-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathcomputational_performance.html
1131 lines (895 loc) · 61.2 KB
/
computational_performance.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en" data-content_root="../" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<meta property="og:title" content="8.2. Computational Performance" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/computing/computational_performance.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="For some applications the performance (mainly latency and throughput at prediction time) of estimators is crucial. It may also be of interest to consider the training throughput but this is often l..." />
<meta property="og:image" content="https://scikit-learn/stable/computing/auto_examples/applications/images/sphx_glr_plot_prediction_latency_001.png" />
<meta property="og:image:alt" content="atomic_prediction_latency" />
<meta name="description" content="For some applications the performance (mainly latency and throughput at prediction time) of estimators is crucial. It may also be of interest to consider the training throughput but this is often l..." />
<title>8.2. Computational Performance — scikit-learn 1.5.2 documentation</title>
<script data-cfasync="false">
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
</script>
<!-- Loaded before other Sphinx assets -->
<link href="../_static/styles/theme.css?digest=dfe6caa3a7d634c4db9b" rel="stylesheet" />
<link href="../_static/styles/bootstrap.css?digest=dfe6caa3a7d634c4db9b" rel="stylesheet" />
<link href="../_static/styles/pydata-sphinx-theme.css?digest=dfe6caa3a7d634c4db9b" rel="stylesheet" />
<link href="../_static/vendor/fontawesome/6.5.2/css/all.min.css?digest=dfe6caa3a7d634c4db9b" rel="stylesheet" />
<link rel="preload" as="font" type="font/woff2" crossorigin href="../_static/vendor/fontawesome/6.5.2/webfonts/fa-solid-900.woff2" />
<link rel="preload" as="font" type="font/woff2" crossorigin href="../_static/vendor/fontawesome/6.5.2/webfonts/fa-brands-400.woff2" />
<link rel="preload" as="font" type="font/woff2" crossorigin href="../_static/vendor/fontawesome/6.5.2/webfonts/fa-regular-400.woff2" />
<link rel="stylesheet" type="text/css" href="../_static/pygments.css?v=a746c00c" />
<link rel="stylesheet" type="text/css" href="../_static/copybutton.css?v=76b2166b" />
<link rel="stylesheet" type="text/css" href="../_static/plot_directive.css" />
<link rel="stylesheet" type="text/css" href="https://fonts.googleapis.com/css?family=Vibur" />
<link rel="stylesheet" type="text/css" href="../_static/jupyterlite_sphinx.css?v=ca70e7f1" />
<link rel="stylesheet" type="text/css" href="../_static/sg_gallery.css?v=d2d258e8" />
<link rel="stylesheet" type="text/css" href="../_static/sg_gallery-binder.css?v=f4aeca0c" />
<link rel="stylesheet" type="text/css" href="../_static/sg_gallery-dataframe.css?v=2082cf3c" />
<link rel="stylesheet" type="text/css" href="../_static/sg_gallery-rendered-html.css?v=1277b6f3" />
<link rel="stylesheet" type="text/css" href="../_static/sphinx-design.min.css?v=95c83b7e" />
<link rel="stylesheet" type="text/css" href="../_static/styles/colors.css?v=cc94ab7d" />
<link rel="stylesheet" type="text/css" href="../_static/styles/custom.css?v=e4cb1417" />
<!-- Pre-loaded scripts that we'll load fully later -->
<link rel="preload" as="script" href="../_static/scripts/bootstrap.js?digest=dfe6caa3a7d634c4db9b" />
<link rel="preload" as="script" href="../_static/scripts/pydata-sphinx-theme.js?digest=dfe6caa3a7d634c4db9b" />
<script src="../_static/vendor/fontawesome/6.5.2/js/all.min.js?digest=dfe6caa3a7d634c4db9b"></script>
<script src="../_static/documentation_options.js?v=73275c37"></script>
<script src="../_static/doctools.js?v=9a2dae69"></script>
<script src="../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../_static/clipboard.min.js?v=a7894cd8"></script>
<script src="../_static/copybutton.js?v=97f0b27d"></script>
<script src="../_static/jupyterlite_sphinx.js?v=d6bdf5f8"></script>
<script src="../_static/design-tabs.js?v=f930bc37"></script>
<script data-domain="scikit-learn.org" defer="defer" src="https://views.scientific-python.org/js/script.js"></script>
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml.js"></script>
<script>DOCUMENTATION_OPTIONS.pagename = 'computing/computational_performance';</script>
<script>
DOCUMENTATION_OPTIONS.theme_version = '0.15.4';
DOCUMENTATION_OPTIONS.theme_switcher_json_url = 'https://scikit-learn.org/dev/_static/versions.json';
DOCUMENTATION_OPTIONS.theme_switcher_version_match = '1.5.2';
DOCUMENTATION_OPTIONS.show_version_warning_banner = true;
</script>
<script src="../_static/scripts/dropdown.js?v=e2048168"></script>
<script src="../_static/scripts/version-switcher.js?v=a6dd8357"></script>
<link rel="canonical" href="https://scikit-learn.org/stable/computing/computational_performance.html" />
<link rel="icon" href="../_static/favicon.ico"/>
<link rel="author" title="About these documents" href="../about.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="8.3. Parallelism, resource management, and configuration" href="parallelism.html" />
<link rel="prev" title="8.1. Strategies to scale computationally: bigger data" href="scaling_strategies.html" />
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<meta name="docsearch:language" content="en"/>
</head>
<body data-bs-spy="scroll" data-bs-target=".bd-toc-nav" data-offset="180" data-bs-root-margin="0px 0px -60%" data-default-mode="">
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
<div id="pst-scroll-pixel-helper"></div>
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
<input type="checkbox"
class="sidebar-toggle"
id="pst-primary-sidebar-checkbox"/>
<label class="overlay overlay-primary" for="pst-primary-sidebar-checkbox"></label>
<input type="checkbox"
class="sidebar-toggle"
id="pst-secondary-sidebar-checkbox"/>
<label class="overlay overlay-secondary" for="pst-secondary-sidebar-checkbox"></label>
<div class="search-button__wrapper">
<div class="search-button__overlay"></div>
<div class="search-button__search-container">
<form class="bd-search d-flex align-items-center"
action="../search.html"
method="get">
<i class="fa-solid fa-magnifying-glass"></i>
<input type="search"
class="form-control"
name="q"
id="search-input"
placeholder="Search the docs ..."
aria-label="Search the docs ..."
autocomplete="off"
autocorrect="off"
autocapitalize="off"
spellcheck="false"/>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
</form></div>
</div>
<div class="pst-async-banner-revealer d-none">
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
</div>
<header class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
<div class="bd-header__inner bd-page-width">
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
<span class="fa-solid fa-bars"></span>
</button>
<div class=" navbar-header-items__start">
<div class="navbar-item">
<a class="navbar-brand logo" href="../index.html">
<img src="../_static/scikit-learn-logo-small.png" class="logo__image only-light" alt="scikit-learn homepage"/>
<script>document.write(`<img src="../_static/scikit-learn-logo-small.png" class="logo__image only-dark" alt="scikit-learn homepage"/>`);</script>
</a></div>
</div>
<div class=" navbar-header-items">
<div class="me-auto navbar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item ">
<a class="nav-link nav-internal" href="../install.html">
Install
</a>
</li>
<li class="nav-item current active">
<a class="nav-link nav-internal" href="../user_guide.html">
User Guide
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../api/index.html">
API
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../auto_examples/index.html">
Examples
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://blog.scikit-learn.org/">
Community
</a>
</li>
<li class="nav-item dropdown">
<button class="btn dropdown-toggle nav-item" type="button" data-bs-toggle="dropdown" aria-expanded="false" aria-controls="pst-nav-more-links">
More
</button>
<ul id="pst-nav-more-links" class="dropdown-menu">
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../getting_started.html">
Getting Started
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../whats_new.html">
Release History
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../glossary.html">
Glossary
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-external" href="https://scikit-learn.org/dev/developers/index.html">
Development
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../faq.html">
FAQ
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../support.html">
Support
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../related_projects.html">
Related Projects
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../roadmap.html">
Roadmap
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../governance.html">
Governance
</a>
</li>
<li class=" ">
<a class="nav-link dropdown-item nav-internal" href="../about.html">
About us
</a>
</li>
</ul>
</li>
</ul>
</nav></div>
</div>
<div class="navbar-header-items__end">
<div class="navbar-item navbar-persistent--container">
<script>
document.write(`
<button class="btn btn-sm pst-navbar-icon search-button search-button__button" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass fa-lg"></i>
</button>
`);
</script>
</div>
<div class="navbar-item">
<script>
document.write(`
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button" title="light/dark" aria-label="light/dark" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto"></i>
</button>
`);
</script></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/scikit-learn/scikit-learn" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
<div class="navbar-item">
<script>
document.write(`
<div class="version-switcher__container dropdown">
<button id="pst-version-switcher-button-2"
type="button"
class="version-switcher__button btn btn-sm dropdown-toggle"
data-bs-toggle="dropdown"
aria-haspopup="listbox"
aria-controls="pst-version-switcher-list-2"
aria-label="Version switcher list"
>
Choose version <!-- this text may get changed later by javascript -->
<span class="caret"></span>
</button>
<div id="pst-version-switcher-list-2"
class="version-switcher__menu dropdown-menu list-group-flush py-0"
role="listbox" aria-labelledby="pst-version-switcher-button-2">
<!-- dropdown will be populated by javascript on page load -->
</div>
</div>
`);
</script></div>
</div>
</div>
<div class="navbar-persistent--mobile">
<script>
document.write(`
<button class="btn btn-sm pst-navbar-icon search-button search-button__button" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass fa-lg"></i>
</button>
`);
</script>
</div>
<button class="pst-navbar-icon sidebar-toggle secondary-toggle" aria-label="On this page">
<span class="fa-solid fa-outdent"></span>
</button>
</div>
</header>
<div class="bd-container">
<div class="bd-container__inner bd-page-width">
<div class="bd-sidebar-primary bd-sidebar">
<div class="sidebar-header-items sidebar-primary__section">
<div class="sidebar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item ">
<a class="nav-link nav-internal" href="../install.html">
Install
</a>
</li>
<li class="nav-item current active">
<a class="nav-link nav-internal" href="../user_guide.html">
User Guide
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../api/index.html">
API
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../auto_examples/index.html">
Examples
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://blog.scikit-learn.org/">
Community
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../getting_started.html">
Getting Started
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../whats_new.html">
Release History
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../glossary.html">
Glossary
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://scikit-learn.org/dev/developers/index.html">
Development
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../faq.html">
FAQ
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../support.html">
Support
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../related_projects.html">
Related Projects
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../roadmap.html">
Roadmap
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../governance.html">
Governance
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="../about.html">
About us
</a>
</li>
</ul>
</nav></div>
</div>
<div class="sidebar-header-items__end">
<div class="navbar-item">
<script>
document.write(`
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button" title="light/dark" aria-label="light/dark" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto"></i>
</button>
`);
</script></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/scikit-learn/scikit-learn" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
<div class="navbar-item">
<script>
document.write(`
<div class="version-switcher__container dropdown">
<button id="pst-version-switcher-button-3"
type="button"
class="version-switcher__button btn btn-sm dropdown-toggle"
data-bs-toggle="dropdown"
aria-haspopup="listbox"
aria-controls="pst-version-switcher-list-3"
aria-label="Version switcher list"
>
Choose version <!-- this text may get changed later by javascript -->
<span class="caret"></span>
</button>
<div id="pst-version-switcher-list-3"
class="version-switcher__menu dropdown-menu list-group-flush py-0"
role="listbox" aria-labelledby="pst-version-switcher-button-3">
<!-- dropdown will be populated by javascript on page load -->
</div>
</div>
`);
</script></div>
</div>
</div>
<div class="sidebar-primary-items__start sidebar-primary__section">
<div class="sidebar-primary-item">
<nav class="bd-docs-nav bd-links"
aria-label="Section Navigation">
<p class="bd-links__title" role="heading" aria-level="1">Section Navigation</p>
<div class="bd-toc-item navbar-nav"><ul class="current nav bd-sidenav">
<li class="toctree-l1 has-children"><a class="reference internal" href="../supervised_learning.html">1. Supervised learning</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/linear_model.html">1.1. Linear Models</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/lda_qda.html">1.2. Linear and Quadratic Discriminant Analysis</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/kernel_ridge.html">1.3. Kernel ridge regression</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/svm.html">1.4. Support Vector Machines</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/sgd.html">1.5. Stochastic Gradient Descent</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/neighbors.html">1.6. Nearest Neighbors</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/gaussian_process.html">1.7. Gaussian Processes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/cross_decomposition.html">1.8. Cross decomposition</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/naive_bayes.html">1.9. Naive Bayes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/tree.html">1.10. Decision Trees</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/ensemble.html">1.11. Ensembles: Gradient boosting, random forests, bagging, voting, stacking</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/multiclass.html">1.12. Multiclass and multioutput algorithms</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/feature_selection.html">1.13. Feature selection</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/semi_supervised.html">1.14. Semi-supervised learning</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/isotonic.html">1.15. Isotonic regression</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/calibration.html">1.16. Probability calibration</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/neural_networks_supervised.html">1.17. Neural network models (supervised)</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../unsupervised_learning.html">2. Unsupervised learning</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/mixture.html">2.1. Gaussian mixture models</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/manifold.html">2.2. Manifold learning</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/clustering.html">2.3. Clustering</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/biclustering.html">2.4. Biclustering</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/decomposition.html">2.5. Decomposing signals in components (matrix factorization problems)</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/covariance.html">2.6. Covariance estimation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/outlier_detection.html">2.7. Novelty and Outlier Detection</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/density.html">2.8. Density Estimation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/neural_networks_unsupervised.html">2.9. Neural network models (unsupervised)</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../model_selection.html">3. Model selection and evaluation</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/cross_validation.html">3.1. Cross-validation: evaluating estimator performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/grid_search.html">3.2. Tuning the hyper-parameters of an estimator</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/classification_threshold.html">3.3. Tuning the decision threshold for class prediction</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/model_evaluation.html">3.4. Metrics and scoring: quantifying the quality of predictions</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/learning_curve.html">3.5. Validation curves: plotting scores to evaluate models</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../inspection.html">4. Inspection</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/partial_dependence.html">4.1. Partial Dependence and Individual Conditional Expectation plots</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/permutation_importance.html">4.2. Permutation feature importance</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../visualizations.html">5. Visualizations</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../data_transforms.html">6. Dataset transformations</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/compose.html">6.1. Pipelines and composite estimators</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/feature_extraction.html">6.2. Feature extraction</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/preprocessing.html">6.3. Preprocessing data</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/impute.html">6.4. Imputation of missing values</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/unsupervised_reduction.html">6.5. Unsupervised dimensionality reduction</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/random_projection.html">6.6. Random Projection</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/kernel_approximation.html">6.7. Kernel Approximation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/metrics.html">6.8. Pairwise metrics, Affinities and Kernels</a></li>
<li class="toctree-l2"><a class="reference internal" href="../modules/preprocessing_targets.html">6.9. Transforming the prediction target (<code class="docutils literal notranslate"><span class="pre">y</span></code>)</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../datasets.html">7. Dataset loading utilities</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../datasets/toy_dataset.html">7.1. Toy datasets</a></li>
<li class="toctree-l2"><a class="reference internal" href="../datasets/real_world.html">7.2. Real world datasets</a></li>
<li class="toctree-l2"><a class="reference internal" href="../datasets/sample_generators.html">7.3. Generated datasets</a></li>
<li class="toctree-l2"><a class="reference internal" href="../datasets/loading_other_datasets.html">7.4. Loading other datasets</a></li>
</ul>
</details></li>
<li class="toctree-l1 current active has-children"><a class="reference internal" href="../computing.html">8. Computing with scikit-learn</a><details open="open"><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="scaling_strategies.html">8.1. Strategies to scale computationally: bigger data</a></li>
<li class="toctree-l2 current active"><a class="current reference internal" href="#">8.2. Computational Performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="parallelism.html">8.3. Parallelism, resource management, and configuration</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../model_persistence.html">9. Model persistence</a></li>
<li class="toctree-l1"><a class="reference internal" href="../common_pitfalls.html">10. Common pitfalls and recommended practices</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../dispatching.html">11. Dispatching</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../modules/array_api.html">11.1. Array API support (experimental)</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../machine_learning_map.html">12. Choosing the right estimator</a></li>
<li class="toctree-l1"><a class="reference internal" href="../presentations.html">13. External Resources, Videos and Talks</a></li>
</ul>
</div>
</nav></div>
</div>
<div class="sidebar-primary-items__end sidebar-primary__section">
</div>
<div id="rtd-footer-container"></div>
</div>
<main id="main-content" class="bd-main" role="main">
<div class="bd-content">
<div class="bd-article-container">
<div class="bd-header-article d-print-none">
<div class="header-article-items header-article__inner">
<div class="header-article-items__start">
<div class="header-article-item">
<nav aria-label="Breadcrumb" class="d-print-none">
<ul class="bd-breadcrumbs">
<li class="breadcrumb-item breadcrumb-home">
<a href="../index.html" class="nav-link" aria-label="Home">
<i class="fa-solid fa-home"></i>
</a>
</li>
<li class="breadcrumb-item"><a href="../user_guide.html" class="nav-link">User Guide</a></li>
<li class="breadcrumb-item"><a href="../computing.html" class="nav-link"><span class="section-number">8. </span>Computing with scikit-learn</a></li>
<li class="breadcrumb-item active" aria-current="page"><span...</li>
</ul>
</nav>
</div>
</div>
</div>
</div>
<div id="searchbox"></div>
<article class="bd-article">
<section id="computational-performance">
<span id="id1"></span><h1><span class="section-number">8.2. </span>Computational Performance<a class="headerlink" href="#computational-performance" title="Link to this heading">#</a></h1>
<p>For some applications the performance (mainly latency and throughput at
prediction time) of estimators is crucial. It may also be of interest to
consider the training throughput but this is often less important in a
production setup (where it often takes place offline).</p>
<p>We will review here the orders of magnitude you can expect from a number of
scikit-learn estimators in different contexts and provide some tips and
tricks for overcoming performance bottlenecks.</p>
<p>Prediction latency is measured as the elapsed time necessary to make a
prediction (e.g. in micro-seconds). Latency is often viewed as a distribution
and operations engineers often focus on the latency at a given percentile of
this distribution (e.g. the 90 percentile).</p>
<p>Prediction throughput is defined as the number of predictions the software can
deliver in a given amount of time (e.g. in predictions per second).</p>
<p>An important aspect of performance optimization is also that it can hurt
prediction accuracy. Indeed, simpler models (e.g. linear instead of
non-linear, or with fewer parameters) often run faster but are not always able
to take into account the same exact properties of the data as more complex ones.</p>
<section id="prediction-latency">
<h2><span class="section-number">8.2.1. </span>Prediction Latency<a class="headerlink" href="#prediction-latency" title="Link to this heading">#</a></h2>
<p>One of the most straight-forward concerns one may have when using/choosing a
machine learning toolkit is the latency at which predictions can be made in a
production environment.</p>
<p>The main factors that influence the prediction latency are</p>
<ol class="arabic simple">
<li><p>Number of features</p></li>
<li><p>Input data representation and sparsity</p></li>
<li><p>Model complexity</p></li>
<li><p>Feature extraction</p></li>
</ol>
<p>A last major parameter is also the possibility to do predictions in bulk or
one-at-a-time mode.</p>
<section id="bulk-versus-atomic-mode">
<h3><span class="section-number">8.2.1.1. </span>Bulk versus Atomic mode<a class="headerlink" href="#bulk-versus-atomic-mode" title="Link to this heading">#</a></h3>
<p>In general doing predictions in bulk (many instances at the same time) is
more efficient for a number of reasons (branching predictability, CPU cache,
linear algebra libraries optimizations etc.). Here we see on a setting
with few features that independently of estimator choice the bulk mode is
always faster, and for some of them by 1 to 2 orders of magnitude:</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_prediction_latency.html"><img alt="atomic_prediction_latency" src="../_images/sphx_glr_plot_prediction_latency_001.png" style="width: 800.0px; height: 480.0px;" /></a></strong></p><p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_prediction_latency.html"><img alt="bulk_prediction_latency" src="../_images/sphx_glr_plot_prediction_latency_002.png" style="width: 800.0px; height: 480.0px;" /></a></strong></p><p>To benchmark different estimators for your case you can simply change the
<code class="docutils literal notranslate"><span class="pre">n_features</span></code> parameter in this example:
<a class="reference internal" href="../auto_examples/applications/plot_prediction_latency.html#sphx-glr-auto-examples-applications-plot-prediction-latency-py"><span class="std std-ref">Prediction Latency</span></a>. This should give
you an estimate of the order of magnitude of the prediction latency.</p>
</section>
<section id="configuring-scikit-learn-for-reduced-validation-overhead">
<h3><span class="section-number">8.2.1.2. </span>Configuring Scikit-learn for reduced validation overhead<a class="headerlink" href="#configuring-scikit-learn-for-reduced-validation-overhead" title="Link to this heading">#</a></h3>
<p>Scikit-learn does some validation on data that increases the overhead per
call to <code class="docutils literal notranslate"><span class="pre">predict</span></code> and similar functions. In particular, checking that
features are finite (not NaN or infinite) involves a full pass over the
data. If you ensure that your data is acceptable, you may suppress
checking for finiteness by setting the environment variable
<code class="docutils literal notranslate"><span class="pre">SKLEARN_ASSUME_FINITE</span></code> to a non-empty string before importing
scikit-learn, or configure it in Python with <a class="reference internal" href="../modules/generated/sklearn.set_config.html#sklearn.set_config" title="sklearn.set_config"><code class="xref py py-func docutils literal notranslate"><span class="pre">set_config</span></code></a>.
For more control than these global settings, a <a class="reference internal" href="../modules/generated/sklearn.config_context.html#sklearn.config_context" title="sklearn.config_context"><code class="xref py py-func docutils literal notranslate"><span class="pre">config_context</span></code></a>
allows you to set this configuration within a specified context:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">sklearn</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="n">sklearn</span><span class="o">.</span><span class="n">config_context</span><span class="p">(</span><span class="n">assume_finite</span><span class="o">=</span><span class="kc">True</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">pass</span> <span class="c1"># do learning/prediction here with reduced validation</span>
</pre></div>
</div>
<p>Note that this will affect all uses of
<a class="reference internal" href="../modules/generated/sklearn.utils.assert_all_finite.html#sklearn.utils.assert_all_finite" title="sklearn.utils.assert_all_finite"><code class="xref py py-func docutils literal notranslate"><span class="pre">assert_all_finite</span></code></a> within the context.</p>
</section>
<section id="influence-of-the-number-of-features">
<h3><span class="section-number">8.2.1.3. </span>Influence of the Number of Features<a class="headerlink" href="#influence-of-the-number-of-features" title="Link to this heading">#</a></h3>
<p>Obviously when the number of features increases so does the memory
consumption of each example. Indeed, for a matrix of <span class="math notranslate nohighlight">\(M\)</span> instances
with <span class="math notranslate nohighlight">\(N\)</span> features, the space complexity is in <span class="math notranslate nohighlight">\(O(NM)\)</span>.
From a computing perspective it also means that the number of basic operations
(e.g., multiplications for vector-matrix products in linear models) increases
too. Here is a graph of the evolution of the prediction latency with the
number of features:</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_prediction_latency.html"><img alt="influence_of_n_features_on_latency" src="../_images/sphx_glr_plot_prediction_latency_003.png" style="width: 800.0px; height: 480.0px;" /></a></strong></p><p>Overall you can expect the prediction time to increase at least linearly with
the number of features (non-linear cases can happen depending on the global
memory footprint and estimator).</p>
</section>
<section id="influence-of-the-input-data-representation">
<h3><span class="section-number">8.2.1.4. </span>Influence of the Input Data Representation<a class="headerlink" href="#influence-of-the-input-data-representation" title="Link to this heading">#</a></h3>
<p>Scipy provides sparse matrix data structures which are optimized for storing
sparse data. The main feature of sparse formats is that you don’t store zeros
so if your data is sparse then you use much less memory. A non-zero value in
a sparse (<a class="reference external" href="https://docs.scipy.org/doc/scipy/reference/sparse.html">CSR or CSC</a>)
representation will only take on average one 32bit integer position + the 64
bit floating point value + an additional 32bit per row or column in the matrix.
Using sparse input on a dense (or sparse) linear model can speedup prediction
by quite a bit as only the non zero valued features impact the dot product
and thus the model predictions. Hence if you have 100 non zeros in 1e6
dimensional space, you only need 100 multiply and add operation instead of 1e6.</p>
<p>Calculation over a dense representation, however, may leverage highly optimized
vector operations and multithreading in BLAS, and tends to result in fewer CPU
cache misses. So the sparsity should typically be quite high (10% non-zeros
max, to be checked depending on the hardware) for the sparse input
representation to be faster than the dense input representation on a machine
with many CPUs and an optimized BLAS implementation.</p>
<p>Here is sample code to test the sparsity of your input:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">sparsity_ratio</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="k">return</span> <span class="mf">1.0</span> <span class="o">-</span> <span class="n">np</span><span class="o">.</span><span class="n">count_nonzero</span><span class="p">(</span><span class="n">X</span><span class="p">)</span> <span class="o">/</span> <span class="nb">float</span><span class="p">(</span><span class="n">X</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">*</span> <span class="n">X</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="nb">print</span><span class="p">(</span><span class="s2">"input sparsity ratio:"</span><span class="p">,</span> <span class="n">sparsity_ratio</span><span class="p">(</span><span class="n">X</span><span class="p">))</span>
</pre></div>
</div>
<p>As a rule of thumb you can consider that if the sparsity ratio is greater
than 90% you can probably benefit from sparse formats. Check Scipy’s sparse
matrix formats <a class="reference external" href="https://docs.scipy.org/doc/scipy/reference/sparse.html">documentation</a>
for more information on how to build (or convert your data to) sparse matrix
formats. Most of the time the <code class="docutils literal notranslate"><span class="pre">CSR</span></code> and <code class="docutils literal notranslate"><span class="pre">CSC</span></code> formats work best.</p>
</section>
<section id="influence-of-the-model-complexity">
<h3><span class="section-number">8.2.1.5. </span>Influence of the Model Complexity<a class="headerlink" href="#influence-of-the-model-complexity" title="Link to this heading">#</a></h3>
<p>Generally speaking, when model complexity increases, predictive power and
latency are supposed to increase. Increasing predictive power is usually
interesting, but for many applications we would better not increase
prediction latency too much. We will now review this idea for different
families of supervised models.</p>
<p>For <a class="reference internal" href="../api/sklearn.linear_model.html#module-sklearn.linear_model" title="sklearn.linear_model"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.linear_model</span></code></a> (e.g. Lasso, ElasticNet,
SGDClassifier/Regressor, Ridge & RidgeClassifier,
PassiveAggressiveClassifier/Regressor, LinearSVC, LogisticRegression…) the
decision function that is applied at prediction time is the same (a dot product)
, so latency should be equivalent.</p>
<p>Here is an example using
<a class="reference internal" href="../modules/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> with the
<code class="docutils literal notranslate"><span class="pre">elasticnet</span></code> penalty. The regularization strength is globally controlled by
the <code class="docutils literal notranslate"><span class="pre">alpha</span></code> parameter. With a sufficiently high <code class="docutils literal notranslate"><span class="pre">alpha</span></code>,
one can then increase the <code class="docutils literal notranslate"><span class="pre">l1_ratio</span></code> parameter of <code class="docutils literal notranslate"><span class="pre">elasticnet</span></code> to
enforce various levels of sparsity in the model coefficients. Higher sparsity
here is interpreted as less model complexity as we need fewer coefficients to
describe it fully. Of course sparsity influences in turn the prediction time
as the sparse dot-product takes time roughly proportional to the number of
non-zero coefficients.</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_model_complexity_influence.html"><img alt="en_model_complexity" src="../_images/sphx_glr_plot_model_complexity_influence_001.png" style="width: 512.0px; height: 384.0px;" /></a></strong></p><p>For the <a class="reference internal" href="../api/sklearn.svm.html#module-sklearn.svm" title="sklearn.svm"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.svm</span></code></a> family of algorithms with a non-linear kernel,
the latency is tied to the number of support vectors (the fewer the faster).
Latency and throughput should (asymptotically) grow linearly with the number
of support vectors in a SVC or SVR model. The kernel will also influence the
latency as it is used to compute the projection of the input vector once per
support vector. In the following graph the <code class="docutils literal notranslate"><span class="pre">nu</span></code> parameter of
<a class="reference internal" href="../modules/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> was used to influence the number of
support vectors.</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_model_complexity_influence.html"><img alt="nusvr_model_complexity" src="../_images/sphx_glr_plot_model_complexity_influence_002.png" style="width: 512.0px; height: 384.0px;" /></a></strong></p><p>For <a class="reference internal" href="../api/sklearn.ensemble.html#module-sklearn.ensemble" title="sklearn.ensemble"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.ensemble</span></code></a> of trees (e.g. RandomForest, GBT,
ExtraTrees, etc.) the number of trees and their depth play the most
important role. Latency and throughput should scale linearly with the number
of trees. In this case we used directly the <code class="docutils literal notranslate"><span class="pre">n_estimators</span></code> parameter of
<a class="reference internal" href="../modules/generated/sklearn.ensemble.GradientBoostingRegressor.html#sklearn.ensemble.GradientBoostingRegressor" title="sklearn.ensemble.GradientBoostingRegressor"><code class="xref py py-class docutils literal notranslate"><span class="pre">GradientBoostingRegressor</span></code></a>.</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_model_complexity_influence.html"><img alt="gbt_model_complexity" src="../_images/sphx_glr_plot_model_complexity_influence_003.png" style="width: 512.0px; height: 384.0px;" /></a></strong></p><p>In any case be warned that decreasing model complexity can hurt accuracy as
mentioned above. For instance a non-linearly separable problem can be handled
with a speedy linear model but prediction power will very likely suffer in
the process.</p>
</section>
<section id="feature-extraction-latency">
<h3><span class="section-number">8.2.1.6. </span>Feature Extraction Latency<a class="headerlink" href="#feature-extraction-latency" title="Link to this heading">#</a></h3>
<p>Most scikit-learn models are usually pretty fast as they are implemented
either with compiled Cython extensions or optimized computing libraries.
On the other hand, in many real world applications the feature extraction
process (i.e. turning raw data like database rows or network packets into
numpy arrays) governs the overall prediction time. For example on the Reuters
text classification task the whole preparation (reading and parsing SGML
files, tokenizing the text and hashing it into a common vector space) is
taking 100 to 500 times more time than the actual prediction code, depending on
the chosen model.</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_out_of_core_classification.html"><img alt="prediction_time" src="../_images/sphx_glr_plot_out_of_core_classification_004.png" style="width: 512.0px; height: 384.0px;" /></a></strong></p><p>In many cases it is thus recommended to carefully time and profile your
feature extraction code as it may be a good place to start optimizing when
your overall latency is too slow for your application.</p>
</section>
</section>
<section id="prediction-throughput">
<h2><span class="section-number">8.2.2. </span>Prediction Throughput<a class="headerlink" href="#prediction-throughput" title="Link to this heading">#</a></h2>
<p>Another important metric to care about when sizing production systems is the
throughput i.e. the number of predictions you can make in a given amount of
time. Here is a benchmark from the
<a class="reference internal" href="../auto_examples/applications/plot_prediction_latency.html#sphx-glr-auto-examples-applications-plot-prediction-latency-py"><span class="std std-ref">Prediction Latency</span></a> example that measures
this quantity for a number of estimators on synthetic data:</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/applications/plot_prediction_latency.html"><img alt="throughput_benchmark" src="../_images/sphx_glr_plot_prediction_latency_004.png" style="width: 800.0px; height: 480.0px;" /></a></strong></p><p>These throughputs are achieved on a single process. An obvious way to
increase the throughput of your application is to spawn additional instances
(usually processes in Python because of the
<a class="reference external" href="https://wiki.python.org/moin/GlobalInterpreterLock">GIL</a>) that share the
same model. One might also add machines to spread the load. A detailed
explanation on how to achieve this is beyond the scope of this documentation
though.</p>
</section>
<section id="tips-and-tricks">
<h2><span class="section-number">8.2.3. </span>Tips and Tricks<a class="headerlink" href="#tips-and-tricks" title="Link to this heading">#</a></h2>
<section id="linear-algebra-libraries">
<h3><span class="section-number">8.2.3.1. </span>Linear algebra libraries<a class="headerlink" href="#linear-algebra-libraries" title="Link to this heading">#</a></h3>
<p>As scikit-learn relies heavily on Numpy/Scipy and linear algebra in general it
makes sense to take explicit care of the versions of these libraries.
Basically, you ought to make sure that Numpy is built using an optimized <a class="reference external" href="https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms">BLAS</a> /
<a class="reference external" href="https://en.wikipedia.org/wiki/LAPACK">LAPACK</a> library.</p>
<p>Not all models benefit from optimized BLAS and Lapack implementations. For
instance models based on (randomized) decision trees typically do not rely on
BLAS calls in their inner loops, nor do kernel SVMs (<code class="docutils literal notranslate"><span class="pre">SVC</span></code>, <code class="docutils literal notranslate"><span class="pre">SVR</span></code>,
<code class="docutils literal notranslate"><span class="pre">NuSVC</span></code>, <code class="docutils literal notranslate"><span class="pre">NuSVR</span></code>). On the other hand a linear model implemented with a
BLAS DGEMM call (via <code class="docutils literal notranslate"><span class="pre">numpy.dot</span></code>) will typically benefit hugely from a tuned
BLAS implementation and lead to orders of magnitude speedup over a
non-optimized BLAS.</p>
<p>You can display the BLAS / LAPACK implementation used by your NumPy / SciPy /
scikit-learn install with the following command:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">python</span> <span class="o">-</span><span class="n">c</span> <span class="s2">"import sklearn; sklearn.show_versions()"</span>
</pre></div>
</div>
<p>Optimized BLAS / LAPACK implementations include:</p>
<ul class="simple">
<li><p>Atlas (need hardware specific tuning by rebuilding on the target machine)</p></li>
<li><p>OpenBLAS</p></li>
<li><p>MKL</p></li>
<li><p>Apple Accelerate and vecLib frameworks (OSX only)</p></li>
</ul>
<p>More information can be found on the <a class="reference external" href="https://numpy.org/install/">NumPy install page</a>
and in this
<a class="reference external" href="https://danielnouri.org/notes/2012/12/19/libblas-and-liblapack-issues-and-speed,-with-scipy-and-ubuntu/">blog post</a>
from Daniel Nouri which has some nice step by step install instructions for
Debian / Ubuntu.</p>
</section>
<section id="limiting-working-memory">
<span id="working-memory"></span><h3><span class="section-number">8.2.3.2. </span>Limiting Working Memory<a class="headerlink" href="#limiting-working-memory" title="Link to this heading">#</a></h3>
<p>Some calculations when implemented using standard numpy vectorized operations
involve using a large amount of temporary memory. This may potentially exhaust
system memory. Where computations can be performed in fixed-memory chunks, we
attempt to do so, and allow the user to hint at the maximum size of this
working memory (defaulting to 1GB) using <a class="reference internal" href="../modules/generated/sklearn.set_config.html#sklearn.set_config" title="sklearn.set_config"><code class="xref py py-func docutils literal notranslate"><span class="pre">set_config</span></code></a> or
<a class="reference internal" href="../modules/generated/sklearn.config_context.html#sklearn.config_context" title="sklearn.config_context"><code class="xref py py-func docutils literal notranslate"><span class="pre">config_context</span></code></a>. The following suggests to limit temporary working
memory to 128 MiB:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">sklearn</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="n">sklearn</span><span class="o">.</span><span class="n">config_context</span><span class="p">(</span><span class="n">working_memory</span><span class="o">=</span><span class="mi">128</span><span class="p">):</span>
<span class="gp">... </span> <span class="k">pass</span> <span class="c1"># do chunked work here</span>
</pre></div>
</div>
<p>An example of a chunked operation adhering to this setting is
<a class="reference internal" href="../modules/generated/sklearn.metrics.pairwise_distances_chunked.html#sklearn.metrics.pairwise_distances_chunked" title="sklearn.metrics.pairwise_distances_chunked"><code class="xref py py-func docutils literal notranslate"><span class="pre">pairwise_distances_chunked</span></code></a>, which facilitates computing
row-wise reductions of a pairwise distance matrix.</p>
</section>
<section id="model-compression">
<h3><span class="section-number">8.2.3.3. </span>Model Compression<a class="headerlink" href="#model-compression" title="Link to this heading">#</a></h3>
<p>Model compression in scikit-learn only concerns linear models for the moment.
In this context it means that we want to control the model sparsity (i.e. the
number of non-zero coordinates in the model vectors). It is generally a good
idea to combine model sparsity with sparse input data representation.</p>
<p>Here is sample code that illustrates the use of the <code class="docutils literal notranslate"><span class="pre">sparsify()</span></code> method:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">clf</span> <span class="o">=</span> <span class="n">SGDRegressor</span><span class="p">(</span><span class="n">penalty</span><span class="o">=</span><span class="s1">'elasticnet'</span><span class="p">,</span> <span class="n">l1_ratio</span><span class="o">=</span><span class="mf">0.25</span><span class="p">)</span>
<span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X_train</span><span class="p">,</span> <span class="n">y_train</span><span class="p">)</span><span class="o">.</span><span class="n">sparsify</span><span class="p">()</span>
<span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test</span><span class="p">)</span>
</pre></div>
</div>
<p>In this example we prefer the <code class="docutils literal notranslate"><span class="pre">elasticnet</span></code> penalty as it is often a good
compromise between model compactness and prediction power. One can also
further tune the <code class="docutils literal notranslate"><span class="pre">l1_ratio</span></code> parameter (in combination with the
regularization strength <code class="docutils literal notranslate"><span class="pre">alpha</span></code>) to control this tradeoff.</p>
<p>A typical <a class="reference external" href="https://github.com/scikit-learn/scikit-learn/blob/main/benchmarks/bench_sparsify.py">benchmark</a>
on synthetic data yields a >30% decrease in latency when both the model and
input are sparse (with 0.000024 and 0.027400 non-zero coefficients ratio
respectively). Your mileage may vary depending on the sparsity and size of
your data and model.
Furthermore, sparsifying can be very useful to reduce the memory usage of
predictive models deployed on production servers.</p>
</section>
<section id="model-reshaping">
<h3><span class="section-number">8.2.3.4. </span>Model Reshaping<a class="headerlink" href="#model-reshaping" title="Link to this heading">#</a></h3>
<p>Model reshaping consists in selecting only a portion of the available features
to fit a model. In other words, if a model discards features during the
learning phase we can then strip those from the input. This has several
benefits. Firstly it reduces memory (and therefore time) overhead of the
model itself. It also allows to discard explicit
feature selection components in a pipeline once we know which features to
keep from a previous run. Finally, it can help reduce processing time and I/O
usage upstream in the data access and feature extraction layers by not
collecting and building features that are discarded by the model. For instance
if the raw data come from a database, it can make it possible to write simpler
and faster queries or reduce I/O usage by making the queries return lighter
records.
At the moment, reshaping needs to be performed manually in scikit-learn.
In the case of sparse input (particularly in <code class="docutils literal notranslate"><span class="pre">CSR</span></code> format), it is generally
sufficient to not generate the relevant features, leaving their columns empty.</p>