="n">train_test_split</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">svm</span>
<span class="gp">>>> </span><span class="n">X</span><span class="p">,</span> <span class="n">y</span> <span class="o">=</span> <span class="n">make_classification</span><span class="p">(</span><span class="n">n_samples</span><span class="o">=</span><span class="mi">10</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X_train</span> <span class="p">,</span> <span class="n">X_test</span> <span class="p">,</span> <span class="n">y_train</span><span class="p">,</span> <span class="n">y_test</span> <span class="o">=</span> <span class="n">train_test_split</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">kernel</span><span class="o">=</span><span class="s1">'precomputed'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="c1"># linear kernel computation</span>
<span class="gp">>>> </span><span class="n">gram_train</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">dot</span><span class="p">(</span><span class="n">X_train</span><span class="p">,</span> <span class="n">X_train</span><span class="o">.</span><span class="n">T</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">gram_train</span><span class="p">,</span> <span class="n">y_train</span><span class="p">)</span>
<span class="go">SVC(kernel='precomputed')</span>
<span class="gp">>>> </span><span class="c1"># predict on training examples</span>
<span class="gp">>>> </span><span class="n">gram_test</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">dot</span><span class="p">(</span><span class="n">X_test</span><span class="p">,</span> <span class="n">X_train</span><span class="o">.</span><span class="n">T</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">gram_test</span><span class="p">)</span>
<span class="go">array([0, 1, 0])</span>
</pre></div>
</div>
<p></div>
</details></p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/svm/plot_custom_kernel.html#sphx-glr-auto-examples-svm-plot-custom-kernel-py"><span class="std std-ref">SVM with custom kernel</span></a>.</p></li>
</ul>
</aside>
</section>
</section>
<section id="mathematical-formulation">
<span id="svm-mathematical-formulation"></span><h2><span class="section-number">1.4.7. </span>Mathematical formulation<a class="headerlink" href="#mathematical-formulation" title="Permalink to this heading">¶</a></h2>
<p>A support vector machine constructs a hyper-plane or set of hyper-planes in a
high or infinite dimensional space, which can be used for
classification, regression or other tasks. Intuitively, a good
separation is achieved by the hyper-plane that has the largest distance
to the nearest training data points of any class (so-called functional
margin), since in general the larger the margin the lower the
generalization error of the classifier. The figure below shows the decision
function for a linearly separable problem, with three samples on the
margin boundaries, called “support vectors”:</p>
<figure class="align-center">
<a class="reference internal image-reference" href="../_images/sphx_glr_plot_separating_hyperplane_001.png"><img alt="../_images/sphx_glr_plot_separating_hyperplane_001.png" src="../_images/sphx_glr_plot_separating_hyperplane_001.png" style="width: 480.0px; height: 360.0px;" /></a>
</figure>
<p>In general, when the problem isn’t linearly separable, the support vectors
are the samples <em>within</em> the margin boundaries.</p>
<p>We recommend <a class="footnote-reference brackets" href="#id15" id="id6" role="doc-noteref"><span class="fn-bracket">[</span>13<span class="fn-bracket">]</span></a> and <a class="footnote-reference brackets" href="#id16" id="id7" role="doc-noteref"><span class="fn-bracket">[</span>14<span class="fn-bracket">]</span></a> as good references for the theory and
practicalities of SVMs.</p>
<section id="svc">
<h3><span class="section-number">1.4.7.1. </span>SVC<a class="headerlink" href="#svc" title="Permalink to this heading">¶</a></h3>
<p>Given training vectors <span class="math notranslate nohighlight">\(x_i \in \mathbb{R}^p\)</span>, i=1,…, n, in two classes, and a
vector <span class="math notranslate nohighlight">\(y \in \{1, -1\}^n\)</span>, our goal is to find <span class="math notranslate nohighlight">\(w \in
\mathbb{R}^p\)</span> and <span class="math notranslate nohighlight">\(b \in \mathbb{R}\)</span> such that the prediction given by
<span class="math notranslate nohighlight">\(\text{sign} (w^T\phi(x) + b)\)</span> is correct for most samples.</p>
<p>SVC solves the following primal problem:</p>
<div class="math notranslate nohighlight">
\[ \begin{align}\begin{aligned}\min_ {w, b, \zeta} \frac{1}{2} w^T w + C \sum_{i=1}^{n} \zeta_i\\\begin{split}\textrm {subject to } & y_i (w^T \phi (x_i) + b) \geq 1 - \zeta_i,\\
& \zeta_i \geq 0, i=1, ..., n\end{split}\end{aligned}\end{align} \]</div>
<p>Intuitively, we’re trying to maximize the margin (by minimizing
<span class="math notranslate nohighlight">\(||w||^2 = w^Tw\)</span>), while incurring a penalty when a sample is
misclassified or within the margin boundary. Ideally, the value <span class="math notranslate nohighlight">\(y_i
(w^T \phi (x_i) + b)\)</span> would be <span class="math notranslate nohighlight">\(\geq 1\)</span> for all samples, which
indicates a perfect prediction. But problems are usually not always perfectly
separable with a hyperplane, so we allow some samples to be at a distance <span class="math notranslate nohighlight">\(\zeta_i\)</span> from
their correct margin boundary. The penalty term <code class="docutils literal notranslate"><span class="pre">C</span></code> controls the strength of
this penalty, and as a result, acts as an inverse regularization parameter
(see note below).</p>
<p>The dual problem to the primal is</p>
<div class="math notranslate nohighlight">
\[ \begin{align}\begin{aligned}\min_{\alpha} \frac{1}{2} \alpha^T Q \alpha - e^T \alpha\\\begin{split}
\textrm {subject to } & y^T \alpha = 0\\
& 0 \leq \alpha_i \leq C, i=1, ..., n\end{split}\end{aligned}\end{align} \]</div>
<p>where <span class="math notranslate nohighlight">\(e\)</span> is the vector of all ones,
and <span class="math notranslate nohighlight">\(Q\)</span> is an <span class="math notranslate nohighlight">\(n\)</span> by <span class="math notranslate nohighlight">\(n\)</span> positive semidefinite matrix,
<span class="math notranslate nohighlight">\(Q_{ij} \equiv y_i y_j K(x_i, x_j)\)</span>, where <span class="math notranslate nohighlight">\(K(x_i, x_j) = \phi (x_i)^T \phi (x_j)\)</span>
is the kernel. The terms <span class="math notranslate nohighlight">\(\alpha_i\)</span> are called the dual coefficients,
and they are upper-bounded by <span class="math notranslate nohighlight">\(C\)</span>.
This dual representation highlights the fact that training vectors are
implicitly mapped into a higher (maybe infinite)
dimensional space by the function <span class="math notranslate nohighlight">\(\phi\)</span>: see <a class="reference external" href="https://en.wikipedia.org/wiki/Kernel_method">kernel trick</a>.</p>
<p>Once the optimization problem is solved, the output of
<a class="reference internal" href="../glossary.html#term-decision_function"><span class="xref std std-term">decision_function</span></a> for a given sample <span class="math notranslate nohighlight">\(x\)</span> becomes:</p>
<div class="math notranslate nohighlight">
\[\sum_{i\in SV} y_i \alpha_i K(x_i, x) + b,\]</div>
<p>and the predicted class correspond to its sign. We only need to sum over the
support vectors (i.e. the samples that lie within the margin) because the
dual coefficients <span class="math notranslate nohighlight">\(\alpha_i\)</span> are zero for the other samples.</p>
<p>These parameters can be accessed through the attributes <code class="docutils literal notranslate"><span class="pre">dual_coef_</span></code>
which holds the product <span class="math notranslate nohighlight">\(y_i \alpha_i\)</span>, <code class="docutils literal notranslate"><span class="pre">support_vectors_</span></code> which
holds the support vectors, and <code class="docutils literal notranslate"><span class="pre">intercept_</span></code> which holds the independent
term <span class="math notranslate nohighlight">\(b\)</span></p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>While SVM models derived from <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a> and <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/liblinear/">liblinear</a> use <code class="docutils literal notranslate"><span class="pre">C</span></code> as
regularization parameter, most other estimators use <code class="docutils literal notranslate"><span class="pre">alpha</span></code>. The exact
equivalence between the amount of regularization of two models depends on
the exact objective function optimized by the model. For example, when the
estimator used is <a class="reference internal" href="generated/sklearn.linear_model.Ridge.html#sklearn.linear_model.Ridge" title="sklearn.linear_model.Ridge"><code class="xref py py-class docutils literal notranslate"><span class="pre">Ridge</span></code></a> regression,
the relation between them is given as <span class="math notranslate nohighlight">\(C = \frac{1}{alpha}\)</span>.</p>
</div>
<p><details>
<summary class="btn btn-light">
<strong>LinearSVC</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>The primal problem can be equivalently formulated as</p>
<div class="math notranslate nohighlight">
\[\min_ {w, b} \frac{1}{2} w^T w + C \sum_{i=1}^{n}\max(0, 1 - y_i (w^T \phi(x_i) + b)),\]</div>
<p>where we make use of the <a class="reference external" href="https://en.wikipedia.org/wiki/Hinge_loss">hinge loss</a>. This is the form that is
directly optimized by <a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a>, but unlike the dual form, this one
does not involve inner products between samples, so the famous kernel trick
cannot be applied. This is why only the linear kernel is supported by
<a class="reference internal" href="generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC" title="sklearn.svm.LinearSVC"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVC</span></code></a> (<span class="math notranslate nohighlight">\(\phi\)</span> is the identity function).</p>
<p></div>
</details></p>
<p id="nu-svc"><details>
<summary class="btn btn-light">
<strong>NuSVC</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>The <span class="math notranslate nohighlight">\(\nu\)</span>-SVC formulation <a class="footnote-reference brackets" href="#id17" id="id8" role="doc-noteref"><span class="fn-bracket">[</span>15<span class="fn-bracket">]</span></a> is a reparameterization of the
<span class="math notranslate nohighlight">\(C\)</span>-SVC and therefore mathematically equivalent.</p>
<p>We introduce a new parameter <span class="math notranslate nohighlight">\(\nu\)</span> (instead of <span class="math notranslate nohighlight">\(C\)</span>) which
controls the number of support vectors and <em>margin errors</em>:
<span class="math notranslate nohighlight">\(\nu \in (0, 1]\)</span> is an upper bound on the fraction of margin errors and
a lower bound of the fraction of support vectors. A margin error corresponds
to a sample that lies on the wrong side of its margin boundary: it is either
misclassified, or it is correctly classified but does not lie beyond the
margin.</p>
<p></div>
</details></p>
</section>
<section id="svr">
<h3><span class="section-number">1.4.7.2. </span>SVR<a class="headerlink" href="#svr" title="Permalink to this heading">¶</a></h3>
<p>Given training vectors <span class="math notranslate nohighlight">\(x_i \in \mathbb{R}^p\)</span>, i=1,…, n, and a
vector <span class="math notranslate nohighlight">\(y \in \mathbb{R}^n\)</span> <span class="math notranslate nohighlight">\(\varepsilon\)</span>-SVR solves the following primal problem:</p>
<div class="math notranslate nohighlight">
\[ \begin{align}\begin{aligned}\min_ {w, b, \zeta, \zeta^*} \frac{1}{2} w^T w + C \sum_{i=1}^{n} (\zeta_i + \zeta_i^*)\\\begin{split}\textrm {subject to } & y_i - w^T \phi (x_i) - b \leq \varepsilon + \zeta_i,\\
& w^T \phi (x_i) + b - y_i \leq \varepsilon + \zeta_i^*,\\
& \zeta_i, \zeta_i^* \geq 0, i=1, ..., n\end{split}\end{aligned}\end{align} \]</div>
<p>Here, we are penalizing samples whose prediction is at least <span class="math notranslate nohighlight">\(\varepsilon\)</span>
away from their true target. These samples penalize the objective by
<span class="math notranslate nohighlight">\(\zeta_i\)</span> or <span class="math notranslate nohighlight">\(\zeta_i^*\)</span>, depending on whether their predictions
lie above or below the <span class="math notranslate nohighlight">\(\varepsilon\)</span> tube.</p>
<p>The dual problem is</p>
<div class="math notranslate nohighlight">
\[ \begin{align}\begin{aligned}\min_{\alpha, \alpha^*} \frac{1}{2} (\alpha - \alpha^*)^T Q (\alpha - \alpha^*) + \varepsilon e^T (\alpha + \alpha^*) - y^T (\alpha - \alpha^*)\\\begin{split}
\textrm {subject to } & e^T (\alpha - \alpha^*) = 0\\
& 0 \leq \alpha_i, \alpha_i^* \leq C, i=1, ..., n\end{split}\end{aligned}\end{align} \]</div>
<p>where <span class="math notranslate nohighlight">\(e\)</span> is the vector of all ones,
<span class="math notranslate nohighlight">\(Q\)</span> is an <span class="math notranslate nohighlight">\(n\)</span> by <span class="math notranslate nohighlight">\(n\)</span> positive semidefinite matrix,
<span class="math notranslate nohighlight">\(Q_{ij} \equiv K(x_i, x_j) = \phi (x_i)^T \phi (x_j)\)</span>
is the kernel. Here training vectors are implicitly mapped into a higher
(maybe infinite) dimensional space by the function <span class="math notranslate nohighlight">\(\phi\)</span>.</p>
<p>The prediction is:</p>
<div class="math notranslate nohighlight">
\[\sum_{i \in SV}(\alpha_i - \alpha_i^*) K(x_i, x) + b\]</div>
<p>These parameters can be accessed through the attributes <code class="docutils literal notranslate"><span class="pre">dual_coef_</span></code>
which holds the difference <span class="math notranslate nohighlight">\(\alpha_i - \alpha_i^*\)</span>, <code class="docutils literal notranslate"><span class="pre">support_vectors_</span></code> which
holds the support vectors, and <code class="docutils literal notranslate"><span class="pre">intercept_</span></code> which holds the independent
term <span class="math notranslate nohighlight">\(b\)</span></p>
<p><details>
<summary class="btn btn-light">
<strong>LinearSVR</strong>
<span class="tooltiptext">Click for more details</span>
</summary>
<div class="card"></p>
<p>The primal problem can be equivalently formulated as</p>
<div class="math notranslate nohighlight">
\[\min_ {w, b} \frac{1}{2} w^T w + C \sum_{i=1}^{n}\max(0, |y_i - (w^T \phi(x_i) + b)| - \varepsilon),\]</div>
<p>where we make use of the epsilon-insensitive loss, i.e. errors of less than
<span class="math notranslate nohighlight">\(\varepsilon\)</span> are ignored. This is the form that is directly optimized
by <a class="reference internal" href="generated/sklearn.svm.LinearSVR.html#sklearn.svm.LinearSVR" title="sklearn.svm.LinearSVR"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearSVR</span></code></a>.</p>
<p></div>
</details></p>
</section>
</section>
<section id="implementation-details">
<span id="svm-implementation-details"></span><h2><span class="section-number">1.4.8. </span>Implementation details<a class="headerlink" href="#implementation-details" title="Permalink to this heading">¶</a></h2>
<p>Internally, we use <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/libsvm/">libsvm</a> <a class="footnote-reference brackets" href="#id14" id="id9" role="doc-noteref"><span class="fn-bracket">[</span>12<span class="fn-bracket">]</span></a> and <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/liblinear/">liblinear</a> <a class="footnote-reference brackets" href="#id13" id="id10" role="doc-noteref"><span class="fn-bracket">[</span>11<span class="fn-bracket">]</span></a> to handle all
computations. These libraries are wrapped using C and Cython.
For a description of the implementation and details of the algorithms
used, please refer to their respective papers.</p>
<aside class="topic">
<p class="topic-title">References:</p>
<aside class="footnote-list brackets">
<aside class="footnote brackets" id="id11" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id2">9</a><span class="fn-bracket">]</span></span>
<p>Platt <a class="reference external" href="https://www.cs.colorado.edu/~mozer/Teaching/syllabi/6622/papers/Platt1999.pdf">“Probabilistic outputs for SVMs and comparisons to
regularized likelihood methods”</a>.</p>
</aside>
<aside class="footnote brackets" id="id12" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id3">10</a><span class="fn-bracket">]</span></span>
<p>Wu, Lin and Weng, <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/papers/svmprob/svmprob.pdf">“Probability estimates for multi-class
classification by pairwise coupling”</a>, JMLR
5:975-1005, 2004.</p>
</aside>
<aside class="footnote brackets" id="id13" role="note">
<span class="label"><span class="fn-bracket">[</span>11<span class="fn-bracket">]</span></span>
<span class="backrefs">(<a role="doc-backlink" href="#id4">1</a>,<a role="doc-backlink" href="#id10">2</a>)</span>
<p>Fan, Rong-En, et al.,
<a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/papers/liblinear.pdf">“LIBLINEAR: A library for large linear classification.”</a>,
Journal of machine learning research 9.Aug (2008): 1871-1874.</p>
</aside>
<aside class="footnote brackets" id="id14" role="note">
<span class="label"><span class="fn-bracket">[</span>12<span class="fn-bracket">]</span></span>
<span class="backrefs">(<a role="doc-backlink" href="#id5">1</a>,<a role="doc-backlink" href="#id9">2</a>)</span>
<p>Chang and Lin, <a class="reference external" href="https://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf">LIBSVM: A Library for Support Vector Machines</a>.</p>
</aside>
<aside class="footnote brackets" id="id15" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id6">13</a><span class="fn-bracket">]</span></span>
<p>Bishop, <a class="reference external" href="https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-and-Machine-Learning-2006.pdf">Pattern recognition and machine learning</a>,
chapter 7 Sparse Kernel Machines</p>
</aside>
<aside class="footnote brackets" id="id16" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id7">14</a><span class="fn-bracket">]</span></span>
<p><a class="reference external" href="https://doi.org/10.1023/B:STCO.0000035301.49549.88">“A Tutorial on Support Vector Regression”</a>
Alex J. Smola, Bernhard Schölkopf - Statistics and Computing archive
Volume 14 Issue 3, August 2004, p. 199-222.</p>
</aside>
<aside class="footnote brackets" id="id17" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id8">15</a><span class="fn-bracket">]</span></span>
<p>Schölkopf et. al <a class="reference external" href="https://www.stat.purdue.edu/~yuzhu/stat598m3/Papers/NewSVM.pdf">New Support Vector Algorithms</a></p>
</aside>
<aside class="footnote brackets" id="id18" role="note">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id1">16</a><span class="fn-bracket">]</span></span>
<p>Crammer and Singer <a class="reference external" href="http://jmlr.csail.mit.edu/papers/volume2/crammer01a/crammer01a.pdf">On the Algorithmic Implementation ofMulticlass
Kernel-based Vector Machines</a>,
JMLR 2001.</p>
</aside>
</aside>
</aside>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2023, scikit-learn developers (BSD License).
<a href="../_sources/modules/svm.rst.txt" rel="nofollow">Show this page source</a>
</footer>
</div>
</div>
</div>
<script src="../_static/js/vendor/bootstrap.min.js"></script>
<script>
window.ga=window.ga||function(){(ga.q=ga.q||[]).push(arguments)};ga.l=+new Date;
ga('create', 'UA-22606712-2', 'auto');
ga('set', 'anonymizeIp', true);
ga('send', 'pageview');
</script>
<script async src='https://www.google-analytics.com/analytics.js'></script>
<script defer data-domain="scikit-learn.org" src="https://views.scientific-python.org/js/script.js">
</script>