-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathcommon_pitfalls.html
913 lines (811 loc) · 83 KB
/
common_pitfalls.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
<!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="10. Common pitfalls and recommended practices" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/common_pitfalls.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="The purpose of this chapter is to illustrate some common pitfalls and anti-patterns that occur when using scikit-learn. It provides examples of what not to do, along with a corresponding correct ex..." />
<meta property="og:image" content="https://scikit-learn.org/stable/_static/scikit-learn-logo-small.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="The purpose of this chapter is to illustrate some common pitfalls and anti-patterns that occur when using scikit-learn. It provides examples of what not to do, along with a corresponding correct ex..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>10. Common pitfalls and recommended practices — scikit-learn 1.4.2 documentation</title>
<link rel="canonical" href="https://scikit-learn.org/stable/common_pitfalls.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="modules/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/v1.4.html" >What's new</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="glossary.html" >Glossary</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://scikit-learn.org/dev/developers/index.html" target="_blank" rel="noopener noreferrer">Development</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="faq.html" >FAQ</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="support.html" >Support</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="related_projects.html" >Related packages</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="roadmap.html" >Roadmap</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="governance.html" >Governance</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="about.html" >About us</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</li>
<li class="nav-item dropdown nav-more-item-dropdown">
<a class="sk-nav-link nav-link dropdown-toggle" href="#" id="navbarDropdown" role="button" data-toggle="dropdown" aria-haspopup="true" aria-expanded="false">More</a>
<div class="dropdown-menu" aria-labelledby="navbarDropdown">
<a class="sk-nav-dropdown-item dropdown-item" href="getting_started.html" >Getting Started</a>
<a class="sk-nav-dropdown-item dropdown-item" href="tutorial/index.html" >Tutorial</a>
<a class="sk-nav-dropdown-item dropdown-item" href="whats_new/v1.4.html" >What's new</a>
<a class="sk-nav-dropdown-item dropdown-item" href="glossary.html" >Glossary</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://scikit-learn.org/dev/developers/index.html" target="_blank" rel="noopener noreferrer">Development</a>
<a class="sk-nav-dropdown-item dropdown-item" href="faq.html" >FAQ</a>
<a class="sk-nav-dropdown-item dropdown-item" href="support.html" >Support</a>
<a class="sk-nav-dropdown-item dropdown-item" href="related_projects.html" >Related packages</a>
<a class="sk-nav-dropdown-item dropdown-item" href="roadmap.html" >Roadmap</a>
<a class="sk-nav-dropdown-item dropdown-item" href="governance.html" >Governance</a>
<a class="sk-nav-dropdown-item dropdown-item" href="about.html" >About us</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</div>
</li>
</ul>
<div id="searchbox" role="search">
<div class="searchformwrapper">
<form class="search" action="search.html" method="get">
<input class="sk-search-text-input" type="text" name="q" aria-labelledby="searchlabel" />
<input class="sk-search-text-btn" type="submit" value="Go" />
</form>
</div>
</div>
</div>
</div>
</nav>
<div class="d-flex" id="sk-doc-wrapper">
<input type="checkbox" name="sk-toggle-checkbox" id="sk-toggle-checkbox">
<label id="sk-sidemenu-toggle" class="sk-btn-toggle-toc btn sk-btn-primary" for="sk-toggle-checkbox">Toggle Menu</label>
<div id="sk-sidebar-wrapper" class="border-right">
<div class="sk-sidebar-toc-wrapper">
<div class="btn-group w-100 mb-2" role="group" aria-label="rellinks">
<a href="model_persistence.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="9. Model persistence">Prev</a><a href="user_guide.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="User Guide">Up</a>
<a href="dispatching.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="11. Dispatching">Next</a>
</div>
<div class="alert alert-danger p-1 mb-2" role="alert">
<p class="text-center mb-0">
<strong>scikit-learn 1.4.2</strong><br/>
<a href="https://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 href="user_guide.html" class="sk-toc-active">User Guide</a>
</li>
<ul>
<li>
<a href="supervised_learning.html" class="">1. Supervised learning</a>
</li>
<li>
<a href="unsupervised_learning.html" class="">2. Unsupervised learning</a>
</li>
<li>
<a href="model_selection.html" class="">3. Model selection and evaluation</a>
</li>
<li>
<a href="inspection.html" class="">4. Inspection</a>
</li>
<li>
<a href="visualizations.html" class="">5. Visualizations</a>
</li>
<li>
<a href="data_transforms.html" class="">6. Dataset transformations</a>
</li>
<li>
<a href="datasets.html" class="">7. Dataset loading utilities</a>
</li>
<li>
<a href="computing.html" class="">8. Computing with scikit-learn</a>
</li>
<li>
<a href="model_persistence.html" class="">9. Model persistence</a>
</li>
<li>
<a href="" class="sk-toc-active">10. Common pitfalls and recommended practices</a>
</li>
<li>
<a href="dispatching.html" class="">11. Dispatching</a>
</li>
</ul>
</ul>
</div>
</div>
</div>
<div id="sk-page-content-wrapper">
<div class="sk-page-content container-fluid body px-md-3" role="main">
<style type="text/css">
div.body div.toctree-wrapper ul {
padding-left: 0;
}
div.body li.toctree-l1 {
padding: 0 0 0.5em 0;
list-style-type: none;
font-size: 150%;
font-weight: bold;
}
div.body li.toctree-l2 {
font-size: 70%;
list-style-type: square;
font-weight: normal;
margin-left: 40px;
}
div.body li.toctree-l3 {
font-size: 85%;
list-style-type: circle;
font-weight: normal;
margin-left: 40px;
}
div.body li.toctree-l4 {
margin-left: 40px;
}
</style><section id="common-pitfalls-and-recommended-practices">
<span id="common-pitfalls"></span><h1><span class="section-number">10. </span>Common pitfalls and recommended practices<a class="headerlink" href="#common-pitfalls-and-recommended-practices" title="Link to this heading">¶</a></h1>
<p>The purpose of this chapter is to illustrate some common pitfalls and
anti-patterns that occur when using scikit-learn. It provides
examples of what <strong>not</strong> to do, along with a corresponding correct
example.</p>
<section id="inconsistent-preprocessing">
<h2><span class="section-number">10.1. </span>Inconsistent preprocessing<a class="headerlink" href="#inconsistent-preprocessing" title="Link to this heading">¶</a></h2>
<p>scikit-learn provides a library of <a class="reference internal" href="data_transforms.html#data-transforms"><span class="std std-ref">Dataset transformations</span></a>, which
may clean (see <a class="reference internal" href="modules/preprocessing.html#preprocessing"><span class="std std-ref">Preprocessing data</span></a>), reduce
(see <a class="reference internal" href="modules/unsupervised_reduction.html#data-reduction"><span class="std std-ref">Unsupervised dimensionality reduction</span></a>), expand (see <a class="reference internal" href="modules/kernel_approximation.html#kernel-approximation"><span class="std std-ref">Kernel Approximation</span></a>)
or generate (see <a class="reference internal" href="modules/feature_extraction.html#feature-extraction"><span class="std std-ref">Feature extraction</span></a>) feature representations.
If these data transforms are used when training a model, they also
must be used on subsequent datasets, whether it’s test data or
data in a production system. Otherwise, the feature space will change,
and the model will not be able to perform effectively.</p>
<p>For the following example, let’s create a synthetic dataset with a
single feature:</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.datasets</span> <span class="kn">import</span> <span class="n">make_regression</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="n">random_state</span> <span class="o">=</span> <span class="mi">42</span>
<span class="gp">>>> </span><span class="n">X</span><span class="p">,</span> <span class="n">y</span> <span class="o">=</span> <span class="n">make_regression</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="n">n_features</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">noise</span><span class="o">=</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">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="n">random_state</span><span class="p">)</span>
</pre></div>
</div>
<p><strong>Wrong</strong></p>
<p>The train dataset is scaled, but not the test dataset, so model
performance on the test dataset is worse than expected:</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">mean_squared_error</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.linear_model</span> <span class="kn">import</span> <span class="n">LinearRegression</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.preprocessing</span> <span class="kn">import</span> <span class="n">StandardScaler</span>
<span class="gp">>>> </span><span class="n">scaler</span> <span class="o">=</span> <span class="n">StandardScaler</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">fit_transform</span><span class="p">(</span><span class="n">X_train</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">model</span> <span class="o">=</span> <span class="n">LinearRegression</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">mean_squared_error</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">model</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test</span><span class="p">))</span>
<span class="go">62.80...</span>
</pre></div>
</div>
<p><strong>Right</strong></p>
<p>Instead of passing the non-transformed <code class="docutils literal notranslate"><span class="pre">X_test</span></code> to <code class="docutils literal notranslate"><span class="pre">predict</span></code>, we should
transform the test data, the same way we transformed the training data:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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">mean_squared_error</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">model</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test_transformed</span><span class="p">))</span>
<span class="go">0.90...</span>
</pre></div>
</div>
<p>Alternatively, we recommend using a <a class="reference internal" href="modules/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>, which makes it easier to chain transformations
with estimators, and reduces the possibility of forgetting a transformation:</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">model</span> <span class="o">=</span> <span class="n">make_pipeline</span><span class="p">(</span><span class="n">StandardScaler</span><span class="p">(),</span> <span class="n">LinearRegression</span><span class="p">())</span>
<span class="gp">>>> </span><span class="n">model</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="go">Pipeline(steps=[('standardscaler', StandardScaler()),</span>
<span class="go"> ('linearregression', LinearRegression())])</span>
<span class="gp">>>> </span><span class="n">mean_squared_error</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">model</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test</span><span class="p">))</span>
<span class="go">0.90...</span>
</pre></div>
</div>
<p>Pipelines also help avoiding another common pitfall: leaking the test data
into the training data.</p>
</section>
<section id="data-leakage">
<span id="id1"></span><h2><span class="section-number">10.2. </span>Data leakage<a class="headerlink" href="#data-leakage" title="Link to this heading">¶</a></h2>
<p>Data leakage occurs when information that would not be available at prediction
time is used when building the model. This results in overly optimistic
performance estimates, for example from <a class="reference internal" href="modules/cross_validation.html#cross-validation"><span class="std std-ref">cross-validation</span></a>, and thus poorer performance when the model is used
on actually novel data, for example during production.</p>
<p>A common cause is not keeping the test and train data subsets separate.
Test data should never be used to make choices about the model.
<strong>The general rule is to never call</strong> <code class="docutils literal notranslate"><span class="pre">fit</span></code> <strong>on the test data</strong>. While this
may sound obvious, this is easy to miss in some cases, for example when
applying certain pre-processing steps.</p>
<p>Although both train and test data subsets should receive the same
preprocessing transformation (as described in the previous section), it is
important that these transformations are only learnt from the training data.
For example, if you have a
normalization step where you divide by the average value, the average should
be the average of the train subset, <strong>not</strong> the average of all the data. If the
test subset is included in the average calculation, information from the test
subset is influencing the model.</p>
<section id="how-to-avoid-data-leakage">
<h3><span class="section-number">10.2.1. </span>How to avoid data leakage<a class="headerlink" href="#how-to-avoid-data-leakage" title="Link to this heading">¶</a></h3>
<p>Below are some tips on avoiding data leakage:</p>
<ul>
<li><p>Always split the data into train and test subsets first, particularly
before any preprocessing steps.</p></li>
<li><p>Never include test data when using the <code class="docutils literal notranslate"><span class="pre">fit</span></code> and <code class="docutils literal notranslate"><span class="pre">fit_transform</span></code>
methods. Using all the data, e.g., <code class="docutils literal notranslate"><span class="pre">fit(X)</span></code>, can result in overly optimistic
scores.</p>
<p>Conversely, the <code class="docutils literal notranslate"><span class="pre">transform</span></code> method should be used on both train and test
subsets as the same preprocessing should be applied to all the data.
This can be achieved by using <code class="docutils literal notranslate"><span class="pre">fit_transform</span></code> on the train subset and
<code class="docutils literal notranslate"><span class="pre">transform</span></code> on the test subset.</p>
</li>
<li><p>The scikit-learn <a class="reference internal" href="modules/compose.html#pipeline"><span class="std std-ref">pipeline</span></a> is a great way to prevent data
leakage as it ensures that the appropriate method is performed on the
correct data subset. The pipeline is ideal for use in cross-validation
and hyper-parameter tuning functions.</p></li>
</ul>
<p>An example of data leakage during preprocessing is detailed below.</p>
</section>
<section id="data-leakage-during-pre-processing">
<h3><span class="section-number">10.2.2. </span>Data leakage during pre-processing<a class="headerlink" href="#data-leakage-during-pre-processing" title="Link to this heading">¶</a></h3>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>We here choose to illustrate data leakage with a feature selection step.
This risk of leakage is however relevant with almost all transformations
in scikit-learn, including (but not limited to)
<a class="reference internal" href="modules/generated/sklearn.preprocessing.StandardScaler.html#sklearn.preprocessing.StandardScaler" title="sklearn.preprocessing.StandardScaler"><code class="xref py py-class docutils literal notranslate"><span class="pre">StandardScaler</span></code></a>,
<a class="reference internal" href="modules/generated/sklearn.impute.SimpleImputer.html#sklearn.impute.SimpleImputer" title="sklearn.impute.SimpleImputer"><code class="xref py py-class docutils literal notranslate"><span class="pre">SimpleImputer</span></code></a>, and
<a class="reference internal" href="modules/generated/sklearn.decomposition.PCA.html#sklearn.decomposition.PCA" title="sklearn.decomposition.PCA"><code class="xref py py-class docutils literal notranslate"><span class="pre">PCA</span></code></a>.</p>
</div>
<p>A number of <a class="reference internal" href="modules/feature_selection.html#feature-selection"><span class="std std-ref">Feature selection</span></a> functions are available in scikit-learn.
They can help remove irrelevant, redundant and noisy features as well as
improve your model build time and performance. As with any other type of
preprocessing, feature selection should <strong>only</strong> use the training data.
Including the test data in feature selection will optimistically bias your
model.</p>
<p>To demonstrate we will create this binary classification problem with
10,000 randomly generated features:</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="n">n_samples</span><span class="p">,</span> <span class="n">n_features</span><span class="p">,</span> <span class="n">n_classes</span> <span class="o">=</span> <span class="mi">200</span><span class="p">,</span> <span class="mi">10000</span><span class="p">,</span> <span class="mi">2</span>
<span class="gp">>>> </span><span class="n">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">42</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">rng</span><span class="o">.</span><span class="n">standard_normal</span><span class="p">((</span><span class="n">n_samples</span><span class="p">,</span> <span class="n">n_features</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">rng</span><span class="o">.</span><span class="n">choice</span><span class="p">(</span><span class="n">n_classes</span><span class="p">,</span> <span class="n">n_samples</span><span class="p">)</span>
</pre></div>
</div>
<p><strong>Wrong</strong></p>
<p>Using all the data to perform feature selection results in an accuracy score
much higher than chance, even though our targets are completely random.
This randomness means that our <code class="docutils literal notranslate"><span class="pre">X</span></code> and <code class="docutils literal notranslate"><span class="pre">y</span></code> are independent and we thus expect
the accuracy to be around 0.5. However, since the feature selection step
‘sees’ the test data, the model has an unfair advantage. In the incorrect
example below we first use all the data for feature selection and then split
the data into training and test subsets for model fitting. The result is a
much higher than expected accuracy score:</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">train_test_split</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.feature_selection</span> <span class="kn">import</span> <span class="n">SelectKBest</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.ensemble</span> <span class="kn">import</span> <span class="n">GradientBoostingClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.metrics</span> <span class="kn">import</span> <span class="n">accuracy_score</span>
<span class="gp">>>> </span><span class="c1"># Incorrect preprocessing: the entire data is transformed</span>
<span class="gp">>>> </span><span class="n">X_selected</span> <span class="o">=</span> <span class="n">SelectKBest</span><span class="p">(</span><span class="n">k</span><span class="o">=</span><span class="mi">25</span><span class="p">)</span><span class="o">.</span><span class="n">fit_transform</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">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_selected</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">42</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">gbc</span> <span class="o">=</span> <span class="n">GradientBoostingClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">gbc</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="go">GradientBoostingClassifier(random_state=1)</span>
<span class="gp">>>> </span><span class="n">y_pred</span> <span class="o">=</span> <span class="n">gbc</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">accuracy_score</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">y_pred</span><span class="p">)</span>
<span class="go">0.76</span>
</pre></div>
</div>
<p><strong>Right</strong></p>
<p>To prevent data leakage, it is good practice to split your data into train
and test subsets <strong>first</strong>. Feature selection can then be formed using just
the train dataset. Notice that whenever we use <code class="docutils literal notranslate"><span class="pre">fit</span></code> or <code class="docutils literal notranslate"><span class="pre">fit_transform</span></code>, we
only use the train dataset. The score is now what we would expect for the
data, close to chance:</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">random_state</span><span class="o">=</span><span class="mi">42</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">select</span> <span class="o">=</span> <span class="n">SelectKBest</span><span class="p">(</span><span class="n">k</span><span class="o">=</span><span class="mi">25</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X_train_selected</span> <span class="o">=</span> <span class="n">select</span><span class="o">.</span><span class="n">fit_transform</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">gbc</span> <span class="o">=</span> <span class="n">GradientBoostingClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">gbc</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X_train_selected</span><span class="p">,</span> <span class="n">y_train</span><span class="p">)</span>
<span class="go">GradientBoostingClassifier(random_state=1)</span>
<span class="gp">>>> </span><span class="n">X_test_selected</span> <span class="o">=</span> <span class="n">select</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">y_pred</span> <span class="o">=</span> <span class="n">gbc</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test_selected</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">accuracy_score</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">y_pred</span><span class="p">)</span>
<span class="go">0.46</span>
</pre></div>
</div>
<p>Here again, we recommend using a <a class="reference internal" href="modules/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> to chain
together the feature selection and model estimators. The pipeline ensures
that only the training data is used when performing <code class="docutils literal notranslate"><span class="pre">fit</span></code> and the test data
is used only for calculating the accuracy score:</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">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">random_state</span><span class="o">=</span><span class="mi">42</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">pipeline</span> <span class="o">=</span> <span class="n">make_pipeline</span><span class="p">(</span><span class="n">SelectKBest</span><span class="p">(</span><span class="n">k</span><span class="o">=</span><span class="mi">25</span><span class="p">),</span>
<span class="gp">... </span> <span class="n">GradientBoostingClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">))</span>
<span class="gp">>>> </span><span class="n">pipeline</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="go">Pipeline(steps=[('selectkbest', SelectKBest(k=25)),</span>
<span class="go"> ('gradientboostingclassifier',</span>
<span class="go"> GradientBoostingClassifier(random_state=1))])</span>
<span class="gp">>>> </span><span class="n">y_pred</span> <span class="o">=</span> <span class="n">pipeline</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X_test</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">accuracy_score</span><span class="p">(</span><span class="n">y_test</span><span class="p">,</span> <span class="n">y_pred</span><span class="p">)</span>
<span class="go">0.46</span>
</pre></div>
</div>
<p>The pipeline can also be fed into a cross-validation
function such as <a class="reference internal" href="modules/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>.
Again, the pipeline ensures that the correct data subset and estimator
method is used during fitting and predicting:</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">scores</span> <span class="o">=</span> <span class="n">cross_val_score</span><span class="p">(</span><span class="n">pipeline</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="sa">f</span><span class="s2">"Mean accuracy: </span><span class="si">{</span><span class="n">scores</span><span class="o">.</span><span class="n">mean</span><span class="p">()</span><span class="si">:</span><span class="s2">.2f</span><span class="si">}</span><span class="s2">+/-</span><span class="si">{</span><span class="n">scores</span><span class="o">.</span><span class="n">std</span><span class="p">()</span><span class="si">:</span><span class="s2">.2f</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
<span class="go">Mean accuracy: 0.46+/-0.07</span>
</pre></div>
</div>
</section>
</section>
<section id="controlling-randomness">
<span id="randomness"></span><h2><span class="section-number">10.3. </span>Controlling randomness<a class="headerlink" href="#controlling-randomness" title="Link to this heading">¶</a></h2>
<p>Some scikit-learn objects are inherently random. These are usually estimators
(e.g. <a class="reference internal" href="modules/generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier" title="sklearn.ensemble.RandomForestClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">RandomForestClassifier</span></code></a>) and cross-validation
splitters (e.g. <a class="reference internal" href="modules/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 randomness of
these objects is controlled via their <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter, as described
in the <a class="reference internal" href="glossary.html#term-random_state"><span class="xref std std-term">Glossary</span></a>. This section expands on the glossary
entry, and describes good practices and common pitfalls w.r.t. this
subtle parameter.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Recommendation summary</p>
<p>For an optimal robustness of cross-validation (CV) results, pass
<code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instances when creating estimators, or leave <code class="docutils literal notranslate"><span class="pre">random_state</span></code>
to <code class="docutils literal notranslate"><span class="pre">None</span></code>. Passing integers to CV splitters is usually the safest option
and is preferable; passing <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instances to splitters may
sometimes be useful to achieve very specific use-cases.
For both estimators and splitters, passing an integer vs passing an
instance (or <code class="docutils literal notranslate"><span class="pre">None</span></code>) leads to subtle but significant differences,
especially for CV procedures. These differences are important to
understand when reporting results.</p>
<p>For reproducible results across executions, remove any use of
<code class="docutils literal notranslate"><span class="pre">random_state=None</span></code>.</p>
</div>
<section id="using-none-or-randomstate-instances-and-repeated-calls-to-fit-and-split">
<h3><span class="section-number">10.3.1. </span>Using <code class="docutils literal notranslate"><span class="pre">None</span></code> or <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instances, and repeated calls to <code class="docutils literal notranslate"><span class="pre">fit</span></code> and <code class="docutils literal notranslate"><span class="pre">split</span></code><a class="headerlink" href="#using-none-or-randomstate-instances-and-repeated-calls-to-fit-and-split" title="Link to this heading">¶</a></h3>
<p>The <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter determines whether multiple calls to <a class="reference internal" href="glossary.html#term-fit"><span class="xref std std-term">fit</span></a>
(for estimators) or to <a class="reference internal" href="glossary.html#term-split"><span class="xref std std-term">split</span></a> (for CV splitters) will produce the same
results, according to these rules:</p>
<ul class="simple">
<li><p>If an integer is passed, calling <code class="docutils literal notranslate"><span class="pre">fit</span></code> or <code class="docutils literal notranslate"><span class="pre">split</span></code> multiple times always
yields the same results.</p></li>
<li><p>If <code class="docutils literal notranslate"><span class="pre">None</span></code> or a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance is passed: <code class="docutils literal notranslate"><span class="pre">fit</span></code> and <code class="docutils literal notranslate"><span class="pre">split</span></code> will
yield different results each time they are called, and the succession of
calls explores all sources of entropy. <code class="docutils literal notranslate"><span class="pre">None</span></code> is the default value for all
<code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameters.</p></li>
</ul>
<p>We here illustrate these rules for both estimators and CV splitters.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Since passing <code class="docutils literal notranslate"><span class="pre">random_state=None</span></code> is equivalent to passing the global
<code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance from <code class="docutils literal notranslate"><span class="pre">numpy</span></code>
(<code class="docutils literal notranslate"><span class="pre">random_state=np.random.mtrand._rand</span></code>), we will not explicitly mention
<code class="docutils literal notranslate"><span class="pre">None</span></code> here. Everything that applies to instances also applies to using
<code class="docutils literal notranslate"><span class="pre">None</span></code>.</p>
</div>
<section id="estimators">
<h4><span class="section-number">10.3.1.1. </span>Estimators<a class="headerlink" href="#estimators" title="Link to this heading">¶</a></h4>
<p>Passing instances means that calling <code class="docutils literal notranslate"><span class="pre">fit</span></code> multiple times will not yield the
same results, even if the estimator is fitted on the same data and with the
same hyper-parameters:</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.linear_model</span> <span class="kn">import</span> <span class="n">SGDClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="kn">import</span> <span class="n">make_classification</span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="gp">>>> </span><span class="n">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">0</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="o">=</span> <span class="n">make_classification</span><span class="p">(</span><span class="n">n_features</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">sgd</span> <span class="o">=</span> <span class="n">SGDClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">sgd</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span><span class="o">.</span><span class="n">coef_</span>
<span class="go">array([[ 8.85418642, 4.79084103, -3.13077794, 8.11915045, -0.56479934]])</span>
<span class="gp">>>> </span><span class="n">sgd</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span><span class="o">.</span><span class="n">coef_</span>
<span class="go">array([[ 6.70814003, 5.25291366, -7.55212743, 5.18197458, 1.37845099]])</span>
</pre></div>
</div>
<p>We can see from the snippet above that repeatedly calling <code class="docutils literal notranslate"><span class="pre">sgd.fit</span></code> has
produced different models, even if the data was the same. This is because the
Random Number Generator (RNG) of the estimator is consumed (i.e. mutated)
when <code class="docutils literal notranslate"><span class="pre">fit</span></code> is called, and this mutated RNG will be used in the subsequent
calls to <code class="docutils literal notranslate"><span class="pre">fit</span></code>. In addition, the <code class="docutils literal notranslate"><span class="pre">rng</span></code> object is shared across all objects
that use it, and as a consequence, these objects become somewhat
inter-dependent. For example, two estimators that share the same
<code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance will influence each other, as we will see later when
we discuss cloning. This point is important to keep in mind when debugging.</p>
<p>If we had passed an integer to the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter of the
<a class="reference internal" href="modules/generated/sklearn.linear_model.SGDClassifier.html#sklearn.linear_model.SGDClassifier" title="sklearn.linear_model.SGDClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">SGDClassifier</span></code></a>, we would have obtained the
same models, and thus the same scores each time. When we pass an integer, the
same RNG is used across all calls to <code class="docutils literal notranslate"><span class="pre">fit</span></code>. What internally happens is that
even though the RNG is consumed when <code class="docutils literal notranslate"><span class="pre">fit</span></code> is called, it is always reset to
its original state at the beginning of <code class="docutils literal notranslate"><span class="pre">fit</span></code>.</p>
</section>
<section id="cv-splitters">
<h4><span class="section-number">10.3.1.2. </span>CV splitters<a class="headerlink" href="#cv-splitters" title="Link to this heading">¶</a></h4>
<p>Randomized CV splitters have a similar behavior when a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code>
instance is passed; calling <code class="docutils literal notranslate"><span class="pre">split</span></code> multiple times yields different data
splits:</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">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="o">=</span> <span class="n">y</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">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</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">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="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="n">rng</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">cv</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="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">)</span>
<span class="go">[0 3 5 6 7] [1 2 4 8 9]</span>
<span class="go">[1 2 4 8 9] [0 3 5 6 7]</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">cv</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="n">train</span><span class="p">,</span> <span class="n">test</span><span class="p">)</span>
<span class="go">[0 4 6 7 8] [1 2 3 5 9]</span>
<span class="go">[1 2 3 5 9] [0 4 6 7 8]</span>
</pre></div>
</div>
<p>We can see that the splits are different from the second time <code class="docutils literal notranslate"><span class="pre">split</span></code> is
called. This may lead to unexpected results if you compare the performance of
multiple estimators by calling <code class="docutils literal notranslate"><span class="pre">split</span></code> many times, as we will see in the next
section.</p>
</section>
</section>
<section id="common-pitfalls-and-subtleties">
<h3><span class="section-number">10.3.2. </span>Common pitfalls and subtleties<a class="headerlink" href="#common-pitfalls-and-subtleties" title="Link to this heading">¶</a></h3>
<p>While the rules that govern the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter are seemingly simple,
they do however have some subtle implications. In some cases, this can even
lead to wrong conclusions.</p>
<section id="id2">
<h4><span class="section-number">10.3.2.1. </span>Estimators<a class="headerlink" href="#id2" title="Link to this heading">¶</a></h4>
<p><strong>Different `random_state` types lead to different cross-validation
procedures</strong></p>
<p>Depending on the type of the <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter, estimators will behave
differently, especially in cross-validation procedures. Consider the
following snippet:</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.ensemble</span> <span class="kn">import</span> <span class="n">RandomForestClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="kn">import</span> <span class="n">make_classification</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.model_selection</span> <span class="kn">import</span> <span class="n">cross_val_score</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">make_classification</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">rf_123</span> <span class="o">=</span> <span class="n">RandomForestClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">123</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">cross_val_score</span><span class="p">(</span><span class="n">rf_123</span><span class="p">,</span> <span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">array([0.85, 0.95, 0.95, 0.9 , 0.9 ])</span>
<span class="gp">>>> </span><span class="n">rf_inst</span> <span class="o">=</span> <span class="n">RandomForestClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</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">rf_inst</span><span class="p">,</span> <span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">array([0.9 , 0.95, 0.95, 0.9 , 0.9 ])</span>
</pre></div>
</div>
<p>We see that the cross-validated scores of <code class="docutils literal notranslate"><span class="pre">rf_123</span></code> and <code class="docutils literal notranslate"><span class="pre">rf_inst</span></code> are
different, as should be expected since we didn’t pass the same <code class="docutils literal notranslate"><span class="pre">random_state</span></code>
parameter. However, the difference between these scores is more subtle than
it looks, and <strong>the cross-validation procedures that were performed by</strong>
<a class="reference internal" href="modules/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> <strong>significantly differ in
each case</strong>:</p>
<ul class="simple">
<li><p>Since <code class="docutils literal notranslate"><span class="pre">rf_123</span></code> was passed an integer, every call to <code class="docutils literal notranslate"><span class="pre">fit</span></code> uses the same RNG:
this means that all random characteristics of the random forest estimator
will be the same for each of the 5 folds of the CV procedure. In
particular, the (randomly chosen) subset of features of the estimator will
be the same across all folds.</p></li>
<li><p>Since <code class="docutils literal notranslate"><span class="pre">rf_inst</span></code> was passed a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance, each call to <code class="docutils literal notranslate"><span class="pre">fit</span></code>
starts from a different RNG. As a result, the random subset of features
will be different for each folds.</p></li>
</ul>
<p>While having a constant estimator RNG across folds isn’t inherently wrong, we
usually want CV results that are robust w.r.t. the estimator’s randomness. As
a result, passing an instance instead of an integer may be preferable, since
it will allow the estimator RNG to vary for each fold.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Here, <a class="reference internal" href="modules/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> will use a
non-randomized CV splitter (as is the default), so both estimators will
be evaluated on the same splits. This section is not about variability in
the splits. Also, whether we pass an integer or an instance to
<a class="reference internal" href="modules/generated/sklearn.datasets.make_classification.html#sklearn.datasets.make_classification" title="sklearn.datasets.make_classification"><code class="xref py py-func docutils literal notranslate"><span class="pre">make_classification</span></code></a> isn’t relevant for our
illustration purpose: what matters is what we pass to the
<a class="reference internal" href="modules/generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier" title="sklearn.ensemble.RandomForestClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">RandomForestClassifier</span></code></a> estimator.</p>
</div>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Cloning</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>
<p>Another subtle side effect of passing <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instances is how
<a class="reference internal" href="modules/generated/sklearn.base.clone.html#sklearn.base.clone" title="sklearn.base.clone"><code class="xref py py-func docutils literal notranslate"><span class="pre">clone</span></code></a> will work:</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">clone</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.ensemble</span> <span class="kn">import</span> <span class="n">RandomForestClassifier</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">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">a</span> <span class="o">=</span> <span class="n">RandomForestClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">b</span> <span class="o">=</span> <span class="n">clone</span><span class="p">(</span><span class="n">a</span><span class="p">)</span>
</pre></div>
</div>
<p>Since a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance was passed to <code class="docutils literal notranslate"><span class="pre">a</span></code>, <code class="docutils literal notranslate"><span class="pre">a</span></code> and <code class="docutils literal notranslate"><span class="pre">b</span></code> are not clones
in the strict sense, but rather clones in the statistical sense: <code class="docutils literal notranslate"><span class="pre">a</span></code> and <code class="docutils literal notranslate"><span class="pre">b</span></code>
will still be different models, even when calling <code class="docutils literal notranslate"><span class="pre">fit(X,</span> <span class="pre">y)</span></code> on the same
data. Moreover, <code class="docutils literal notranslate"><span class="pre">a</span></code> and <code class="docutils literal notranslate"><span class="pre">b</span></code> will influence each-other since they share the
same internal RNG: calling <code class="docutils literal notranslate"><span class="pre">a.fit</span></code> will consume <code class="docutils literal notranslate"><span class="pre">b</span></code>’s RNG, and calling
<code class="docutils literal notranslate"><span class="pre">b.fit</span></code> will consume <code class="docutils literal notranslate"><span class="pre">a</span></code>’s RNG, since they are the same. This bit is true for
any estimators that share a <code class="docutils literal notranslate"><span class="pre">random_state</span></code> parameter; it is not specific to
clones.</p>
<p>If an integer were passed, <code class="docutils literal notranslate"><span class="pre">a</span></code> and <code class="docutils literal notranslate"><span class="pre">b</span></code> would be exact clones and they would not
influence each other.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>Even though <a class="reference internal" href="modules/generated/sklearn.base.clone.html#sklearn.base.clone" title="sklearn.base.clone"><code class="xref py py-func docutils literal notranslate"><span class="pre">clone</span></code></a> is rarely used in user code, it is
called pervasively throughout scikit-learn codebase: in particular, most
meta-estimators that accept non-fitted estimators call
<a class="reference internal" href="modules/generated/sklearn.base.clone.html#sklearn.base.clone" title="sklearn.base.clone"><code class="xref py py-func docutils literal notranslate"><span class="pre">clone</span></code></a> internally
(<a class="reference internal" href="modules/generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">GridSearchCV</span></code></a>,
<a class="reference internal" href="modules/generated/sklearn.ensemble.StackingClassifier.html#sklearn.ensemble.StackingClassifier" title="sklearn.ensemble.StackingClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">StackingClassifier</span></code></a>,
<a class="reference internal" href="modules/generated/sklearn.calibration.CalibratedClassifierCV.html#sklearn.calibration.CalibratedClassifierCV" title="sklearn.calibration.CalibratedClassifierCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">CalibratedClassifierCV</span></code></a>, etc.).</p>
</div>
<p></div>
</details></p>
</section>
<section id="id3">
<h4><span class="section-number">10.3.2.2. </span>CV splitters<a class="headerlink" href="#id3" title="Link to this heading">¶</a></h4>
<p>When passed a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance, CV splitters yield different splits
each time <code class="docutils literal notranslate"><span class="pre">split</span></code> is called. When comparing different estimators, this can
lead to overestimating the variance of the difference in performance between
the estimators:</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.naive_bayes</span> <span class="kn">import</span> <span class="n">GaussianNB</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.discriminant_analysis</span> <span class="kn">import</span> <span class="n">LinearDiscriminantAnalysis</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="kn">import</span> <span class="n">make_classification</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.model_selection</span> <span class="kn">import</span> <span class="n">KFold</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="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">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">0</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="o">=</span> <span class="n">make_classification</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">cv</span> <span class="o">=</span> <span class="n">KFold</span><span class="p">(</span><span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">lda</span> <span class="o">=</span> <span class="n">LinearDiscriminantAnalysis</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">nb</span> <span class="o">=</span> <span class="n">GaussianNB</span><span class="p">()</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">est</span> <span class="ow">in</span> <span class="p">(</span><span class="n">lda</span><span class="p">,</span> <span class="n">nb</span><span class="p">):</span>
<span class="gp">... </span> <span class="nb">print</span><span class="p">(</span><span class="n">cross_val_score</span><span class="p">(</span><span class="n">est</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">[0.8 0.75 0.75 0.7 0.85]</span>
<span class="go">[0.85 0.95 0.95 0.85 0.95]</span>
</pre></div>
</div>
<p>Directly comparing the performance of the
<a class="reference internal" href="modules/generated/sklearn.discriminant_analysis.LinearDiscriminantAnalysis.html#sklearn.discriminant_analysis.LinearDiscriminantAnalysis" title="sklearn.discriminant_analysis.LinearDiscriminantAnalysis"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearDiscriminantAnalysis</span></code></a> estimator
vs the <a class="reference internal" href="modules/generated/sklearn.naive_bayes.GaussianNB.html#sklearn.naive_bayes.GaussianNB" title="sklearn.naive_bayes.GaussianNB"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianNB</span></code></a> estimator <strong>on each fold</strong> would
be a mistake: <strong>the splits on which the estimators are evaluated are
different</strong>. Indeed, <a class="reference internal" href="modules/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> will
internally call <code class="docutils literal notranslate"><span class="pre">cv.split</span></code> on the same
<a class="reference internal" href="modules/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> instance, but the splits will be
different each time. This is also true for any tool that performs model
selection via cross-validation, e.g.
<a class="reference internal" href="modules/generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">GridSearchCV</span></code></a> and
<a class="reference internal" href="modules/generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV" title="sklearn.model_selection.RandomizedSearchCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">RandomizedSearchCV</span></code></a>: scores are not
comparable fold-to-fold across different calls to <code class="docutils literal notranslate"><span class="pre">search.fit</span></code>, since
<code class="docutils literal notranslate"><span class="pre">cv.split</span></code> would have been called multiple times. Within a single call to
<code class="docutils literal notranslate"><span class="pre">search.fit</span></code>, however, fold-to-fold comparison is possible since the search
estimator only calls <code class="docutils literal notranslate"><span class="pre">cv.split</span></code> once.</p>
<p>For comparable fold-to-fold results in all scenarios, one should pass an
integer to the CV splitter: <code class="docutils literal notranslate"><span class="pre">cv</span> <span class="pre">=</span> <span class="pre">KFold(shuffle=True,</span> <span class="pre">random_state=0)</span></code>.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>While fold-to-fold comparison is not advisable with <code class="docutils literal notranslate"><span class="pre">RandomState</span></code>
instances, one can however expect that average scores allow to conclude
whether one estimator is better than another, as long as enough folds and
data are used.</p>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>What matters in this example is what was passed to
<a class="reference internal" href="modules/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>. Whether we pass a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code>
instance or an integer to <a class="reference internal" href="modules/generated/sklearn.datasets.make_classification.html#sklearn.datasets.make_classification" title="sklearn.datasets.make_classification"><code class="xref py py-func docutils literal notranslate"><span class="pre">make_classification</span></code></a>
is not relevant for our illustration purpose. Also, neither
<a class="reference internal" href="modules/generated/sklearn.discriminant_analysis.LinearDiscriminantAnalysis.html#sklearn.discriminant_analysis.LinearDiscriminantAnalysis" title="sklearn.discriminant_analysis.LinearDiscriminantAnalysis"><code class="xref py py-class docutils literal notranslate"><span class="pre">LinearDiscriminantAnalysis</span></code></a> nor
<a class="reference internal" href="modules/generated/sklearn.naive_bayes.GaussianNB.html#sklearn.naive_bayes.GaussianNB" title="sklearn.naive_bayes.GaussianNB"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianNB</span></code></a> are randomized estimators.</p>
</div>
</section>
</section>
<section id="general-recommendations">
<h3><span class="section-number">10.3.3. </span>General recommendations<a class="headerlink" href="#general-recommendations" title="Link to this heading">¶</a></h3>
<section id="getting-reproducible-results-across-multiple-executions">
<h4><span class="section-number">10.3.3.1. </span>Getting reproducible results across multiple executions<a class="headerlink" href="#getting-reproducible-results-across-multiple-executions" title="Link to this heading">¶</a></h4>
<p>In order to obtain reproducible (i.e. constant) results across multiple
<em>program executions</em>, we need to remove all uses of <code class="docutils literal notranslate"><span class="pre">random_state=None</span></code>, which
is the default. The recommended way is to declare a <code class="docutils literal notranslate"><span class="pre">rng</span></code> variable at the top
of the program, and pass it down to any object that accepts a <code class="docutils literal notranslate"><span class="pre">random_state</span></code>
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.ensemble</span> <span class="kn">import</span> <span class="n">RandomForestClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="kn">import</span> <span class="n">make_classification</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.model_selection</span> <span class="kn">import</span> <span class="n">train_test_split</span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="gp">>>> </span><span class="n">rng</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">random</span><span class="o">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">0</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="o">=</span> <span class="n">make_classification</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">rf</span> <span class="o">=</span> <span class="n">RandomForestClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X_train</span><span class="p">,</span> <span class="n">X_test</span><span class="p">,</span> <span class="n">y_train</span><span class="p">,</span> <span class="n">y_test</span> <span class="o">=</span> <span class="n">train_test_split</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">random_state</span><span class="o">=</span><span class="n">rng</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">rf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X_train</span><span class="p">,</span> <span class="n">y_train</span><span class="p">)</span><span class="o">.</span><span class="n">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">0.84</span>
</pre></div>
</div>
<p>We are now guaranteed that the result of this script will always be 0.84, no
matter how many times we run it. Changing the global <code class="docutils literal notranslate"><span class="pre">rng</span></code> variable to a
different value should affect the results, as expected.</p>
<p>It is also possible to declare the <code class="docutils literal notranslate"><span class="pre">rng</span></code> variable as an integer. This may
however lead to less robust cross-validation results, as we will see in the
next section.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>We do not recommend setting the global <code class="docutils literal notranslate"><span class="pre">numpy</span></code> seed by calling
<code class="docutils literal notranslate"><span class="pre">np.random.seed(0)</span></code>. See <a class="reference external" href="https://stackoverflow.com/questions/5836335/consistently-create-same-random-numpy-array/5837352#comment6712034_5837352">here</a>
for a discussion.</p>
</div>
</section>
<section id="robustness-of-cross-validation-results">
<h4><span class="section-number">10.3.3.2. </span>Robustness of cross-validation results<a class="headerlink" href="#robustness-of-cross-validation-results" title="Link to this heading">¶</a></h4>
<p>When we evaluate a randomized estimator performance by cross-validation, we
want to make sure that the estimator can yield accurate predictions for new
data, but we also want to make sure that the estimator is robust w.r.t. its
random initialization. For example, we would like the random weights
initialization of a <a class="reference internal" href="modules/generated/sklearn.linear_model.SGDClassifier.html#sklearn.linear_model.SGDClassifier" title="sklearn.linear_model.SGDClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">SGDClassifier</span></code></a> to be
consistently good across all folds: otherwise, when we train that estimator
on new data, we might get unlucky and the random initialization may lead to
bad performance. Similarly, we want a random forest to be robust w.r.t the
set of randomly selected features that each tree will be using.</p>
<p>For these reasons, it is preferable to evaluate the cross-validation
performance by letting the estimator use a different RNG on each fold. This
is done by passing a <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instance (or <code class="docutils literal notranslate"><span class="pre">None</span></code>) to the estimator
initialization.</p>
<p>When we pass an integer, the estimator will use the same RNG on each fold:
if the estimator performs well (or bad), as evaluated by CV, it might just be
because we got lucky (or unlucky) with that specific seed. Passing instances
leads to more robust CV results, and makes the comparison between various
algorithms fairer. It also helps limiting the temptation to treat the
estimator’s RNG as a hyper-parameter that can be tuned.</p>
<p>Whether we pass <code class="docutils literal notranslate"><span class="pre">RandomState</span></code> instances or integers to CV splitters has no
impact on robustness, as long as <code class="docutils literal notranslate"><span class="pre">split</span></code> is only called once. When <code class="docutils literal notranslate"><span class="pre">split</span></code>
is called multiple times, fold-to-fold comparison isn’t possible anymore. As
a result, passing integer to CV splitters is usually safer and covers most
use-cases.</p>
</section>
</section>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2024, scikit-learn developers (BSD License).
<a href="_sources/common_pitfalls.rst.txt" rel="nofollow">Show this page source</a>
</footer>
</div>
</div>
</div>
<script src="_static/js/vendor/bootstrap.min.js"></script>
<script>
window.ga=window.ga||function(){(ga.q=ga.q||[]).push(arguments)};ga.l=+new Date;
ga('create', 'UA-22606712-2', 'auto');
ga('set', 'anonymizeIp', true);
ga('send', 'pageview');
</script>
<script async src='https://www.google-analytics.com/analytics.js'></script>
<script defer data-domain="scikit-learn.org" src="https://views.scientific-python.org/js/script.js">
</script>
<script src="_static/clipboard.min.js"></script>
<script src="_static/copybutton.js"></script>
<script>
$(document).ready(function() {
/* Add a [>>>] button on the top-right corner of code samples to hide
* the >>> and ... prompts and the output and thus make the code
* copyable. */
var div = $('.highlight-python .highlight,' +
'.highlight-python3 .highlight,' +
'.highlight-pycon .highlight,' +
'.highlight-default .highlight')
var pre = div.find('pre');
// get the styles from the current theme
pre.parent().parent().css('position', 'relative');
// create and add the button to all the code blocks that contain >>>
div.each(function(index) {
var jthis = $(this);
// tracebacks (.gt) contain bare text elements that need to be
// wrapped in a span to work with .nextUntil() (see later)
jthis.find('pre:has(.gt)').contents().filter(function() {
return ((this.nodeType == 3) && (this.data.trim().length > 0));
}).wrap('<span>');
});
/*** Add permalink buttons next to glossary terms ***/
$('dl.glossary > dt[id]').append(function() {
return ('<a class="headerlink" href="#' +
this.getAttribute('id') +
'" title="Permalink to this term">¶</a>');
});
});
</script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml.js"></script>
<script src="https://scikit-learn.org/versionwarning.js"></script>
</body>
</html>