forked from scikit-learn/scikit-learn.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgrid_search.html
575 lines (541 loc) · 41.3 KB
/
grid_search.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
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>3.2. Tuning the hyper-parameters of an estimator — scikit-learn 0.20.dev0 documentation</title>
<!-- htmltitle is before nature.css - we use this hack to load bootstrap first -->
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<link rel="stylesheet" href="../_static/css/bootstrap.min.css" media="screen" />
<link rel="stylesheet" href="../_static/css/bootstrap-responsive.css"/>
<link rel="stylesheet" href="../_static/nature.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT: '../',
VERSION: '0.20.dev0',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true,
SOURCELINK_SUFFIX: '.txt'
};
</script>
<script type="text/javascript" src="../_static/jquery.js"></script>
<script type="text/javascript" src="../_static/underscore.js"></script>
<script type="text/javascript" src="../_static/doctools.js"></script>
<script type="text/javascript" src="../_static/js/copybutton.js"></script>
<script type="text/javascript" src="../_static/js/extra.js"></script>
<script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS_SVG"></script>
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<link rel="author" title="About these documents" href="../about.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="3.2.4.1.1. sklearn.linear_model.ElasticNetCV" href="generated/sklearn.linear_model.ElasticNetCV.html" />
<link rel="prev" title="3.1. Cross-validation: evaluating estimator performance" href="cross_validation.html" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<script src="../_static/js/bootstrap.min.js" type="text/javascript"></script>
<script>
VERSION_SUBDIR = (function(groups) {
return groups ? groups[1] : null;
})(location.href.match(/^https?:\/\/scikit-learn.org\/([^\/]+)/));
</script>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/grid_search.html" />
<script type="text/javascript">
$("div.buttonNext, div.buttonPrevious").hover(
function () {
$(this).css('background-color', '#FF9C34');
},
function () {
$(this).css('background-color', '#A7D6E2');
}
);
function showMenu() {
var topNav = document.getElementById("scikit-navbar");
if (topNav.className === "navbar") {
topNav.className += " responsive";
} else {
topNav.className = "navbar";
}
};
</script>
</head>
<body>
<div class="header-wrapper">
<div class="header">
<p class="logo"><a href="../index.html">
<img src="../_static/scikit-learn-logo-small.png" alt="Logo"/>
</a>
</p><div class="navbar" id="scikit-navbar">
<ul>
<li><a href="../index.html">Home</a></li>
<li><a href="../install.html">Installation</a></li>
<li class="btn-li"><div class="btn-group">
<a href="../documentation.html">Documentation</a>
<a class="btn dropdown-toggle" data-toggle="dropdown">
<span class="caret"></span>
</a>
<ul class="dropdown-menu">
<li class="link-title">Scikit-learn <script>document.write(DOCUMENTATION_OPTIONS.VERSION + (VERSION_SUBDIR ? " (" + VERSION_SUBDIR + ")" : ""));</script></li>
<li><a href="../tutorial/index.html">Tutorials</a></li>
<li><a href="../user_guide.html">User guide</a></li>
<li><a href="classes.html">API</a></li>
<li><a href="../glossary.html">Glossary</a></li>
<li><a href="../faq.html">FAQ</a></li>
<li><a href="../developers/contributing.html">Contributing</a></li>
<li class="divider"></li>
<script>if (VERSION_SUBDIR != "stable") document.write('<li><a href="http://scikit-learn.org/stable/documentation.html">Stable version</a></li>')</script>
<script>if (VERSION_SUBDIR != "dev") document.write('<li><a href="http://scikit-learn.org/dev/documentation.html">Development version</a></li>')</script>
<li><a href="http://scikit-learn.org/dev/versions.html">All available versions</a></li>
<li><a href="../_downloads/scikit-learn-docs.pdf">PDF documentation</a></li>
</ul>
</div>
</li>
<li><a href="../auto_examples/index.html">Examples</a></li>
</ul>
<a href="javascript:void(0);" onclick="showMenu()">
<div class="nav-icon">
<div class="hamburger-line"></div>
<div class="hamburger-line"></div>
<div class="hamburger-line"></div>
</div>
</a>
<div class="search_form">
<div class="gcse-search" id="cse" style="width: 100%;"></div>
</div>
</div> <!-- end navbar --></div>
</div>
<!-- GitHub "fork me" ribbon -->
<a href="https://github.com/scikit-learn/scikit-learn">
<img class="fork-me"
style="position: absolute; top: 0; right: 0; border: 0;"
src="../_static/img/forkme.png"
alt="Fork me on GitHub" />
</a>
<div class="content-wrapper">
<div class="sphinxsidebar">
<div class="sphinxsidebarwrapper">
<div class="rel">
<div class="rellink">
<a href="cross_validation.html"
accesskey="P">Previous
<br/>
<span class="smallrellink">
3.1. Cross-va...
</span>
<span class="hiddenrellink">
3.1. Cross-validation: evaluating estimator performance
</span>
</a>
</div>
<div class="spacer">
</div>
<div class="rellink">
<a href="generated/sklearn.linear_model.ElasticNetCV.html"
accesskey="N">Next
<br/>
<span class="smallrellink">
3.2.4.1.1. sk...
</span>
<span class="hiddenrellink">
3.2.4.1.1. sklearn.linear_model.ElasticNetCV
</span>
</a>
</div>
<!-- Ad a link to the 'up' page -->
<div class="spacer">
</div>
<div class="rellink">
<a href="../model_selection.html">
Up
<br/>
<span class="smallrellink">
3. Model sele...
</span>
<span class="hiddenrellink">
3. Model selection and evaluation
</span>
</a>
</div>
</div>
<p class="doc-version"><b>scikit-learn v0.20.dev0</b><br/>
<a href="http://scikit-learn.org/stable/support.html#documentation-resources">Other versions</a></p>
<p class="citing">Please <b><a href="../about.html#citing-scikit-learn" style="font-size: 110%;">cite us </a></b>if you use the software.</p>
<ul>
<li><a class="reference internal" href="#">3.2. Tuning the hyper-parameters of an estimator</a><ul>
<li><a class="reference internal" href="#exhaustive-grid-search">3.2.1. Exhaustive Grid Search</a></li>
<li><a class="reference internal" href="#randomized-parameter-optimization">3.2.2. Randomized Parameter Optimization</a></li>
<li><a class="reference internal" href="#tips-for-parameter-search">3.2.3. Tips for parameter search</a><ul>
<li><a class="reference internal" href="#specifying-an-objective-metric">3.2.3.1. Specifying an objective metric</a></li>
<li><a class="reference internal" href="#specifying-multiple-metrics-for-evaluation">3.2.3.2. Specifying multiple metrics for evaluation</a></li>
<li><a class="reference internal" href="#composite-estimators-and-parameter-spaces">3.2.3.3. Composite estimators and parameter spaces</a></li>
<li><a class="reference internal" href="#model-selection-development-and-evaluation">3.2.3.4. Model selection: development and evaluation</a></li>
<li><a class="reference internal" href="#parallelism">3.2.3.5. Parallelism</a></li>
<li><a class="reference internal" href="#robustness-to-failure">3.2.3.6. Robustness to failure</a></li>
</ul>
</li>
<li><a class="reference internal" href="#alternatives-to-brute-force-parameter-search">3.2.4. Alternatives to brute force parameter search</a><ul>
<li><a class="reference internal" href="#model-specific-cross-validation">3.2.4.1. Model specific cross-validation</a></li>
<li><a class="reference internal" href="#information-criterion">3.2.4.2. Information Criterion</a></li>
<li><a class="reference internal" href="#out-of-bag-estimates">3.2.4.3. Out of Bag Estimates</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
</div>
<input type="checkbox" id="nav-trigger" class="nav-trigger" checked />
<label for="nav-trigger"></label>
<div class="content">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body" role="main">
<div class="section" id="tuning-the-hyper-parameters-of-an-estimator">
<span id="grid-search"></span><h1>3.2. Tuning the hyper-parameters of an estimator<a class="headerlink" href="#tuning-the-hyper-parameters-of-an-estimator" title="Permalink to this headline">¶</a></h1>
<p>Hyper-parameters are parameters that are not directly learnt within estimators.
In scikit-learn they are passed as arguments to the constructor of the
estimator classes. Typical examples include <code class="docutils literal"><span class="pre">C</span></code>, <code class="docutils literal"><span class="pre">kernel</span></code> and <code class="docutils literal"><span class="pre">gamma</span></code>
for Support Vector Classifier, <code class="docutils literal"><span class="pre">alpha</span></code> for Lasso, etc.</p>
<p>It is possible and recommended to search the hyper-parameter space for the
best <a class="reference internal" href="cross_validation.html#cross-validation"><span class="std std-ref">cross validation</span></a> score.</p>
<p>Any parameter provided when constructing an estimator may be optimized in this
manner. Specifically, to find the names and current values for all parameters
for a given estimator, use:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">estimator</span><span class="o">.</span><span class="n">get_params</span><span class="p">()</span>
</pre></div>
</div>
<p>A search consists of:</p>
<ul class="simple">
<li>an estimator (regressor or classifier such as <code class="docutils literal"><span class="pre">sklearn.svm.SVC()</span></code>);</li>
<li>a parameter space;</li>
<li>a method for searching or sampling candidates;</li>
<li>a cross-validation scheme; and</li>
<li>a <a class="reference internal" href="#gridsearch-scoring"><span class="std std-ref">score function</span></a>.</li>
</ul>
<p>Some models allow for specialized, efficient parameter search strategies,
<a class="reference internal" href="#alternative-cv"><span class="std std-ref">outlined below</span></a>.
Two generic approaches to sampling search candidates are provided in
scikit-learn: for given values, <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a> exhaustively considers
all parameter combinations, while <a class="reference internal" href="generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV" title="sklearn.model_selection.RandomizedSearchCV"><code class="xref py py-class docutils literal"><span class="pre">RandomizedSearchCV</span></code></a> can sample a
given number of candidates from a parameter space with a specified
distribution. After describing these tools we detail
<a class="reference internal" href="#grid-search-tips"><span class="std std-ref">best practice</span></a> applicable to both approaches.</p>
<p>Note that it is common that a small subset of those parameters can have a large
impact on the predictive or computation performance of the model while others
can be left to their default values. It is recommended to read the docstring of
the estimator class to get a finer understanding of their expected behavior,
possibly by reading the enclosed reference to the literature.</p>
<div class="section" id="exhaustive-grid-search">
<h2>3.2.1. Exhaustive Grid Search<a class="headerlink" href="#exhaustive-grid-search" title="Permalink to this headline">¶</a></h2>
<p>The grid search provided by <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a> exhaustively generates
candidates from a grid of parameter values specified with the <code class="docutils literal"><span class="pre">param_grid</span></code>
parameter. For instance, the following <code class="docutils literal"><span class="pre">param_grid</span></code>:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="n">param_grid</span> <span class="o">=</span> <span class="p">[</span>
<span class="p">{</span><span class="s1">'C'</span><span class="p">:</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">10</span><span class="p">,</span> <span class="mi">100</span><span class="p">,</span> <span class="mi">1000</span><span class="p">],</span> <span class="s1">'kernel'</span><span class="p">:</span> <span class="p">[</span><span class="s1">'linear'</span><span class="p">]},</span>
<span class="p">{</span><span class="s1">'C'</span><span class="p">:</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">10</span><span class="p">,</span> <span class="mi">100</span><span class="p">,</span> <span class="mi">1000</span><span class="p">],</span> <span class="s1">'gamma'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.001</span><span class="p">,</span> <span class="mf">0.0001</span><span class="p">],</span> <span class="s1">'kernel'</span><span class="p">:</span> <span class="p">[</span><span class="s1">'rbf'</span><span class="p">]},</span>
<span class="p">]</span>
</pre></div>
</div>
<p>specifies that two grids should be explored: one with a linear kernel and
C values in [1, 10, 100, 1000], and the second one with an RBF kernel,
and the cross-product of C values ranging in [1, 10, 100, 1000] and gamma
values in [0.001, 0.0001].</p>
<p>The <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a> instance implements the usual estimator API: when
“fitting” it on a dataset all the possible combinations of parameter values are
evaluated and the best combination is retained.</p>
<div class="topic">
<p class="topic-title first">Examples:</p>
<ul class="simple">
<li>See <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">Parameter estimation using grid search with cross-validation</span></a> for an example of
Grid Search computation on the digits dataset.</li>
<li>See <a class="reference internal" href="../auto_examples/model_selection/grid_search_text_feature_extraction.html#sphx-glr-auto-examples-model-selection-grid-search-text-feature-extraction-py"><span class="std std-ref">Sample pipeline for text feature extraction and evaluation</span></a> for an example
of Grid Search coupling parameters from a text documents feature
extractor (n-gram count vectorizer and TF-IDF transformer) with a
classifier (here a linear SVM trained with SGD with either elastic
net or L2 penalty) using a <code class="xref py py-class docutils literal"><span class="pre">pipeline.Pipeline</span></code> instance.</li>
<li>See <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>
for an example of Grid Search within a cross validation loop on the iris
dataset. This is the best practice for evaluating the performance of a
model with grid search.</li>
<li>See <a class="reference internal" href="../auto_examples/model_selection/plot_multi_metric_evaluation.html#sphx-glr-auto-examples-model-selection-plot-multi-metric-evaluation-py"><span class="std std-ref">Demonstration of multi-metric evaluation on cross_val_score and GridSearchCV</span></a>
for an example of <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a> being used to evaluate multiple
metrics simultaneously.</li>
</ul>
</div>
</div>
<div class="section" id="randomized-parameter-optimization">
<span id="randomized-parameter-search"></span><h2>3.2.2. Randomized Parameter Optimization<a class="headerlink" href="#randomized-parameter-optimization" title="Permalink to this headline">¶</a></h2>
<p>While using a grid of parameter settings is currently the most widely used
method for parameter optimization, other search methods have more
favourable properties.
<a class="reference internal" href="generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV" title="sklearn.model_selection.RandomizedSearchCV"><code class="xref py py-class docutils literal"><span class="pre">RandomizedSearchCV</span></code></a> implements a randomized search over parameters,
where each setting is sampled from a distribution over possible parameter values.
This has two main benefits over an exhaustive search:</p>
<ul class="simple">
<li>A budget can be chosen independent of the number of parameters and possible values.</li>
<li>Adding parameters that do not influence the performance does not decrease efficiency.</li>
</ul>
<p>Specifying how parameters should be sampled is done using a dictionary, very
similar to specifying parameters for <a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a>. Additionally,
a computation budget, being the number of sampled candidates or sampling
iterations, is specified using the <code class="docutils literal"><span class="pre">n_iter</span></code> parameter.
For each parameter, either a distribution over possible values or a list of
discrete choices (which will be sampled uniformly) can be specified:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="p">{</span><span class="s1">'C'</span><span class="p">:</span> <span class="n">scipy</span><span class="o">.</span><span class="n">stats</span><span class="o">.</span><span class="n">expon</span><span class="p">(</span><span class="n">scale</span><span class="o">=</span><span class="mi">100</span><span class="p">),</span> <span class="s1">'gamma'</span><span class="p">:</span> <span class="n">scipy</span><span class="o">.</span><span class="n">stats</span><span class="o">.</span><span class="n">expon</span><span class="p">(</span><span class="n">scale</span><span class="o">=.</span><span class="mi">1</span><span class="p">),</span>
<span class="s1">'kernel'</span><span class="p">:</span> <span class="p">[</span><span class="s1">'rbf'</span><span class="p">],</span> <span class="s1">'class_weight'</span><span class="p">:[</span><span class="s1">'balanced'</span><span class="p">,</span> <span class="kc">None</span><span class="p">]}</span>
</pre></div>
</div>
<p>This example uses the <code class="docutils literal"><span class="pre">scipy.stats</span></code> module, which contains many useful
distributions for sampling parameters, such as <code class="docutils literal"><span class="pre">expon</span></code>, <code class="docutils literal"><span class="pre">gamma</span></code>,
<code class="docutils literal"><span class="pre">uniform</span></code> or <code class="docutils literal"><span class="pre">randint</span></code>.
In principle, any function can be passed that provides a <code class="docutils literal"><span class="pre">rvs</span></code> (random
variate sample) method to sample a value. A call to the <code class="docutils literal"><span class="pre">rvs</span></code> function should
provide independent random samples from possible parameter values on
consecutive calls.</p>
<blockquote>
<div><div class="admonition warning">
<p class="first admonition-title">Warning</p>
<p class="last">The distributions in <code class="docutils literal"><span class="pre">scipy.stats</span></code> prior to version scipy 0.16
do not allow specifying a random state. Instead, they use the global
numpy random state, that can be seeded via <code class="docutils literal"><span class="pre">np.random.seed</span></code> or set
using <code class="docutils literal"><span class="pre">np.random.set_state</span></code>. However, beginning scikit-learn 0.18,
the <a class="reference internal" href="classes.html#module-sklearn.model_selection" title="sklearn.model_selection"><code class="xref py py-mod docutils literal"><span class="pre">sklearn.model_selection</span></code></a> module sets the random state provided
by the user if scipy >= 0.16 is also available.</p>
</div>
</div></blockquote>
<p>For continuous parameters, such as <code class="docutils literal"><span class="pre">C</span></code> above, it is important to specify
a continuous distribution to take full advantage of the randomization. This way,
increasing <code class="docutils literal"><span class="pre">n_iter</span></code> will always lead to a finer search.</p>
<div class="topic">
<p class="topic-title first">Examples:</p>
<ul class="simple">
<li><a class="reference internal" href="../auto_examples/model_selection/plot_randomized_search.html#sphx-glr-auto-examples-model-selection-plot-randomized-search-py"><span class="std std-ref">Comparing randomized search and grid search for hyperparameter estimation</span></a> compares the usage and efficiency
of randomized search and grid search.</li>
</ul>
</div>
<div class="topic">
<p class="topic-title first">References:</p>
<ul class="simple">
<li>Bergstra, J. and Bengio, Y.,
Random search for hyper-parameter optimization,
The Journal of Machine Learning Research (2012)</li>
</ul>
</div>
</div>
<div class="section" id="tips-for-parameter-search">
<span id="grid-search-tips"></span><h2>3.2.3. Tips for parameter search<a class="headerlink" href="#tips-for-parameter-search" title="Permalink to this headline">¶</a></h2>
<div class="section" id="specifying-an-objective-metric">
<span id="gridsearch-scoring"></span><h3>3.2.3.1. Specifying an objective metric<a class="headerlink" href="#specifying-an-objective-metric" title="Permalink to this headline">¶</a></h3>
<p>By default, parameter search uses the <code class="docutils literal"><span class="pre">score</span></code> function of the estimator
to evaluate a parameter setting. These are the
<a class="reference internal" href="generated/sklearn.metrics.accuracy_score.html#sklearn.metrics.accuracy_score" title="sklearn.metrics.accuracy_score"><code class="xref py py-func docutils literal"><span class="pre">sklearn.metrics.accuracy_score</span></code></a> for classification and
<a class="reference internal" href="generated/sklearn.metrics.r2_score.html#sklearn.metrics.r2_score" title="sklearn.metrics.r2_score"><code class="xref py py-func docutils literal"><span class="pre">sklearn.metrics.r2_score</span></code></a> for regression. For some applications,
other scoring functions are better suited (for example in unbalanced
classification, the accuracy score is often uninformative). An alternative
scoring function can be specified via the <code class="docutils literal"><span class="pre">scoring</span></code> parameter to
<a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a>, <a class="reference internal" href="generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV" title="sklearn.model_selection.RandomizedSearchCV"><code class="xref py py-class docutils literal"><span class="pre">RandomizedSearchCV</span></code></a> and many of the
specialized cross-validation tools described below.
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 more details.</p>
</div>
<div class="section" id="specifying-multiple-metrics-for-evaluation">
<span id="multimetric-grid-search"></span><h3>3.2.3.2. Specifying multiple metrics for evaluation<a class="headerlink" href="#specifying-multiple-metrics-for-evaluation" title="Permalink to this headline">¶</a></h3>
<p><code class="docutils literal"><span class="pre">GridSearchCV</span></code> and <code class="docutils literal"><span class="pre">RandomizedSearchCV</span></code> allow specifying multiple metrics
for the <code class="docutils literal"><span class="pre">scoring</span></code> parameter.</p>
<p>Multimetric scoring can either be specified as a list of strings of predefined
scores names or a dict mapping the scorer name to the scorer function and/or
the predefined scorer name(s). See <a class="reference internal" href="model_evaluation.html#multimetric-scoring"><span class="std std-ref">Using multiple metric evaluation</span></a> for more details.</p>
<p>When specifying multiple metrics, the <code class="docutils literal"><span class="pre">refit</span></code> parameter must be set to the
metric (string) for which the <code class="docutils literal"><span class="pre">best_params_</span></code> will be found and used to build
the <code class="docutils literal"><span class="pre">best_estimator_</span></code> on the whole dataset. If the search should not be
refit, set <code class="docutils literal"><span class="pre">refit=False</span></code>. Leaving refit to the default value <code class="docutils literal"><span class="pre">None</span></code> will
result in an error when using multiple metrics.</p>
<p>See <a class="reference internal" href="../auto_examples/model_selection/plot_multi_metric_evaluation.html#sphx-glr-auto-examples-model-selection-plot-multi-metric-evaluation-py"><span class="std std-ref">Demonstration of multi-metric evaluation on cross_val_score and GridSearchCV</span></a>
for an example usage.</p>
</div>
<div class="section" id="composite-estimators-and-parameter-spaces">
<h3>3.2.3.3. Composite estimators and parameter spaces<a class="headerlink" href="#composite-estimators-and-parameter-spaces" title="Permalink to this headline">¶</a></h3>
<p><a class="reference internal" href="compose.html#pipeline"><span class="std std-ref">Pipeline: chaining estimators</span></a> describes building composite estimators whose
parameter space can be searched with these tools.</p>
</div>
<div class="section" id="model-selection-development-and-evaluation">
<h3>3.2.3.4. Model selection: development and evaluation<a class="headerlink" href="#model-selection-development-and-evaluation" title="Permalink to this headline">¶</a></h3>
<p>Model selection by evaluating various parameter settings can be seen as a way
to use the labeled data to “train” the parameters of the grid.</p>
<p>When evaluating the resulting model it is important to do it on
held-out samples that were not seen during the grid search process:
it is recommended to split the data into a <strong>development set</strong> (to
be fed to the <code class="docutils literal"><span class="pre">GridSearchCV</span></code> instance) and an <strong>evaluation set</strong>
to compute performance metrics.</p>
<p>This can be done by using 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"><span class="pre">train_test_split</span></code></a>
utility function.</p>
</div>
<div class="section" id="parallelism">
<h3>3.2.3.5. Parallelism<a class="headerlink" href="#parallelism" title="Permalink to this headline">¶</a></h3>
<p><a class="reference internal" href="generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV" title="sklearn.model_selection.GridSearchCV"><code class="xref py py-class docutils literal"><span class="pre">GridSearchCV</span></code></a> and <a class="reference internal" href="generated/sklearn.model_selection.RandomizedSearchCV.html#sklearn.model_selection.RandomizedSearchCV" title="sklearn.model_selection.RandomizedSearchCV"><code class="xref py py-class docutils literal"><span class="pre">RandomizedSearchCV</span></code></a> evaluate each parameter
setting independently. Computations can be run in parallel if your OS
supports it, by using the keyword <code class="docutils literal"><span class="pre">n_jobs=-1</span></code>. See function signature for
more details.</p>
</div>
<div class="section" id="robustness-to-failure">
<h3>3.2.3.6. Robustness to failure<a class="headerlink" href="#robustness-to-failure" title="Permalink to this headline">¶</a></h3>
<p>Some parameter settings may result in a failure to <code class="docutils literal"><span class="pre">fit</span></code> one or more folds
of the data. By default, this will cause the entire search to fail, even if
some parameter settings could be fully evaluated. Setting <code class="docutils literal"><span class="pre">error_score=0</span></code>
(or <cite>=np.NaN</cite>) will make the procedure robust to such failure, issuing a
warning and setting the score for that fold to 0 (or <cite>NaN</cite>), but completing
the search.</p>
</div>
</div>
<div class="section" id="alternatives-to-brute-force-parameter-search">
<span id="alternative-cv"></span><h2>3.2.4. Alternatives to brute force parameter search<a class="headerlink" href="#alternatives-to-brute-force-parameter-search" title="Permalink to this headline">¶</a></h2>
<div class="section" id="model-specific-cross-validation">
<h3>3.2.4.1. Model specific cross-validation<a class="headerlink" href="#model-specific-cross-validation" title="Permalink to this headline">¶</a></h3>
<p>Some models can fit data for a range of values of some parameter almost
as efficiently as fitting the estimator for a single value of the
parameter. This feature can be leveraged to perform a more efficient
cross-validation used for model selection of this parameter.</p>
<p>The most common parameter amenable to this strategy is the parameter
encoding the strength of the regularizer. In this case we say that we
compute the <strong>regularization path</strong> of the estimator.</p>
<p>Here is the list of such models:</p>
<table border="1" class="longtable docutils">
<colgroup>
<col width="10%" />
<col width="90%" />
</colgroup>
<tbody valign="top">
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.ElasticNetCV.html#sklearn.linear_model.ElasticNetCV" title="sklearn.linear_model.ElasticNetCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.ElasticNetCV</span></code></a>([l1_ratio, eps, …])</td>
<td>Elastic Net model with iterative fitting along a regularization path</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.linear_model.LarsCV.html#sklearn.linear_model.LarsCV" title="sklearn.linear_model.LarsCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.LarsCV</span></code></a>([fit_intercept, …])</td>
<td>Cross-validated Least Angle Regression model</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.LassoCV.html#sklearn.linear_model.LassoCV" title="sklearn.linear_model.LassoCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.LassoCV</span></code></a>([eps, n_alphas, …])</td>
<td>Lasso linear model with iterative fitting along a regularization path</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.linear_model.LassoLarsCV.html#sklearn.linear_model.LassoLarsCV" title="sklearn.linear_model.LassoLarsCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.LassoLarsCV</span></code></a>([fit_intercept, …])</td>
<td>Cross-validated Lasso, using the LARS algorithm</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.LogisticRegressionCV.html#sklearn.linear_model.LogisticRegressionCV" title="sklearn.linear_model.LogisticRegressionCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.LogisticRegressionCV</span></code></a>([Cs, …])</td>
<td>Logistic Regression CV (aka logit, MaxEnt) classifier.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.linear_model.MultiTaskElasticNetCV.html#sklearn.linear_model.MultiTaskElasticNetCV" title="sklearn.linear_model.MultiTaskElasticNetCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.MultiTaskElasticNetCV</span></code></a>([…])</td>
<td>Multi-task L1/L2 ElasticNet with built-in cross-validation.</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.MultiTaskLassoCV.html#sklearn.linear_model.MultiTaskLassoCV" title="sklearn.linear_model.MultiTaskLassoCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.MultiTaskLassoCV</span></code></a>([eps, …])</td>
<td>Multi-task L1/L2 Lasso with built-in cross-validation.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.linear_model.OrthogonalMatchingPursuitCV.html#sklearn.linear_model.OrthogonalMatchingPursuitCV" title="sklearn.linear_model.OrthogonalMatchingPursuitCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.OrthogonalMatchingPursuitCV</span></code></a>([…])</td>
<td>Cross-validated Orthogonal Matching Pursuit model (OMP)</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.RidgeCV.html#sklearn.linear_model.RidgeCV" title="sklearn.linear_model.RidgeCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.RidgeCV</span></code></a>([alphas, …])</td>
<td>Ridge regression with built-in cross-validation.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.linear_model.RidgeClassifierCV.html#sklearn.linear_model.RidgeClassifierCV" title="sklearn.linear_model.RidgeClassifierCV"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.RidgeClassifierCV</span></code></a>([alphas, …])</td>
<td>Ridge classifier with built-in cross-validation.</td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="information-criterion">
<h3>3.2.4.2. Information Criterion<a class="headerlink" href="#information-criterion" title="Permalink to this headline">¶</a></h3>
<p>Some models can offer an information-theoretic closed-form formula of the
optimal estimate of the regularization parameter by computing a single
regularization path (instead of several when using cross-validation).</p>
<p>Here is the list of models benefiting from the Akaike Information
Criterion (AIC) or the Bayesian Information Criterion (BIC) for automated
model selection:</p>
<table border="1" class="longtable docutils">
<colgroup>
<col width="10%" />
<col width="90%" />
</colgroup>
<tbody valign="top">
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.linear_model.LassoLarsIC.html#sklearn.linear_model.LassoLarsIC" title="sklearn.linear_model.LassoLarsIC"><code class="xref py py-obj docutils literal"><span class="pre">linear_model.LassoLarsIC</span></code></a>([criterion, …])</td>
<td>Lasso model fit with Lars using BIC or AIC for model selection</td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="out-of-bag-estimates">
<span id="out-of-bag"></span><h3>3.2.4.3. Out of Bag Estimates<a class="headerlink" href="#out-of-bag-estimates" title="Permalink to this headline">¶</a></h3>
<p>When using ensemble methods base upon bagging, i.e. generating new
training sets using sampling with replacement, part of the training set
remains unused. For each classifier in the ensemble, a different part
of the training set is left out.</p>
<p>This left out portion can be used to estimate the generalization error
without having to rely on a separate validation set. This estimate
comes “for free” as no additional data is needed and can be used for
model selection.</p>
<p>This is currently implemented in the following classes:</p>
<table border="1" class="longtable docutils">
<colgroup>
<col width="10%" />
<col width="90%" />
</colgroup>
<tbody valign="top">
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier" title="sklearn.ensemble.RandomForestClassifier"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.RandomForestClassifier</span></code></a>([…])</td>
<td>A random forest classifier.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.ensemble.RandomForestRegressor.html#sklearn.ensemble.RandomForestRegressor" title="sklearn.ensemble.RandomForestRegressor"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.RandomForestRegressor</span></code></a>([…])</td>
<td>A random forest regressor.</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.ensemble.ExtraTreesClassifier.html#sklearn.ensemble.ExtraTreesClassifier" title="sklearn.ensemble.ExtraTreesClassifier"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.ExtraTreesClassifier</span></code></a>([…])</td>
<td>An extra-trees classifier.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.ensemble.ExtraTreesRegressor.html#sklearn.ensemble.ExtraTreesRegressor" title="sklearn.ensemble.ExtraTreesRegressor"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.ExtraTreesRegressor</span></code></a>([n_estimators, …])</td>
<td>An extra-trees regressor.</td>
</tr>
<tr class="row-odd"><td><a class="reference internal" href="generated/sklearn.ensemble.GradientBoostingClassifier.html#sklearn.ensemble.GradientBoostingClassifier" title="sklearn.ensemble.GradientBoostingClassifier"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.GradientBoostingClassifier</span></code></a>([loss, …])</td>
<td>Gradient Boosting for classification.</td>
</tr>
<tr class="row-even"><td><a class="reference internal" href="generated/sklearn.ensemble.GradientBoostingRegressor.html#sklearn.ensemble.GradientBoostingRegressor" title="sklearn.ensemble.GradientBoostingRegressor"><code class="xref py py-obj docutils literal"><span class="pre">ensemble.GradientBoostingRegressor</span></code></a>([loss, …])</td>
<td>Gradient Boosting for regression.</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
</div>
<div class="footer">
© 2007 - 2018, scikit-learn developers (BSD License).
<a href="../_sources/modules/grid_search.rst.txt" rel="nofollow">Show this page source</a>
</div>
<div class="rel">
<div class="buttonPrevious">
<a href="cross_validation.html">Previous
</a>
</div>
<div class="buttonNext">
<a href="generated/sklearn.linear_model.ElasticNetCV.html">Next
</a>
</div>
</div>
<script type="text/javascript">
var _gaq = _gaq || [];
_gaq.push(['_setAccount', 'UA-22606712-2']);
_gaq.push(['_trackPageview']);
(function() {
var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true;
ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s);
})();
</script>
<script>
(function() {
var cx = '016639176250731907682:tjtqbvtvij0';
var gcse = document.createElement('script'); gcse.type = 'text/javascript'; gcse.async = true;
gcse.src = 'https://cse.google.com/cse.js?cx=' + cx;
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(gcse, s);
})();
</script>
</body>
</html>