-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathcross_validation.html
1188 lines (1143 loc) · 138 KB
/
cross_validation.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>
<!-- data-theme below is forced to be "light" but should be changed if we use pydata-theme-sphinx in the future -->
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" data-content_root="../" data-theme="light"> <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" data-content_root="../" data-theme="light"> <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1" />
<meta property="og:title" content="3.1. Cross-validation: evaluating estimator performance" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/modules/cross_validation.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="Learning the parameters of a prediction function and testing it on the same data is a methodological mistake: a model that would just repeat the labels of the samples that it has just seen would ha..." />
<meta property="og:image" content="https://scikit-learn/stable/_images/grid_search_workflow.png" />
<meta property="og:image:alt" content="Grid Search Workflow" />
<meta name="description" content="Learning the parameters of a prediction function and testing it on the same data is a methodological mistake: a model that would just repeat the labels of the samples that it has just seen would ha..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>3.1. Cross-validation: evaluating estimator performance — scikit-learn 1.5.dev0 documentation</title>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/cross_validation.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>
<script src="../_static/js/details-permalink.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/v1.5.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="../developers/index.html" >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/v1.5.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="../developers/index.html" >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="../model_selection.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="3. Model selection and evaluation">Prev</a><a href="../model_selection.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="3. Model selection and evaluation">Up</a>
<a href="grid_search.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="3.2. Tuning the hyper-parameters of an estimator">Next</a>
</div>
<div class="alert alert-danger p-1 mb-2" role="alert">
<p class="text-center mb-0">
<strong>scikit-learn 1.5.dev0</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="#">3.1. Cross-validation: evaluating estimator performance</a><ul>
<li><a class="reference internal" href="#computing-cross-validated-metrics">3.1.1. Computing cross-validated metrics</a><ul>
<li><a class="reference internal" href="#the-cross-validate-function-and-multiple-metric-evaluation">3.1.1.1. The cross_validate function and multiple metric evaluation</a></li>
<li><a class="reference internal" href="#obtaining-predictions-by-cross-validation">3.1.1.2. Obtaining predictions by cross-validation</a></li>
</ul>
</li>
<li><a class="reference internal" href="#cross-validation-iterators">3.1.2. Cross validation iterators</a><ul>
<li><a class="reference internal" href="#cross-validation-iterators-for-i-i-d-data">3.1.2.1. Cross-validation iterators for i.i.d. data</a><ul>
<li><a class="reference internal" href="#k-fold">3.1.2.1.1. K-fold</a></li>
<li><a class="reference internal" href="#repeated-k-fold">3.1.2.1.2. Repeated K-Fold</a></li>
<li><a class="reference internal" href="#leave-one-out-loo">3.1.2.1.3. Leave One Out (LOO)</a></li>
<li><a class="reference internal" href="#leave-p-out-lpo">3.1.2.1.4. Leave P Out (LPO)</a></li>
<li><a class="reference internal" href="#random-permutations-cross-validation-a-k-a-shuffle-split">3.1.2.1.5. Random permutations cross-validation a.k.a. Shuffle & Split</a></li>
</ul>
</li>
<li><a class="reference internal" href="#cross-validation-iterators-with-stratification-based-on-class-labels">3.1.2.2. Cross-validation iterators with stratification based on class labels</a><ul>
<li><a class="reference internal" href="#stratified-k-fold">3.1.2.2.1. Stratified k-fold</a></li>
<li><a class="reference internal" href="#stratified-shuffle-split">3.1.2.2.2. Stratified Shuffle Split</a></li>
</ul>
</li>
<li><a class="reference internal" href="#predefined-fold-splits-validation-sets">3.1.2.3. Predefined fold-splits / Validation-sets</a></li>
<li><a class="reference internal" href="#cross-validation-iterators-for-grouped-data">3.1.2.4. Cross-validation iterators for grouped data</a><ul>
<li><a class="reference internal" href="#group-k-fold">3.1.2.4.1. Group k-fold</a></li>
<li><a class="reference internal" href="#stratifiedgroupkfold">3.1.2.4.2. StratifiedGroupKFold</a></li>
<li><a class="reference internal" href="#leave-one-group-out">3.1.2.4.3. Leave One Group Out</a></li>
<li><a class="reference internal" href="#leave-p-groups-out">3.1.2.4.4. Leave P Groups Out</a></li>
<li><a class="reference internal" href="#group-shuffle-split">3.1.2.4.5. Group Shuffle Split</a></li>
</ul>
</li>
<li><a class="reference internal" href="#using-cross-validation-iterators-to-split-train-and-test">3.1.2.5. Using cross-validation iterators to split train and test</a></li>
<li><a class="reference internal" href="#cross-validation-of-time-series-data">3.1.2.6. Cross validation of time series data</a><ul>
<li><a class="reference internal" href="#time-series-split">3.1.2.6.1. Time Series Split</a></li>
</ul>
</li>
</ul>
</li>
<li><a class="reference internal" href="#a-note-on-shuffling">3.1.3. A note on shuffling</a></li>
<li><a class="reference internal" href="#cross-validation-and-model-selection">3.1.4. Cross validation and model selection</a></li>
<li><a class="reference internal" href="#permutation-test-score">3.1.5. Permutation test score</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="cross-validation-evaluating-estimator-performance">
<span id="cross-validation"></span><h1><span class="section-number">3.1. </span>Cross-validation: evaluating estimator performance<a class="headerlink" href="#cross-validation-evaluating-estimator-performance" title="Link to this heading">¶</a></h1>
<p>Learning the parameters of a prediction function and testing it on the
same data is a methodological mistake: a model that would just repeat
the labels of the samples that it has just seen would have a perfect
score but would fail to predict anything useful on yet-unseen data.
This situation is called <strong>overfitting</strong>.
To avoid it, it is common practice when performing
a (supervised) machine learning experiment
to hold out part of the available data as a <strong>test set</strong> <code class="docutils literal notranslate"><span class="pre">X_test,</span> <span class="pre">y_test</span></code>.
Note that the word “experiment” is not intended
to denote academic use only,
because even in commercial settings
machine learning usually starts out experimentally.
Here is a flowchart of typical cross validation workflow in model training.
The best parameters can be determined by
<a class="reference internal" href="grid_search.html#grid-search"><span class="std std-ref">grid search</span></a> techniques.</p>
<a class="reference internal image-reference" href="../_images/grid_search_workflow.png"><img alt="Grid Search Workflow" class="align-center" src="../_images/grid_search_workflow.png" style="width: 400px; height: 240px;" /></a>
<p>In scikit-learn a random split into training and test sets
can be quickly computed with the <a class="reference internal" href="generated/sklearn.model_selection.train_test_split.html#sklearn.model_selection.train_test_split" title="sklearn.model_selection.train_test_split"><code class="xref py py-func docutils literal notranslate"><span class="pre">train_test_split</span></code></a> helper function.
Let’s load the iris data set to fit a linear support vector machine on it:</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.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">datasets</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">datasets</span><span class="o">.</span><span class="n">load_iris</span><span class="p">(</span><span class="n">return_X_y</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X</span><span class="o">.</span><span class="n">shape</span><span class="p">,</span> <span class="n">y</span><span class="o">.</span><span class="n">shape</span>
<span class="go">((150, 4), (150,))</span>
</pre></div>
</div>
<p>We can now quickly sample a training set while holding out 40% of the
data for testing (evaluating) our classifier:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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="gp">... </span> <span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">test_size</span><span class="o">=</span><span class="mf">0.4</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="o">.</span><span class="n">shape</span><span class="p">,</span> <span class="n">y_train</span><span class="o">.</span><span class="n">shape</span>
<span class="go">((90, 4), (90,))</span>
<span class="gp">>>> </span><span class="n">X_test</span><span class="o">.</span><span class="n">shape</span><span class="p">,</span> <span class="n">y_test</span><span class="o">.</span><span class="n">shape</span>
<span class="go">((60, 4), (60,))</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">'linear'</span><span class="p">,</span> <span class="n">C</span><span class="o">=</span><span class="mi">1</span><span class="p">)</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="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">score</span><span class="p">(</span><span class="n">X_test</span><span class="p">,</span> <span class="n">y_test</span><span class="p">)</span>
<span class="go">0.96...</span>
</pre></div>
</div>
<p>When evaluating different settings (“hyperparameters”) for estimators,
such as the <code class="docutils literal notranslate"><span class="pre">C</span></code> setting that must be manually set for an SVM,
there is still a risk of overfitting <em>on the test set</em>
because the parameters can be tweaked until the estimator performs optimally.
This way, knowledge about the test set can “leak” into the model
and evaluation metrics no longer report on generalization performance.
To solve this problem, yet another part of the dataset can be held out
as a so-called “validation set”: training proceeds on the training set,
after which evaluation is done on the validation set,
and when the experiment seems to be successful,
final evaluation can be done on the test set.</p>
<p>However, by partitioning the available data into three sets,
we drastically reduce the number of samples
which can be used for learning the model,
and the results can depend on a particular random choice for the pair of
(train, validation) sets.</p>
<p>A solution to this problem is a procedure called
<a class="reference external" href="https://en.wikipedia.org/wiki/Cross-validation_(statistics)">cross-validation</a>
(CV for short).
A test set should still be held out for final evaluation,
but the validation set is no longer needed when doing CV.
In the basic approach, called <em>k</em>-fold CV,
the training set is split into <em>k</em> smaller sets
(other approaches are described below,
but generally follow the same principles).
The following procedure is followed for each of the <em>k</em> “folds”:</p>
<ul class="simple">
<li><p>A model is trained using <span class="math notranslate nohighlight">\(k-1\)</span> of the folds as training data;</p></li>
<li><p>the resulting model is validated on the remaining part of the data
(i.e., it is used as a test set to compute a performance measure
such as accuracy).</p></li>
</ul>
<p>The performance measure reported by <em>k</em>-fold cross-validation
is then the average of the values computed in the loop.
This approach can be computationally expensive,
but does not waste too much data
(as is the case when fixing an arbitrary validation set),
which is a major advantage in problems such as inverse inference
where the number of samples is very small.</p>
<a class="reference internal image-reference" href="../_images/grid_search_cross_validation.png"><img alt="A depiction of a 5 fold cross validation on a training set, while holding out a test set." class="align-center" src="../_images/grid_search_cross_validation.png" style="width: 500px; height: 300px;" /></a>
<section id="computing-cross-validated-metrics">
<h2><span class="section-number">3.1.1. </span>Computing cross-validated metrics<a class="headerlink" href="#computing-cross-validated-metrics" title="Link to this heading">¶</a></h2>
<p>The simplest way to use cross-validation is to call the
<a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a> helper function on the estimator and the dataset.</p>
<p>The following example demonstrates how to estimate the accuracy of a linear
kernel support vector machine on the iris dataset by splitting the data, fitting
a model and computing the score 5 consecutive times (with different splits each
time):</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.model_selection</span> <span class="kn">import</span> <span class="n">cross_val_score</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">'linear'</span><span class="p">,</span> <span class="n">C</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">42</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">scores</span> <span class="o">=</span> <span class="n">cross_val_score</span><span class="p">(</span><span class="n">clf</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">cv</span><span class="o">=</span><span class="mi">5</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">scores</span>
<span class="go">array([0.96..., 1. , 0.96..., 0.96..., 1. ])</span>
</pre></div>
</div>
<p>The mean score and the standard deviation are hence given by:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%0.2f</span><span class="s2"> accuracy with a standard deviation of </span><span class="si">%0.2f</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">scores</span><span class="o">.</span><span class="n">mean</span><span class="p">(),</span> <span class="n">scores</span><span class="o">.</span><span class="n">std</span><span class="p">()))</span>
<span class="go">0.98 accuracy with a standard deviation of 0.02</span>
</pre></div>
</div>
<p>By default, the score computed at each CV iteration is the <code class="docutils literal notranslate"><span class="pre">score</span></code>
method of the estimator. It is possible to change this by using the
scoring parameter:</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">metrics</span>
<span class="gp">>>> </span><span class="n">scores</span> <span class="o">=</span> <span class="n">cross_val_score</span><span class="p">(</span>
<span class="gp">... </span> <span class="n">clf</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">cv</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">scoring</span><span class="o">=</span><span class="s1">'f1_macro'</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">scores</span>
<span class="go">array([0.96..., 1. ..., 0.96..., 0.96..., 1. ])</span>
</pre></div>
</div>
<p>See <a class="reference internal" href="model_evaluation.html#scoring-parameter"><span class="std std-ref">The scoring parameter: defining model evaluation rules</span></a> for details.
In the case of the Iris dataset, the samples are balanced across target
classes hence the accuracy and the F1-score are almost equal.</p>
<p>When the <code class="docutils literal notranslate"><span class="pre">cv</span></code> argument is an integer, <a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a> uses the
<a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> or <a class="reference internal" href="generated/sklearn.model_selection.StratifiedKFold.html#sklearn.model_selection.StratifiedKFold" title="sklearn.model_selection.StratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedKFold</span></code></a> strategies by default, the latter
being used if the estimator derives from <a class="reference internal" href="generated/sklearn.base.ClassifierMixin.html#sklearn.base.ClassifierMixin" title="sklearn.base.ClassifierMixin"><code class="xref py py-class docutils literal notranslate"><span class="pre">ClassifierMixin</span></code></a>.</p>
<p>It is also possible to use other cross validation strategies by passing a cross
validation iterator instead, for instance:</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.model_selection</span> <span class="kn">import</span> <span class="n">ShuffleSplit</span>
<span class="gp">>>> </span><span class="n">n_samples</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">0</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">cv</span> <span class="o">=</span> <span class="n">ShuffleSplit</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">test_size</span><span class="o">=</span><span class="mf">0.3</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">cross_val_score</span><span class="p">(</span><span class="n">clf</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">cv</span><span class="o">=</span><span class="n">cv</span><span class="p">)</span>
<span class="go">array([0.977..., 0.977..., 1. ..., 0.955..., 1. ])</span>
</pre></div>
</div>
<p>Another option is to use an iterable yielding (train, test) splits as arrays of
indices, for example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="k">def</span> <span class="nf">custom_cv_2folds</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="n">n</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">0</span><span class="p">]</span>
<span class="gp">... </span> <span class="n">i</span> <span class="o">=</span> <span class="mi">1</span>
<span class="gp">... </span> <span class="k">while</span> <span class="n">i</span> <span class="o"><=</span> <span class="mi">2</span><span class="p">:</span>
<span class="gp">... </span> <span class="n">idx</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="n">n</span> <span class="o">*</span> <span class="p">(</span><span class="n">i</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span> <span class="o">/</span> <span class="mi">2</span><span class="p">,</span> <span class="n">n</span> <span class="o">*</span> <span class="n">i</span> <span class="o">/</span> <span class="mi">2</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="nb">int</span><span class="p">)</span>
<span class="gp">... </span> <span class="k">yield</span> <span class="n">idx</span><span class="p">,</span> <span class="n">idx</span>
<span class="gp">... </span> <span class="n">i</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="gp">...</span>
<span class="gp">>>> </span><span class="n">custom_cv</span> <span class="o">=</span> <span class="n">custom_cv_2folds</span><span class="p">(</span><span class="n">X</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">cross_val_score</span><span class="p">(</span><span class="n">clf</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">cv</span><span class="o">=</span><span class="n">custom_cv</span><span class="p">)</span>
<span class="go">array([1. , 0.973...])</span>
</pre></div>
</div>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Data transformation with held out data</strong>
<span class="tooltiptext">Click for more details</span>
<a class="headerlink" href="#summary-anchor" title="Permalink to this heading">¶</a>
</summary>
<div class="card"></p>
<blockquote>
<div><p>Just as it is important to test a predictor on data held-out from
training, preprocessing (such as standardization, feature selection, etc.)
and similar <a class="reference internal" href="../data_transforms.html#data-transforms"><span class="std std-ref">data transformations</span></a> similarly should
be learnt from a training set and applied to held-out data for prediction:</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">preprocessing</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="gp">... </span> <span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">test_size</span><span class="o">=</span><span class="mf">0.4</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">scaler</span> <span class="o">=</span> <span class="n">preprocessing</span><span class="o">.</span><span class="n">StandardScaler</span><span class="p">()</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="gp">>>> </span><span class="n">X_train_transformed</span> <span class="o">=</span> <span class="n">scaler</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">X_train</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">C</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X_train_transformed</span><span class="p">,</span> <span class="n">y_train</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X_test_transformed</span> <span class="o">=</span> <span class="n">scaler</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">X_test</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">score</span><span class="p">(</span><span class="n">X_test_transformed</span><span class="p">,</span> <span class="n">y_test</span><span class="p">)</span>
<span class="go">0.9333...</span>
</pre></div>
</div>
<p>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> makes it easier to compose
estimators, providing this behavior under cross-validation:</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="n">clf</span> <span class="o">=</span> <span class="n">make_pipeline</span><span class="p">(</span><span class="n">preprocessing</span><span class="o">.</span><span class="n">StandardScaler</span><span class="p">(),</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">(</span><span class="n">C</span><span class="o">=</span><span class="mi">1</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">cross_val_score</span><span class="p">(</span><span class="n">clf</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">cv</span><span class="o">=</span><span class="n">cv</span><span class="p">)</span>
<span class="go">array([0.977..., 0.933..., 0.955..., 0.933..., 0.977...])</span>
</pre></div>
</div>
<p>See <a class="reference internal" href="compose.html#combining-estimators"><span class="std std-ref">Pipelines and composite estimators</span></a>.</p>
</div></blockquote>
<p></div>
</details></p>
<section id="the-cross-validate-function-and-multiple-metric-evaluation">
<span id="multimetric-cross-validation"></span><h3><span class="section-number">3.1.1.1. </span>The cross_validate function and multiple metric evaluation<a class="headerlink" href="#the-cross-validate-function-and-multiple-metric-evaluation" title="Link to this heading">¶</a></h3>
<p>The <a class="reference internal" href="generated/sklearn.model_selection.cross_validate.html#sklearn.model_selection.cross_validate" title="sklearn.model_selection.cross_validate"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_validate</span></code></a> function differs from <a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a> in
two ways:</p>
<ul class="simple">
<li><p>It allows specifying multiple metrics for evaluation.</p></li>
<li><p>It returns a dict containing fit-times, score-times
(and optionally training scores, fitted estimators, train-test split indices)
in addition to the test score.</p></li>
</ul>
<p>For single metric evaluation, where the scoring parameter is a string,
callable or None, the keys will be - <code class="docutils literal notranslate"><span class="pre">['test_score',</span> <span class="pre">'fit_time',</span> <span class="pre">'score_time']</span></code></p>
<p>And for multiple metric evaluation, the return value is a dict with the
following keys -
<code class="docutils literal notranslate"><span class="pre">['test_<scorer1_name>',</span> <span class="pre">'test_<scorer2_name>',</span> <span class="pre">'test_<scorer...>',</span> <span class="pre">'fit_time',</span> <span class="pre">'score_time']</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">return_train_score</span></code> is set to <code class="docutils literal notranslate"><span class="pre">False</span></code> by default to save computation time.
To evaluate the scores on the training set as well you need to set it to
<code class="docutils literal notranslate"><span class="pre">True</span></code>. You may also retain the estimator fitted on each training set by
setting <code class="docutils literal notranslate"><span class="pre">return_estimator=True</span></code>. Similarly, you may set
<code class="docutils literal notranslate"><span class="pre">return_indices=True</span></code> to retain the training and testing indices used to split
the dataset into train and test sets for each cv split.</p>
<p>The multiple metrics can be specified either as a list, tuple or set of
predefined scorer names:</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.model_selection</span> <span class="kn">import</span> <span class="n">cross_validate</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.metrics</span> <span class="kn">import</span> <span class="n">recall_score</span>
<span class="gp">>>> </span><span class="n">scoring</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'precision_macro'</span><span class="p">,</span> <span class="s1">'recall_macro'</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">'linear'</span><span class="p">,</span> <span class="n">C</span><span class="o">=</span><span class="mi">1</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">scores</span> <span class="o">=</span> <span class="n">cross_validate</span><span class="p">(</span><span class="n">clf</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">scoring</span><span class="o">=</span><span class="n">scoring</span><span class="p">)</span>
<span class="gp">>>> </span><span class="nb">sorted</span><span class="p">(</span><span class="n">scores</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
<span class="go">['fit_time', 'score_time', 'test_precision_macro', 'test_recall_macro']</span>
<span class="gp">>>> </span><span class="n">scores</span><span class="p">[</span><span class="s1">'test_recall_macro'</span><span class="p">]</span>
<span class="go">array([0.96..., 1. ..., 0.96..., 0.96..., 1. ])</span>
</pre></div>
</div>
<p>Or as a dict mapping scorer name to a predefined or custom scoring function:</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.metrics</span> <span class="kn">import</span> <span class="n">make_scorer</span>
<span class="gp">>>> </span><span class="n">scoring</span> <span class="o">=</span> <span class="p">{</span><span class="s1">'prec_macro'</span><span class="p">:</span> <span class="s1">'precision_macro'</span><span class="p">,</span>
<span class="gp">... </span> <span class="s1">'rec_macro'</span><span class="p">:</span> <span class="n">make_scorer</span><span class="p">(</span><span class="n">recall_score</span><span class="p">,</span> <span class="n">average</span><span class="o">=</span><span class="s1">'macro'</span><span class="p">)}</span>
<span class="gp">>>> </span><span class="n">scores</span> <span class="o">=</span> <span class="n">cross_validate</span><span class="p">(</span><span class="n">clf</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">scoring</span><span class="o">=</span><span class="n">scoring</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">cv</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">return_train_score</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="nb">sorted</span><span class="p">(</span><span class="n">scores</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
<span class="go">['fit_time', 'score_time', 'test_prec_macro', 'test_rec_macro',</span>
<span class="go"> 'train_prec_macro', 'train_rec_macro']</span>
<span class="gp">>>> </span><span class="n">scores</span><span class="p">[</span><span class="s1">'train_rec_macro'</span><span class="p">]</span>
<span class="go">array([0.97..., 0.97..., 0.99..., 0.98..., 0.98...])</span>
</pre></div>
</div>
<p>Here is an example of <code class="docutils literal notranslate"><span class="pre">cross_validate</span></code> using a single metric:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">scores</span> <span class="o">=</span> <span class="n">cross_validate</span><span class="p">(</span><span class="n">clf</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="n">scoring</span><span class="o">=</span><span class="s1">'precision_macro'</span><span class="p">,</span> <span class="n">cv</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">return_estimator</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="nb">sorted</span><span class="p">(</span><span class="n">scores</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
<span class="go">['estimator', 'fit_time', 'score_time', 'test_score']</span>
</pre></div>
</div>
</section>
<section id="obtaining-predictions-by-cross-validation">
<h3><span class="section-number">3.1.1.2. </span>Obtaining predictions by cross-validation<a class="headerlink" href="#obtaining-predictions-by-cross-validation" title="Link to this heading">¶</a></h3>
<p>The function <a class="reference internal" href="generated/sklearn.model_selection.cross_val_predict.html#sklearn.model_selection.cross_val_predict" title="sklearn.model_selection.cross_val_predict"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_predict</span></code></a> has a similar interface to
<a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a>, but returns, for each element in the input, the
prediction that was obtained for that element when it was in the test set. Only
cross-validation strategies that assign all elements to a test set exactly once
can be used (otherwise, an exception is raised).</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Note on inappropriate usage of cross_val_predict</p>
<p>The result of <a class="reference internal" href="generated/sklearn.model_selection.cross_val_predict.html#sklearn.model_selection.cross_val_predict" title="sklearn.model_selection.cross_val_predict"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_predict</span></code></a> may be different from those
obtained using <a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a> as the elements are grouped in
different ways. The function <a class="reference internal" href="generated/sklearn.model_selection.cross_val_score.html#sklearn.model_selection.cross_val_score" title="sklearn.model_selection.cross_val_score"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_score</span></code></a> takes an average
over cross-validation folds, whereas <a class="reference internal" href="generated/sklearn.model_selection.cross_val_predict.html#sklearn.model_selection.cross_val_predict" title="sklearn.model_selection.cross_val_predict"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_predict</span></code></a> simply
returns the labels (or probabilities) from several distinct models
undistinguished. Thus, <a class="reference internal" href="generated/sklearn.model_selection.cross_val_predict.html#sklearn.model_selection.cross_val_predict" title="sklearn.model_selection.cross_val_predict"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_predict</span></code></a> is not an appropriate
measure of generalization error.</p>
</div>
<dl class="simple">
<dt>The function <a class="reference internal" href="generated/sklearn.model_selection.cross_val_predict.html#sklearn.model_selection.cross_val_predict" title="sklearn.model_selection.cross_val_predict"><code class="xref py py-func docutils literal notranslate"><span class="pre">cross_val_predict</span></code></a> is appropriate for:</dt><dd><ul class="simple">
<li><p>Visualization of predictions obtained from different models.</p></li>
<li><p>Model blending: When predictions of one supervised estimator are used to
train another estimator in ensemble methods.</p></li>
</ul>
</dd>
</dl>
<p>The available cross validation iterators are introduced in the following
section.</p>
<aside class="topic">
<p class="topic-title">Examples</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/model_selection/plot_roc_crossval.html#sphx-glr-auto-examples-model-selection-plot-roc-crossval-py"><span class="std std-ref">Receiver Operating Characteristic (ROC) with cross validation</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/feature_selection/plot_rfe_with_cross_validation.html#sphx-glr-auto-examples-feature-selection-plot-rfe-with-cross-validation-py"><span class="std std-ref">Recursive feature elimination with cross-validation</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/model_selection/plot_grid_search_digits.html#sphx-glr-auto-examples-model-selection-plot-grid-search-digits-py"><span class="std std-ref">Custom refit strategy of a grid search with cross-validation</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/model_selection/plot_grid_search_text_feature_extraction.html#sphx-glr-auto-examples-model-selection-plot-grid-search-text-feature-extraction-py"><span class="std std-ref">Sample pipeline for text feature extraction and evaluation</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/model_selection/plot_cv_predict.html#sphx-glr-auto-examples-model-selection-plot-cv-predict-py"><span class="std std-ref">Plotting Cross-Validated Predictions</span></a>,</p></li>
<li><p><a class="reference internal" href="../auto_examples/model_selection/plot_nested_cross_validation_iris.html#sphx-glr-auto-examples-model-selection-plot-nested-cross-validation-iris-py"><span class="std std-ref">Nested versus non-nested cross-validation</span></a>.</p></li>
</ul>
</aside>
</section>
</section>
<section id="cross-validation-iterators">
<h2><span class="section-number">3.1.2. </span>Cross validation iterators<a class="headerlink" href="#cross-validation-iterators" title="Link to this heading">¶</a></h2>
<p>The following sections list utilities to generate indices
that can be used to generate dataset splits according to different cross
validation strategies.</p>
<section id="cross-validation-iterators-for-i-i-d-data">
<span id="iid-cv"></span><h3><span class="section-number">3.1.2.1. </span>Cross-validation iterators for i.i.d. data<a class="headerlink" href="#cross-validation-iterators-for-i-i-d-data" title="Link to this heading">¶</a></h3>
<p>Assuming that some data is Independent and Identically Distributed (i.i.d.) is
making the assumption that all samples stem from the same generative process
and that the generative process is assumed to have no memory of past generated
samples.</p>
<p>The following cross-validators can be used in such cases.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>While i.i.d. data is a common assumption in machine learning theory, it rarely
holds in practice. If one knows that the samples have been generated using a
time-dependent process, it is safer to
use a <a class="reference internal" href="#timeseries-cv"><span class="std std-ref">time-series aware cross-validation scheme</span></a>.
Similarly, if we know that the generative process has a group structure
(samples collected from different subjects, experiments, measurement
devices), it is safer to use <a class="reference internal" href="#group-cv"><span class="std std-ref">group-wise cross-validation</span></a>.</p>
</div>
<section id="k-fold">
<span id="id2"></span><h4><span class="section-number">3.1.2.1.1. </span>K-fold<a class="headerlink" href="#k-fold" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> divides all the samples in <span class="math notranslate nohighlight">\(k\)</span> groups of samples,
called folds (if <span class="math notranslate nohighlight">\(k = n\)</span>, this is equivalent to the <em>Leave One
Out</em> strategy), of equal sizes (if possible). The prediction function is
learned using <span class="math notranslate nohighlight">\(k - 1\)</span> folds, and the fold left out is used for test.</p>
<p>Example of 2-fold cross-validation on a dataset with 4 samples:</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.model_selection</span> <span class="kn">import</span> <span class="n">KFold</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"a"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"d"</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">kf</span> <span class="o">=</span> <span class="n">KFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">kf</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[2 3] [0 1]</span>
<span class="go">[0 1] [2 3]</span>
</pre></div>
</div>
<p>Here is a visualization of the cross-validation behavior. Note that
<a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> is not affected by classes or groups.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_006.png" src="../_images/sphx_glr_plot_cv_indices_006.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
<p>Each fold is constituted by two arrays: the first one is related to the
<em>training set</em>, and the second one to the <em>test set</em>.
Thus, one can create the training/test sets using numpy indexing:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([[</span><span class="mf">0.</span><span class="p">,</span> <span class="mf">0.</span><span class="p">],</span> <span class="p">[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">],</span> <span class="p">[</span><span class="o">-</span><span class="mf">1.</span><span class="p">,</span> <span class="o">-</span><span class="mf">1.</span><span class="p">],</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="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</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">0</span><span class="p">,</span> <span class="mi">1</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">X</span><span class="p">[</span><span class="n">train</span><span class="p">],</span> <span class="n">X</span><span class="p">[</span><span class="n">test</span><span class="p">],</span> <span class="n">y</span><span class="p">[</span><span class="n">train</span><span class="p">],</span> <span class="n">y</span><span class="p">[</span><span class="n">test</span><span class="p">]</span>
</pre></div>
</div>
</section>
<section id="repeated-k-fold">
<span id="id3"></span><h4><span class="section-number">3.1.2.1.2. </span>Repeated K-Fold<a class="headerlink" href="#repeated-k-fold" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.RepeatedKFold.html#sklearn.model_selection.RepeatedKFold" title="sklearn.model_selection.RepeatedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">RepeatedKFold</span></code></a> repeats K-Fold n times. It can be used when one
requires to run <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> n times, producing different splits in
each repetition.</p>
<p>Example of 2-fold K-Fold repeated 2 times:</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.model_selection</span> <span class="kn">import</span> <span class="n">RepeatedKFold</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</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="p">[</span><span class="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">],</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="p">[</span><span class="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">]])</span>
<span class="gp">>>> </span><span class="n">random_state</span> <span class="o">=</span> <span class="mi">12883823</span>
<span class="gp">>>> </span><span class="n">rkf</span> <span class="o">=</span> <span class="n">RepeatedKFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">n_repeats</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="n">random_state</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">rkf</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="gp">...</span>
<span class="go">[2 3] [0 1]</span>
<span class="go">[0 1] [2 3]</span>
<span class="go">[0 2] [1 3]</span>
<span class="go">[1 3] [0 2]</span>
</pre></div>
</div>
<p>Similarly, <a class="reference internal" href="generated/sklearn.model_selection.RepeatedStratifiedKFold.html#sklearn.model_selection.RepeatedStratifiedKFold" title="sklearn.model_selection.RepeatedStratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">RepeatedStratifiedKFold</span></code></a> repeats Stratified K-Fold n times
with different randomization in each repetition.</p>
</section>
<section id="leave-one-out-loo">
<span id="leave-one-out"></span><h4><span class="section-number">3.1.2.1.3. </span>Leave One Out (LOO)<a class="headerlink" href="#leave-one-out-loo" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.LeaveOneOut.html#sklearn.model_selection.LeaveOneOut" title="sklearn.model_selection.LeaveOneOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneOut</span></code></a> (or LOO) is a simple cross-validation. Each learning
set is created by taking all the samples except one, the test set being
the sample left out. Thus, for <span class="math notranslate nohighlight">\(n\)</span> samples, we have <span class="math notranslate nohighlight">\(n\)</span> different
training sets and <span class="math notranslate nohighlight">\(n\)</span> different tests set. This cross-validation
procedure does not waste much data as only one sample is removed from the
training set:</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.model_selection</span> <span class="kn">import</span> <span class="n">LeaveOneOut</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</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="mi">4</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">loo</span> <span class="o">=</span> <span class="n">LeaveOneOut</span><span class="p">()</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">loo</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[1 2 3] [0]</span>
<span class="go">[0 2 3] [1]</span>
<span class="go">[0 1 3] [2]</span>
<span class="go">[0 1 2] [3]</span>
</pre></div>
</div>
<p>Potential users of LOO for model selection should weigh a few known caveats.
When compared with <span class="math notranslate nohighlight">\(k\)</span>-fold cross validation, one builds <span class="math notranslate nohighlight">\(n\)</span> models
from <span class="math notranslate nohighlight">\(n\)</span> samples instead of <span class="math notranslate nohighlight">\(k\)</span> models, where <span class="math notranslate nohighlight">\(n > k\)</span>.
Moreover, each is trained on <span class="math notranslate nohighlight">\(n - 1\)</span> samples rather than
<span class="math notranslate nohighlight">\((k-1) n / k\)</span>. In both ways, assuming <span class="math notranslate nohighlight">\(k\)</span> is not too large
and <span class="math notranslate nohighlight">\(k < n\)</span>, LOO is more computationally expensive than <span class="math notranslate nohighlight">\(k\)</span>-fold
cross validation.</p>
<p>In terms of accuracy, LOO often results in high variance as an estimator for the
test error. Intuitively, since <span class="math notranslate nohighlight">\(n - 1\)</span> of
the <span class="math notranslate nohighlight">\(n\)</span> samples are used to build each model, models constructed from
folds are virtually identical to each other and to the model built from the
entire training set.</p>
<p>However, if the learning curve is steep for the training size in question,
then 5- or 10- fold cross validation can overestimate the generalization error.</p>
<p>As a general rule, most authors, and empirical evidence, suggest that 5- or 10-
fold cross validation should be preferred to LOO.</p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>References</strong>
<span class="tooltiptext">Click for more details</span>
<a class="headerlink" href="#summary-anchor" title="Permalink to this heading">¶</a>
</summary>
<div class="card"></p>
<blockquote>
<div><ul class="simple">
<li><p><a class="reference external" href="http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-12.html">http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-12.html</a>;</p></li>
<li><p>T. Hastie, R. Tibshirani, J. Friedman, <a class="reference external" href="https://web.stanford.edu/~hastie/ElemStatLearn/">The Elements of Statistical Learning</a>, Springer 2009</p></li>
<li><p>L. Breiman, P. Spector <a class="reference external" href="https://digitalassets.lib.berkeley.edu/sdtr/ucb/text/197.pdf">Submodel selection and evaluation in regression: The X-random case</a>, International Statistical Review 1992;</p></li>
<li><p>R. Kohavi, <a class="reference external" href="https://www.ijcai.org/Proceedings/95-2/Papers/016.pdf">A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection</a>, Intl. Jnt. Conf. AI</p></li>
<li><p>R. Bharat Rao, G. Fung, R. Rosales, <a class="reference external" href="https://people.csail.mit.edu/romer/papers/CrossVal_SDM08.pdf">On the Dangers of Cross-Validation. An Experimental Evaluation</a>, SIAM 2008;</p></li>
<li><p>G. James, D. Witten, T. Hastie, R Tibshirani, <a class="reference external" href="https://www.statlearning.com">An Introduction to
Statistical Learning</a>, Springer 2013.</p></li>
</ul>
</div></blockquote>
<p></div>
</details></p>
</section>
<section id="leave-p-out-lpo">
<span id="leave-p-out"></span><h4><span class="section-number">3.1.2.1.4. </span>Leave P Out (LPO)<a class="headerlink" href="#leave-p-out-lpo" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.LeavePOut.html#sklearn.model_selection.LeavePOut" title="sklearn.model_selection.LeavePOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePOut</span></code></a> is very similar to <a class="reference internal" href="generated/sklearn.model_selection.LeaveOneOut.html#sklearn.model_selection.LeaveOneOut" title="sklearn.model_selection.LeaveOneOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneOut</span></code></a> as it creates all
the possible training/test sets by removing <span class="math notranslate nohighlight">\(p\)</span> samples from the complete
set. For <span class="math notranslate nohighlight">\(n\)</span> samples, this produces <span class="math notranslate nohighlight">\({n \choose p}\)</span> train-test
pairs. Unlike <a class="reference internal" href="generated/sklearn.model_selection.LeaveOneOut.html#sklearn.model_selection.LeaveOneOut" title="sklearn.model_selection.LeaveOneOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneOut</span></code></a> and <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a>, the test sets will
overlap for <span class="math notranslate nohighlight">\(p > 1\)</span>.</p>
<p>Example of Leave-2-Out on a dataset with 4 samples:</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.model_selection</span> <span class="kn">import</span> <span class="n">LeavePOut</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">4</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">lpo</span> <span class="o">=</span> <span class="n">LeavePOut</span><span class="p">(</span><span class="n">p</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">lpo</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[2 3] [0 1]</span>
<span class="go">[1 3] [0 2]</span>
<span class="go">[1 2] [0 3]</span>
<span class="go">[0 3] [1 2]</span>
<span class="go">[0 2] [1 3]</span>
<span class="go">[0 1] [2 3]</span>
</pre></div>
</div>
</section>
<section id="random-permutations-cross-validation-a-k-a-shuffle-split">
<span id="shufflesplit"></span><h4><span class="section-number">3.1.2.1.5. </span>Random permutations cross-validation a.k.a. Shuffle & Split<a class="headerlink" href="#random-permutations-cross-validation-a-k-a-shuffle-split" title="Link to this heading">¶</a></h4>
<p>The <a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a> iterator will generate a user defined number of
independent train / test dataset splits. Samples are first shuffled and
then split into a pair of train and test sets.</p>
<p>It is possible to control the randomness for reproducibility of the
results by explicitly seeding the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> pseudo random number
generator.</p>
<p>Here is a usage example:</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.model_selection</span> <span class="kn">import</span> <span class="n">ShuffleSplit</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">ss</span> <span class="o">=</span> <span class="n">ShuffleSplit</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">test_size</span><span class="o">=</span><span class="mf">0.25</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="k">for</span> <span class="n">train_index</span><span class="p">,</span> <span class="n">test_index</span> <span class="ow">in</span> <span class="n">ss</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="n">X</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train_index</span><span class="p">,</span> <span class="n">test_index</span><span class="p">))</span>
<span class="go">[9 1 6 7 3 0 5] [2 8 4]</span>
<span class="go">[2 9 8 0 6 7 4] [3 5 1]</span>
<span class="go">[4 5 1 0 6 9 7] [2 3 8]</span>
<span class="go">[2 7 5 8 0 3 4] [6 1 9]</span>
<span class="go">[4 1 0 6 8 9 3] [5 2 7]</span>
</pre></div>
</div>
<p>Here is a visualization of the cross-validation behavior. Note that
<a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a> is not affected by classes or groups.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_008.png" src="../_images/sphx_glr_plot_cv_indices_008.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
<p><a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a> is thus a good alternative to <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> cross
validation that allows a finer control on the number of iterations and
the proportion of samples on each side of the train / test split.</p>
</section>
</section>
<section id="cross-validation-iterators-with-stratification-based-on-class-labels">
<span id="stratification"></span><h3><span class="section-number">3.1.2.2. </span>Cross-validation iterators with stratification based on class labels<a class="headerlink" href="#cross-validation-iterators-with-stratification-based-on-class-labels" title="Link to this heading">¶</a></h3>
<p>Some classification problems can exhibit a large imbalance in the distribution
of the target classes: for instance there could be several times more negative
samples than positive samples. In such cases it is recommended to use
stratified sampling as implemented in <a class="reference internal" href="generated/sklearn.model_selection.StratifiedKFold.html#sklearn.model_selection.StratifiedKFold" title="sklearn.model_selection.StratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedKFold</span></code></a> and
<a class="reference internal" href="generated/sklearn.model_selection.StratifiedShuffleSplit.html#sklearn.model_selection.StratifiedShuffleSplit" title="sklearn.model_selection.StratifiedShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedShuffleSplit</span></code></a> to ensure that relative class frequencies is
approximately preserved in each train and validation fold.</p>
<section id="stratified-k-fold">
<span id="id4"></span><h4><span class="section-number">3.1.2.2.1. </span>Stratified k-fold<a class="headerlink" href="#stratified-k-fold" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.StratifiedKFold.html#sklearn.model_selection.StratifiedKFold" title="sklearn.model_selection.StratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedKFold</span></code></a> is a variation of <em>k-fold</em> which returns <em>stratified</em>
folds: each set contains approximately the same percentage of samples of each
target class as the complete set.</p>
<p>Here is an example of stratified 3-fold cross-validation on a dataset with 50 samples from
two unbalanced classes. We show the number of samples in each class and compare with
<a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a>.</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></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">StratifiedKFold</span><span class="p">,</span> <span class="n">KFold</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="n">X</span><span class="p">,</span> <span class="n">y</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">ones</span><span class="p">((</span><span class="mi">50</span><span class="p">,</span> <span class="mi">1</span><span class="p">)),</span> <span class="n">np</span><span class="o">.</span><span class="n">hstack</span><span class="p">(([</span><span class="mi">0</span><span class="p">]</span> <span class="o">*</span> <span class="mi">45</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">*</span> <span class="mi">5</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">skf</span> <span class="o">=</span> <span class="n">StratifiedKFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">skf</span><span class="o">.</span><span class="n">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="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s1">'train - </span><span class="si">{}</span><span class="s1"> | test - </span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span>
<span class="gp">... </span> <span class="n">np</span><span class="o">.</span><span class="n">bincount</span><span class="p">(</span><span class="n">y</span><span class="p">[</span><span class="n">train</span><span class="p">]),</span> <span class="n">np</span><span class="o">.</span><span class="n">bincount</span><span class="p">(</span><span class="n">y</span><span class="p">[</span><span class="n">test</span><span class="p">])))</span>
<span class="go">train - [30 3] | test - [15 2]</span>
<span class="go">train - [30 3] | test - [15 2]</span>
<span class="go">train - [30 4] | test - [15 1]</span>
<span class="gp">>>> </span><span class="n">kf</span> <span class="o">=</span> <span class="n">KFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">kf</span><span class="o">.</span><span class="n">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="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s1">'train - </span><span class="si">{}</span><span class="s1"> | test - </span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span>
<span class="gp">... </span> <span class="n">np</span><span class="o">.</span><span class="n">bincount</span><span class="p">(</span><span class="n">y</span><span class="p">[</span><span class="n">train</span><span class="p">]),</span> <span class="n">np</span><span class="o">.</span><span class="n">bincount</span><span class="p">(</span><span class="n">y</span><span class="p">[</span><span class="n">test</span><span class="p">])))</span>
<span class="go">train - [28 5] | test - [17]</span>
<span class="go">train - [28 5] | test - [17]</span>
<span class="go">train - [34] | test - [11 5]</span>
</pre></div>
</div>
<p>We can see that <a class="reference internal" href="generated/sklearn.model_selection.StratifiedKFold.html#sklearn.model_selection.StratifiedKFold" title="sklearn.model_selection.StratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedKFold</span></code></a> preserves the class ratios
(approximately 1 / 10) in both train and test dataset.</p>
<p>Here is a visualization of the cross-validation behavior.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_009.png" src="../_images/sphx_glr_plot_cv_indices_009.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
<p><a class="reference internal" href="generated/sklearn.model_selection.RepeatedStratifiedKFold.html#sklearn.model_selection.RepeatedStratifiedKFold" title="sklearn.model_selection.RepeatedStratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">RepeatedStratifiedKFold</span></code></a> can be used to repeat Stratified K-Fold n times
with different randomization in each repetition.</p>
</section>
<section id="stratified-shuffle-split">
<span id="id5"></span><h4><span class="section-number">3.1.2.2.2. </span>Stratified Shuffle Split<a class="headerlink" href="#stratified-shuffle-split" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.StratifiedShuffleSplit.html#sklearn.model_selection.StratifiedShuffleSplit" title="sklearn.model_selection.StratifiedShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedShuffleSplit</span></code></a> is a variation of <em>ShuffleSplit</em>, which returns
stratified splits, <em>i.e</em> which creates splits by preserving the same
percentage for each target class as in the complete set.</p>
<p>Here is a visualization of the cross-validation behavior.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_012.png" src="../_images/sphx_glr_plot_cv_indices_012.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
</section>
</section>
<section id="predefined-fold-splits-validation-sets">
<span id="predefined-split"></span><h3><span class="section-number">3.1.2.3. </span>Predefined fold-splits / Validation-sets<a class="headerlink" href="#predefined-fold-splits-validation-sets" title="Link to this heading">¶</a></h3>
<p>For some datasets, a pre-defined split of the data into training- and
validation fold or into several cross-validation folds already
exists. Using <a class="reference internal" href="generated/sklearn.model_selection.PredefinedSplit.html#sklearn.model_selection.PredefinedSplit" title="sklearn.model_selection.PredefinedSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">PredefinedSplit</span></code></a> it is possible to use these folds
e.g. when searching for hyperparameters.</p>
<p>For example, when using a validation set, set the <code class="docutils literal notranslate"><span class="pre">test_fold</span></code> to 0 for all
samples that are part of the validation set, and to -1 for all other samples.</p>
</section>
<section id="cross-validation-iterators-for-grouped-data">
<span id="group-cv"></span><h3><span class="section-number">3.1.2.4. </span>Cross-validation iterators for grouped data<a class="headerlink" href="#cross-validation-iterators-for-grouped-data" title="Link to this heading">¶</a></h3>
<p>The i.i.d. assumption is broken if the underlying generative process yield
groups of dependent samples.</p>
<p>Such a grouping of data is domain specific. An example would be when there is
medical data collected from multiple patients, with multiple samples taken from
each patient. And such data is likely to be dependent on the individual group.
In our example, the patient id for each sample will be its group identifier.</p>
<p>In this case we would like to know if a model trained on a particular set of
groups generalizes well to the unseen groups. To measure this, we need to
ensure that all the samples in the validation fold come from groups that are
not represented at all in the paired training fold.</p>
<p>The following cross-validation splitters can be used to do that.
The grouping identifier for the samples is specified via the <code class="docutils literal notranslate"><span class="pre">groups</span></code>
parameter.</p>
<section id="group-k-fold">
<span id="id6"></span><h4><span class="section-number">3.1.2.4.1. </span>Group k-fold<a class="headerlink" href="#group-k-fold" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> is a variation of k-fold which ensures that the same group is
not represented in both testing and training sets. For example if the data is
obtained from different subjects with several samples per-subject and if the
model is flexible enough to learn from highly person specific features it
could fail to generalize to new subjects. <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> makes it possible
to detect this kind of overfitting situations.</p>
<p>Imagine you have three subjects, each with an associated number from 1 to 3:</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.model_selection</span> <span class="kn">import</span> <span class="n">GroupKFold</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[</span><span class="mf">0.1</span><span class="p">,</span> <span class="mf">0.2</span><span class="p">,</span> <span class="mf">2.2</span><span class="p">,</span> <span class="mf">2.4</span><span class="p">,</span> <span class="mf">2.3</span><span class="p">,</span> <span class="mf">4.55</span><span class="p">,</span> <span class="mf">5.8</span><span class="p">,</span> <span class="mf">8.8</span><span class="p">,</span> <span class="mi">9</span><span class="p">,</span> <span class="mi">10</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"a"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"d"</span><span class="p">,</span> <span class="s2">"d"</span><span class="p">,</span> <span class="s2">"d"</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">groups</span> <span class="o">=</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="mi">1</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="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">gkf</span> <span class="o">=</span> <span class="n">GroupKFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">gkf</span><span class="o">.</span><span class="n">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">groups</span><span class="o">=</span><span class="n">groups</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[0 1 2 3 4 5] [6 7 8 9]</span>
<span class="go">[0 1 2 6 7 8 9] [3 4 5]</span>
<span class="go">[3 4 5 6 7 8 9] [0 1 2]</span>
</pre></div>
</div>
<p>Each subject is in a different testing fold, and the same subject is never in
both testing and training. Notice that the folds do not have exactly the same
size due to the imbalance in the data. If class proportions must be balanced
across folds, <a class="reference internal" href="generated/sklearn.model_selection.StratifiedGroupKFold.html#sklearn.model_selection.StratifiedGroupKFold" title="sklearn.model_selection.StratifiedGroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedGroupKFold</span></code></a> is a better option.</p>
<p>Here is a visualization of the cross-validation behavior.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_007.png" src="../_images/sphx_glr_plot_cv_indices_007.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
<p>Similar to <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a>, the test sets from <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> will form a
complete partition of all the data. Unlike <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a>, <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a>
is not randomized at all, whereas <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> is randomized when
<code class="docutils literal notranslate"><span class="pre">shuffle=True</span></code>.</p>
</section>
<section id="stratifiedgroupkfold">
<span id="stratified-group-k-fold"></span><h4><span class="section-number">3.1.2.4.2. </span>StratifiedGroupKFold<a class="headerlink" href="#stratifiedgroupkfold" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.StratifiedGroupKFold.html#sklearn.model_selection.StratifiedGroupKFold" title="sklearn.model_selection.StratifiedGroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedGroupKFold</span></code></a> is a cross-validation scheme that combines both
<a class="reference internal" href="generated/sklearn.model_selection.StratifiedKFold.html#sklearn.model_selection.StratifiedKFold" title="sklearn.model_selection.StratifiedKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">StratifiedKFold</span></code></a> and <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a>. The idea is to try to
preserve the distribution of classes in each split while keeping each group
within a single split. That might be useful when you have an unbalanced
dataset so that using just <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> might produce skewed splits.</p>
<p>Example:</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.model_selection</span> <span class="kn">import</span> <span class="n">StratifiedGroupKFold</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="mi">18</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">1</span><span class="p">]</span> <span class="o">*</span> <span class="mi">6</span> <span class="o">+</span> <span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">*</span> <span class="mi">12</span>
<span class="gp">>>> </span><span class="n">groups</span> <span class="o">=</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="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</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="mi">2</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="mi">4</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">6</span><span class="p">,</span> <span class="mi">6</span><span class="p">,</span> <span class="mi">6</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">sgkf</span> <span class="o">=</span> <span class="n">StratifiedGroupKFold</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">3</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">sgkf</span><span class="o">.</span><span class="n">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">groups</span><span class="o">=</span><span class="n">groups</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[ 0 2 3 4 5 6 7 10 11 15 16 17] [ 1 8 9 12 13 14]</span>
<span class="go">[ 0 1 4 5 6 7 8 9 11 12 13 14] [ 2 3 10 15 16 17]</span>
<span class="go">[ 1 2 3 8 9 10 12 13 14 15 16 17] [ 0 4 5 6 7 11]</span>
</pre></div>
</div>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Implementation notes</strong>
<span class="tooltiptext">Click for more details</span>
<a class="headerlink" href="#summary-anchor" title="Permalink to this heading">¶</a>
</summary>
<div class="card"></p>
<ul>
<li><p>With the current implementation full shuffle is not possible in most
scenarios. When shuffle=True, the following happens:</p>
<ol class="arabic simple">
<li><p>All groups are shuffled.</p></li>
<li><p>Groups are sorted by standard deviation of classes using stable sort.</p></li>
<li><p>Sorted groups are iterated over and assigned to folds.</p></li>
</ol>
<p>That means that only groups with the same standard deviation of class
distribution will be shuffled, which might be useful when each group has only
a single class.</p>
</li>
<li><p>The algorithm greedily assigns each group to one of n_splits test sets,
choosing the test set that minimises the variance in class distribution
across test sets. Group assignment proceeds from groups with highest to
lowest variance in class frequency, i.e. large groups peaked on one or few
classes are assigned first.</p></li>
<li><p>This split is suboptimal in a sense that it might produce imbalanced splits
even if perfect stratification is possible. If you have relatively close
distribution of classes in each group, using <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> is better.</p></li>
</ul>
<p></div>
</details></p>
<p>Here is a visualization of cross-validation behavior for uneven groups:</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_005.png" src="../_images/sphx_glr_plot_cv_indices_005.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
</section>
<section id="leave-one-group-out">
<span id="id7"></span><h4><span class="section-number">3.1.2.4.3. </span>Leave One Group Out<a class="headerlink" href="#leave-one-group-out" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.LeaveOneGroupOut.html#sklearn.model_selection.LeaveOneGroupOut" title="sklearn.model_selection.LeaveOneGroupOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneGroupOut</span></code></a> is a cross-validation scheme where each split holds
out samples belonging to one specific group. Group information is
provided via an array that encodes the group of each sample.</p>
<p>Each training set is thus constituted by all the samples except the ones
related to a specific group. This is the same as <a class="reference internal" href="generated/sklearn.model_selection.LeavePGroupsOut.html#sklearn.model_selection.LeavePGroupsOut" title="sklearn.model_selection.LeavePGroupsOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePGroupsOut</span></code></a> with
<code class="docutils literal notranslate"><span class="pre">n_groups=1</span></code> and the same as <a class="reference internal" href="generated/sklearn.model_selection.GroupKFold.html#sklearn.model_selection.GroupKFold" title="sklearn.model_selection.GroupKFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupKFold</span></code></a> with <code class="docutils literal notranslate"><span class="pre">n_splits</span></code> equal to the
number of unique labels passed to the <code class="docutils literal notranslate"><span class="pre">groups</span></code> parameter.</p>
<p>For example, in the cases of multiple experiments, <a class="reference internal" href="generated/sklearn.model_selection.LeaveOneGroupOut.html#sklearn.model_selection.LeaveOneGroupOut" title="sklearn.model_selection.LeaveOneGroupOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneGroupOut</span></code></a>
can be used to create a cross-validation based on the different experiments:
we create a training set using the samples of all the experiments except one:</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.model_selection</span> <span class="kn">import</span> <span class="n">LeaveOneGroupOut</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">10</span><span class="p">,</span> <span class="mi">50</span><span class="p">,</span> <span class="mi">60</span><span class="p">,</span> <span class="mi">70</span><span class="p">,</span> <span class="mi">80</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">1</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="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">groups</span> <span class="o">=</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="mi">2</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="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">logo</span> <span class="o">=</span> <span class="n">LeaveOneGroupOut</span><span class="p">()</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">logo</span><span class="o">.</span><span class="n">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">groups</span><span class="o">=</span><span class="n">groups</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[2 3 4 5 6] [0 1]</span>
<span class="go">[0 1 4 5 6] [2 3]</span>
<span class="go">[0 1 2 3] [4 5 6]</span>
</pre></div>
</div>
<p>Another common application is to use time information: for instance the
groups could be the year of collection of the samples and thus allow
for cross-validation against time-based splits.</p>
</section>
<section id="leave-p-groups-out">
<span id="id8"></span><h4><span class="section-number">3.1.2.4.4. </span>Leave P Groups Out<a class="headerlink" href="#leave-p-groups-out" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.LeavePGroupsOut.html#sklearn.model_selection.LeavePGroupsOut" title="sklearn.model_selection.LeavePGroupsOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePGroupsOut</span></code></a> is similar as <a class="reference internal" href="generated/sklearn.model_selection.LeaveOneGroupOut.html#sklearn.model_selection.LeaveOneGroupOut" title="sklearn.model_selection.LeaveOneGroupOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeaveOneGroupOut</span></code></a>, but removes
samples related to <span class="math notranslate nohighlight">\(P\)</span> groups for each training/test set. All possible
combinations of <span class="math notranslate nohighlight">\(P\)</span> groups are left out, meaning test sets will overlap
for <span class="math notranslate nohighlight">\(P>1\)</span>.</p>
<p>Example of Leave-2-Group Out:</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.model_selection</span> <span class="kn">import</span> <span class="n">LeavePGroupsOut</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">6</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">1</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="mi">2</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">groups</span> <span class="o">=</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="mi">2</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="mi">3</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">lpgo</span> <span class="o">=</span> <span class="n">LeavePGroupsOut</span><span class="p">(</span><span class="n">n_groups</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">lpgo</span><span class="o">.</span><span class="n">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">groups</span><span class="o">=</span><span class="n">groups</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="go">[4 5] [0 1 2 3]</span>
<span class="go">[2 3] [0 1 4 5]</span>
<span class="go">[0 1] [2 3 4 5]</span>
</pre></div>
</div>
</section>
<section id="group-shuffle-split">
<span id="id9"></span><h4><span class="section-number">3.1.2.4.5. </span>Group Shuffle Split<a class="headerlink" href="#group-shuffle-split" title="Link to this heading">¶</a></h4>
<p>The <a class="reference internal" href="generated/sklearn.model_selection.GroupShuffleSplit.html#sklearn.model_selection.GroupShuffleSplit" title="sklearn.model_selection.GroupShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupShuffleSplit</span></code></a> iterator behaves as a combination of
<a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a> and <a class="reference internal" href="generated/sklearn.model_selection.LeavePGroupsOut.html#sklearn.model_selection.LeavePGroupsOut" title="sklearn.model_selection.LeavePGroupsOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePGroupsOut</span></code></a>, and generates a
sequence of randomized partitions in which a subset of groups are held
out for each split. Each train/test split is performed independently meaning
there is no guaranteed relationship between successive test sets.</p>
<p>Here is a usage example:</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.model_selection</span> <span class="kn">import</span> <span class="n">GroupShuffleSplit</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[</span><span class="mf">0.1</span><span class="p">,</span> <span class="mf">0.2</span><span class="p">,</span> <span class="mf">2.2</span><span class="p">,</span> <span class="mf">2.4</span><span class="p">,</span> <span class="mf">2.3</span><span class="p">,</span> <span class="mf">4.55</span><span class="p">,</span> <span class="mf">5.8</span><span class="p">,</span> <span class="mf">0.001</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"a"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"a"</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">groups</span> <span class="o">=</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="mi">2</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="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">gss</span> <span class="o">=</span> <span class="n">GroupShuffleSplit</span><span class="p">(</span><span class="n">n_splits</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span> <span class="n">test_size</span><span class="o">=</span><span class="mf">0.5</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="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">test</span> <span class="ow">in</span> <span class="n">gss</span><span class="o">.</span><span class="n">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">groups</span><span class="o">=</span><span class="n">groups</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">%s</span><span class="s2"> </span><span class="si">%s</span><span class="s2">"</span> <span class="o">%</span> <span class="p">(</span><span class="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">))</span>
<span class="gp">...</span>
<span class="go">[0 1 2 3] [4 5 6 7]</span>
<span class="go">[2 3 6 7] [0 1 4 5]</span>
<span class="go">[2 3 4 5] [0 1 6 7]</span>
<span class="go">[4 5 6 7] [0 1 2 3]</span>
</pre></div>
</div>
<p>Here is a visualization of the cross-validation behavior.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/model_selection/plot_cv_indices.html"><img alt="../_images/sphx_glr_plot_cv_indices_011.png" src="../_images/sphx_glr_plot_cv_indices_011.png" style="width: 450.0px; height: 225.0px;" /></a>
</figure>
<p>This class is useful when the behavior of <a class="reference internal" href="generated/sklearn.model_selection.LeavePGroupsOut.html#sklearn.model_selection.LeavePGroupsOut" title="sklearn.model_selection.LeavePGroupsOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePGroupsOut</span></code></a> is
desired, but the number of groups is large enough that generating all
possible partitions with <span class="math notranslate nohighlight">\(P\)</span> groups withheld would be prohibitively
expensive. In such a scenario, <a class="reference internal" href="generated/sklearn.model_selection.GroupShuffleSplit.html#sklearn.model_selection.GroupShuffleSplit" title="sklearn.model_selection.GroupShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">GroupShuffleSplit</span></code></a> provides
a random sample (with replacement) of the train / test splits
generated by <a class="reference internal" href="generated/sklearn.model_selection.LeavePGroupsOut.html#sklearn.model_selection.LeavePGroupsOut" title="sklearn.model_selection.LeavePGroupsOut"><code class="xref py py-class docutils literal notranslate"><span class="pre">LeavePGroupsOut</span></code></a>.</p>
</section>
</section>
<section id="using-cross-validation-iterators-to-split-train-and-test">
<h3><span class="section-number">3.1.2.5. </span>Using cross-validation iterators to split train and test<a class="headerlink" href="#using-cross-validation-iterators-to-split-train-and-test" title="Link to this heading">¶</a></h3>
<p>The above group cross-validation functions may also be useful for splitting a
dataset into training and testing subsets. Note that the convenience
function <a class="reference internal" href="generated/sklearn.model_selection.train_test_split.html#sklearn.model_selection.train_test_split" title="sklearn.model_selection.train_test_split"><code class="xref py py-func docutils literal notranslate"><span class="pre">train_test_split</span></code></a> is a wrapper around <a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-func docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a>
and thus only allows for stratified splitting (using the class labels)
and cannot account for groups.</p>
<p>To perform the train and test split, use the indices for the train and test
subsets yielded by the generator output by the <code class="docutils literal notranslate"><span class="pre">split()</span></code> method of the
cross-validation splitter. For example:</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.model_selection</span> <span class="kn">import</span> <span class="n">GroupShuffleSplit</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">0.1</span><span class="p">,</span> <span class="mf">0.2</span><span class="p">,</span> <span class="mf">2.2</span><span class="p">,</span> <span class="mf">2.4</span><span class="p">,</span> <span class="mf">2.3</span><span class="p">,</span> <span class="mf">4.55</span><span class="p">,</span> <span class="mf">5.8</span><span class="p">,</span> <span class="mf">0.001</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="s2">"a"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"b"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"c"</span><span class="p">,</span> <span class="s2">"a"</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">groups</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</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="mi">2</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="mi">3</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">4</span><span class="p">])</span>
<span class="gp">>>> </span><span class="n">train_indx</span><span class="p">,</span> <span class="n">test_indx</span> <span class="o">=</span> <span class="nb">next</span><span class="p">(</span>
<span class="gp">... </span> <span class="n">GroupShuffleSplit</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">7</span><span class="p">)</span><span class="o">.</span><span class="n">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">groups</span><span class="p">)</span>
<span class="gp">... </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="gp">... </span> <span class="n">X</span><span class="p">[</span><span class="n">train_indx</span><span class="p">],</span> <span class="n">X</span><span class="p">[</span><span class="n">test_indx</span><span class="p">],</span> <span class="n">y</span><span class="p">[</span><span class="n">train_indx</span><span class="p">],</span> <span class="n">y</span><span class="p">[</span><span class="n">test_indx</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">X_train</span><span class="o">.</span><span class="n">shape</span><span class="p">,</span> <span class="n">X_test</span><span class="o">.</span><span class="n">shape</span>
<span class="go">((6,), (2,))</span>
<span class="gp">>>> </span><span class="n">np</span><span class="o">.</span><span class="n">unique</span><span class="p">(</span><span class="n">groups</span><span class="p">[</span><span class="n">train_indx</span><span class="p">]),</span> <span class="n">np</span><span class="o">.</span><span class="n">unique</span><span class="p">(</span><span class="n">groups</span><span class="p">[</span><span class="n">test_indx</span><span class="p">])</span>
<span class="go">(array([1, 2, 4]), array([3]))</span>
</pre></div>
</div>
</section>
<section id="cross-validation-of-time-series-data">
<span id="timeseries-cv"></span><h3><span class="section-number">3.1.2.6. </span>Cross validation of time series data<a class="headerlink" href="#cross-validation-of-time-series-data" title="Link to this heading">¶</a></h3>
<p>Time series data is characterized by the correlation between observations
that are near in time (<em>autocorrelation</em>). However, classical
cross-validation techniques such as <a class="reference internal" href="generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold" title="sklearn.model_selection.KFold"><code class="xref py py-class docutils literal notranslate"><span class="pre">KFold</span></code></a> and
<a class="reference internal" href="generated/sklearn.model_selection.ShuffleSplit.html#sklearn.model_selection.ShuffleSplit" title="sklearn.model_selection.ShuffleSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShuffleSplit</span></code></a> assume the samples are independent and
identically distributed, and would result in unreasonable correlation
between training and testing instances (yielding poor estimates of
generalization error) on time series data. Therefore, it is very important
to evaluate our model for time series data on the “future” observations
least like those that are used to train the model. To achieve this, one
solution is provided by <a class="reference internal" href="generated/sklearn.model_selection.TimeSeriesSplit.html#sklearn.model_selection.TimeSeriesSplit" title="sklearn.model_selection.TimeSeriesSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">TimeSeriesSplit</span></code></a>.</p>
<section id="time-series-split">
<span id="id10"></span><h4><span class="section-number">3.1.2.6.1. </span>Time Series Split<a class="headerlink" href="#time-series-split" title="Link to this heading">¶</a></h4>
<p><a class="reference internal" href="generated/sklearn.model_selection.TimeSeriesSplit.html#sklearn.model_selection.TimeSeriesSplit" title="sklearn.model_selection.TimeSeriesSplit"><code class="xref py py-class docutils literal notranslate"><span class="pre">TimeSeriesSplit</span></code></a> is a variation of <em>k-fold</em> which
returns first <span class="math notranslate nohighlight">\(k\)</span> folds as train set and the <span class="math notranslate nohighlight">\((k+1)\)</span> th
fold as test set. Note that unlike standard cross-validation methods,
successive training sets are supersets of those that come before them.
Also, it adds all surplus data to the first training partition, which
is always used to train the model.</p>
<p>This class can be used to cross-validate time series data samples
that are observed at fixed time intervals.</p>
<p>Example of 3-split time series cross-validation on a dataset with 6 samples:</p>