-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathcovariance.html
596 lines (564 loc) · 41.8 KB
/
covariance.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
<!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="2.6. Covariance estimation" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/modules/covariance.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="Many statistical problems require the estimation of a population’s covariance matrix, which can be seen as an estimation of data set scatter plot shape. Most of the time, such an estimation has to ..." />
<meta property="og:image" content="https://scikit-learn/stable/_images/sphx_glr_plot_covariance_estimation_001.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="Many statistical problems require the estimation of a population’s covariance matrix, which can be seen as an estimation of data set scatter plot shape. Most of the time, such an estimation has to ..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>2.6. Covariance estimation — scikit-learn 1.5.dev0 documentation</title>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/covariance.html" />
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<link rel="stylesheet" href="../_static/css/vendor/bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="../_static/plot_directive.css" type="text/css" />
<link rel="stylesheet" href="../https://fonts.googleapis.com/css?family=Vibur" type="text/css" />
<link rel="stylesheet" href="../_static/jupyterlite_sphinx.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<script id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
<script src="../_static/js/vendor/jquery-3.6.3.slim.min.js"></script>
<script src="../_static/js/details-permalink.js"></script>
</head>
<body>
<nav id="navbar" class="sk-docs-navbar navbar navbar-expand-md navbar-light bg-light py-0">
<div class="container-fluid sk-docs-container px-0">
<a class="navbar-brand py-0" href="../index.html">
<img
class="sk-brand-img"
src="../_static/scikit-learn-logo-small.png"
alt="logo"/>
</a>
<button
id="sk-navbar-toggler"
class="navbar-toggler"
type="button"
data-toggle="collapse"
data-target="#navbarSupportedContent"
aria-controls="navbarSupportedContent"
aria-expanded="false"
aria-label="Toggle navigation"
>
<span class="navbar-toggler-icon"></span>
</button>
<div class="sk-navbar-collapse collapse navbar-collapse" id="navbarSupportedContent">
<ul class="navbar-nav mr-auto">
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../install.html">Install</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../user_guide.html">User Guide</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="classes.html">API</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="../auto_examples/index.html">Examples</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" target="_blank" rel="noopener noreferrer" href="https://blog.scikit-learn.org/">Community</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../getting_started.html" >Getting Started</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../tutorial/index.html" >Tutorial</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../whats_new/v1.5.html" >What's new</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../glossary.html" >Glossary</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../developers/index.html" >Development</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../faq.html" >FAQ</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../support.html" >Support</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../related_projects.html" >Related packages</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../roadmap.html" >Roadmap</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../governance.html" >Governance</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="../about.html" >About us</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</li>
<li class="nav-item dropdown nav-more-item-dropdown">
<a class="sk-nav-link nav-link dropdown-toggle" href="#" id="navbarDropdown" role="button" data-toggle="dropdown" aria-haspopup="true" aria-expanded="false">More</a>
<div class="dropdown-menu" aria-labelledby="navbarDropdown">
<a class="sk-nav-dropdown-item dropdown-item" href="../getting_started.html" >Getting Started</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../tutorial/index.html" >Tutorial</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../whats_new/v1.5.html" >What's new</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../glossary.html" >Glossary</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../developers/index.html" >Development</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../faq.html" >FAQ</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../support.html" >Support</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../related_projects.html" >Related packages</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../roadmap.html" >Roadmap</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../governance.html" >Governance</a>
<a class="sk-nav-dropdown-item dropdown-item" href="../about.html" >About us</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://github.com/scikit-learn/scikit-learn" >GitHub</a>
<a class="sk-nav-dropdown-item dropdown-item" href="https://scikit-learn.org/dev/versions.html" >Other Versions and Download</a>
</div>
</li>
</ul>
<div id="searchbox" role="search">
<div class="searchformwrapper">
<form class="search" action="../search.html" method="get">
<input class="sk-search-text-input" type="text" name="q" aria-labelledby="searchlabel" />
<input class="sk-search-text-btn" type="submit" value="Go" />
</form>
</div>
</div>
</div>
</div>
</nav>
<div class="d-flex" id="sk-doc-wrapper">
<input type="checkbox" name="sk-toggle-checkbox" id="sk-toggle-checkbox">
<label id="sk-sidemenu-toggle" class="sk-btn-toggle-toc btn sk-btn-primary" for="sk-toggle-checkbox">Toggle Menu</label>
<div id="sk-sidebar-wrapper" class="border-right">
<div class="sk-sidebar-toc-wrapper">
<div class="btn-group w-100 mb-2" role="group" aria-label="rellinks">
<a href="decomposition.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2.5. Decomposing signals in components (matrix factorization problems)">Prev</a><a href="../unsupervised_learning.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2. Unsupervised learning">Up</a>
<a href="outlier_detection.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2.7. Novelty and Outlier Detection">Next</a>
</div>
<div class="alert alert-danger p-1 mb-2" role="alert">
<p class="text-center mb-0">
<strong>scikit-learn 1.5.dev0</strong><br/>
<a href="http://scikit-learn.org/dev/versions.html">Other versions</a>
</p>
</div>
<div class="alert alert-warning p-1 mb-2" role="alert">
<p class="text-center mb-0">
Please <a class="font-weight-bold" href="../about.html#citing-scikit-learn"><string>cite us</string></a> if you use the software.
</p>
</div>
<div class="sk-sidebar-toc">
<ul>
<li><a class="reference internal" href="#">2.6. Covariance estimation</a><ul>
<li><a class="reference internal" href="#empirical-covariance">2.6.1. Empirical covariance</a></li>
<li><a class="reference internal" href="#shrunk-covariance">2.6.2. Shrunk Covariance</a><ul>
<li><a class="reference internal" href="#basic-shrinkage">2.6.2.1. Basic shrinkage</a></li>
<li><a class="reference internal" href="#ledoit-wolf-shrinkage">2.6.2.2. Ledoit-Wolf shrinkage</a></li>
<li><a class="reference internal" href="#oracle-approximating-shrinkage">2.6.2.3. Oracle Approximating Shrinkage</a></li>
</ul>
</li>
<li><a class="reference internal" href="#sparse-inverse-covariance">2.6.3. Sparse inverse covariance</a></li>
<li><a class="reference internal" href="#robust-covariance-estimation">2.6.4. Robust Covariance Estimation</a><ul>
<li><a class="reference internal" href="#minimum-covariance-determinant">2.6.4.1. Minimum Covariance Determinant</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
</div>
</div>
<div id="sk-page-content-wrapper">
<div class="sk-page-content container-fluid body px-md-3" role="main">
<section id="covariance-estimation">
<span id="covariance"></span><h1><span class="section-number">2.6. </span>Covariance estimation<a class="headerlink" href="#covariance-estimation" title="Link to this heading">¶</a></h1>
<p>Many statistical problems require the estimation of a
population’s covariance matrix, which can be seen as an estimation of
data set scatter plot shape. Most of the time, such an estimation has
to be done on a sample whose properties (size, structure, homogeneity)
have a large influence on the estimation’s quality. The
<a class="reference internal" href="classes.html#module-sklearn.covariance" title="sklearn.covariance"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.covariance</span></code></a> package provides tools for accurately estimating
a population’s covariance matrix under various settings.</p>
<p>We assume that the observations are independent and identically
distributed (i.i.d.).</p>
<section id="empirical-covariance">
<h2><span class="section-number">2.6.1. </span>Empirical covariance<a class="headerlink" href="#empirical-covariance" title="Link to this heading">¶</a></h2>
<p>The covariance matrix of a data set is known to be well approximated
by the classical <em>maximum likelihood estimator</em> (or “empirical
covariance”), provided the number of observations is large enough
compared to the number of features (the variables describing the
observations). More precisely, the Maximum Likelihood Estimator of a
sample is an asymptotically unbiased estimator of the corresponding
population’s covariance matrix.</p>
<p>The empirical covariance matrix of a sample can be computed using the
<a class="reference internal" href="generated/sklearn.covariance.empirical_covariance.html#sklearn.covariance.empirical_covariance" title="sklearn.covariance.empirical_covariance"><code class="xref py py-func docutils literal notranslate"><span class="pre">empirical_covariance</span></code></a> function of the package, or by fitting an
<a class="reference internal" href="generated/sklearn.covariance.EmpiricalCovariance.html#sklearn.covariance.EmpiricalCovariance" title="sklearn.covariance.EmpiricalCovariance"><code class="xref py py-class docutils literal notranslate"><span class="pre">EmpiricalCovariance</span></code></a> object to the data sample with the
<a class="reference internal" href="generated/sklearn.covariance.EmpiricalCovariance.html#sklearn.covariance.EmpiricalCovariance.fit" title="sklearn.covariance.EmpiricalCovariance.fit"><code class="xref py py-meth docutils literal notranslate"><span class="pre">EmpiricalCovariance.fit</span></code></a> method. Be careful that results depend
on whether the data are centered, so one may want to use the
<code class="docutils literal notranslate"><span class="pre">assume_centered</span></code> parameter accurately. More precisely, if
<code class="docutils literal notranslate"><span class="pre">assume_centered=False</span></code>, then the test set is supposed to have the
same mean vector as the training set. If not, both should be centered
by the user, and <code class="docutils literal notranslate"><span class="pre">assume_centered=True</span></code> should be used.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_covariance_estimation.html#sphx-glr-auto-examples-covariance-plot-covariance-estimation-py"><span class="std std-ref">Shrinkage covariance estimation: LedoitWolf vs OAS and max-likelihood</span></a> for
an example on how to fit an <a class="reference internal" href="generated/sklearn.covariance.EmpiricalCovariance.html#sklearn.covariance.EmpiricalCovariance" title="sklearn.covariance.EmpiricalCovariance"><code class="xref py py-class docutils literal notranslate"><span class="pre">EmpiricalCovariance</span></code></a> object
to data.</p></li>
</ul>
</aside>
</section>
<section id="shrunk-covariance">
<span id="id1"></span><h2><span class="section-number">2.6.2. </span>Shrunk Covariance<a class="headerlink" href="#shrunk-covariance" title="Link to this heading">¶</a></h2>
<section id="basic-shrinkage">
<h3><span class="section-number">2.6.2.1. </span>Basic shrinkage<a class="headerlink" href="#basic-shrinkage" title="Link to this heading">¶</a></h3>
<p>Despite being an asymptotically unbiased estimator of the covariance matrix,
the Maximum Likelihood Estimator is not a good estimator of the
eigenvalues of the covariance matrix, so the precision matrix obtained
from its inversion is not accurate. Sometimes, it even occurs that the
empirical covariance matrix cannot be inverted for numerical
reasons. To avoid such an inversion problem, a transformation of the
empirical covariance matrix has been introduced: the <code class="docutils literal notranslate"><span class="pre">shrinkage</span></code>.</p>
<p>In scikit-learn, this transformation (with a user-defined shrinkage
coefficient) can be directly applied to a pre-computed covariance with
the <a class="reference internal" href="generated/sklearn.covariance.shrunk_covariance.html#sklearn.covariance.shrunk_covariance" title="sklearn.covariance.shrunk_covariance"><code class="xref py py-func docutils literal notranslate"><span class="pre">shrunk_covariance</span></code></a> method. Also, a shrunk estimator of the
covariance can be fitted to data with a <a class="reference internal" href="generated/sklearn.covariance.ShrunkCovariance.html#sklearn.covariance.ShrunkCovariance" title="sklearn.covariance.ShrunkCovariance"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShrunkCovariance</span></code></a> object
and its <a class="reference internal" href="generated/sklearn.covariance.ShrunkCovariance.html#sklearn.covariance.ShrunkCovariance.fit" title="sklearn.covariance.ShrunkCovariance.fit"><code class="xref py py-meth docutils literal notranslate"><span class="pre">ShrunkCovariance.fit</span></code></a> method. Again, results depend on
whether the data are centered, so one may want to use the
<code class="docutils literal notranslate"><span class="pre">assume_centered</span></code> parameter accurately.</p>
<p>Mathematically, this shrinkage consists in reducing the ratio between the
smallest and the largest eigenvalues of the empirical covariance matrix.
It can be done by simply shifting every eigenvalue according to a given
offset, which is equivalent of finding the l2-penalized Maximum
Likelihood Estimator of the covariance matrix. In practice, shrinkage
boils down to a simple a convex transformation : <span class="math notranslate nohighlight">\(\Sigma_{\rm
shrunk} = (1-\alpha)\hat{\Sigma} + \alpha\frac{{\rm
Tr}\hat{\Sigma}}{p}\rm Id\)</span>.</p>
<p>Choosing the amount of shrinkage, <span class="math notranslate nohighlight">\(\alpha\)</span> amounts to setting a
bias/variance trade-off, and is discussed below.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_covariance_estimation.html#sphx-glr-auto-examples-covariance-plot-covariance-estimation-py"><span class="std std-ref">Shrinkage covariance estimation: LedoitWolf vs OAS and max-likelihood</span></a> for
an example on how to fit a <a class="reference internal" href="generated/sklearn.covariance.ShrunkCovariance.html#sklearn.covariance.ShrunkCovariance" title="sklearn.covariance.ShrunkCovariance"><code class="xref py py-class docutils literal notranslate"><span class="pre">ShrunkCovariance</span></code></a> object
to data.</p></li>
</ul>
</aside>
</section>
<section id="ledoit-wolf-shrinkage">
<h3><span class="section-number">2.6.2.2. </span>Ledoit-Wolf shrinkage<a class="headerlink" href="#ledoit-wolf-shrinkage" title="Link to this heading">¶</a></h3>
<p>In their 2004 paper <a class="footnote-reference brackets" href="#id3" id="id2" role="doc-noteref"><span class="fn-bracket">[</span>1<span class="fn-bracket">]</span></a>, O. Ledoit and M. Wolf propose a formula
to compute the optimal shrinkage coefficient <span class="math notranslate nohighlight">\(\alpha\)</span> that
minimizes the Mean Squared Error between the estimated and the real
covariance matrix.</p>
<p>The Ledoit-Wolf estimator of the covariance matrix can be computed on
a sample with the <a class="reference internal" href="generated/sklearn.covariance.ledoit_wolf.html#sklearn.covariance.ledoit_wolf" title="sklearn.covariance.ledoit_wolf"><code class="xref py py-meth docutils literal notranslate"><span class="pre">ledoit_wolf</span></code></a> function of the
<a class="reference internal" href="classes.html#module-sklearn.covariance" title="sklearn.covariance"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.covariance</span></code></a> package, or it can be otherwise obtained by
fitting a <a class="reference internal" href="generated/sklearn.covariance.LedoitWolf.html#sklearn.covariance.LedoitWolf" title="sklearn.covariance.LedoitWolf"><code class="xref py py-class docutils literal notranslate"><span class="pre">LedoitWolf</span></code></a> object to the same sample.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><strong>Case when population covariance matrix is isotropic</strong></p>
<p>It is important to note that when the number of samples is much larger than
the number of features, one would expect that no shrinkage would be
necessary. The intuition behind this is that if the population covariance
is full rank, when the number of sample grows, the sample covariance will
also become positive definite. As a result, no shrinkage would necessary
and the method should automatically do this.</p>
<p>This, however, is not the case in the Ledoit-Wolf procedure when the
population covariance happens to be a multiple of the identity matrix. In
this case, the Ledoit-Wolf shrinkage estimate approaches 1 as the number of
samples increases. This indicates that the optimal estimate of the
covariance matrix in the Ledoit-Wolf sense is multiple of the identity.
Since the population covariance is already a multiple of the identity
matrix, the Ledoit-Wolf solution is indeed a reasonable estimate.</p>
</div>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_covariance_estimation.html#sphx-glr-auto-examples-covariance-plot-covariance-estimation-py"><span class="std std-ref">Shrinkage covariance estimation: LedoitWolf vs OAS and max-likelihood</span></a> for
an example on how to fit a <a class="reference internal" href="generated/sklearn.covariance.LedoitWolf.html#sklearn.covariance.LedoitWolf" title="sklearn.covariance.LedoitWolf"><code class="xref py py-class docutils literal notranslate"><span class="pre">LedoitWolf</span></code></a> object to data and
for visualizing the performances of the Ledoit-Wolf estimator in
terms of likelihood.</p></li>
</ul>
</aside>
<aside class="topic">
<p class="topic-title">References:</p>
<aside class="footnote-list brackets">
<aside class="footnote brackets" id="id3" role="doc-footnote">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id2">1</a><span class="fn-bracket">]</span></span>
<p>O. Ledoit and M. Wolf, “A Well-Conditioned Estimator for Large-Dimensional
Covariance Matrices”, Journal of Multivariate Analysis, Volume 88, Issue 2,
February 2004, pages 365-411.</p>
</aside>
</aside>
</aside>
</section>
<section id="oracle-approximating-shrinkage">
<span id="id4"></span><h3><span class="section-number">2.6.2.3. </span>Oracle Approximating Shrinkage<a class="headerlink" href="#oracle-approximating-shrinkage" title="Link to this heading">¶</a></h3>
<p>Under the assumption that the data are Gaussian distributed, Chen et
al. <a class="footnote-reference brackets" href="#id6" id="id5" role="doc-noteref"><span class="fn-bracket">[</span>2<span class="fn-bracket">]</span></a> derived a formula aimed at choosing a shrinkage coefficient that
yields a smaller Mean Squared Error than the one given by Ledoit and
Wolf’s formula. The resulting estimator is known as the Oracle
Shrinkage Approximating estimator of the covariance.</p>
<p>The OAS estimator of the covariance matrix can be computed on a sample
with the <a class="reference internal" href="generated/oas-function.html#sklearn.covariance.oas" title="sklearn.covariance.oas"><code class="xref py py-meth docutils literal notranslate"><span class="pre">oas</span></code></a> function of the <a class="reference internal" href="classes.html#module-sklearn.covariance" title="sklearn.covariance"><code class="xref py py-mod docutils literal notranslate"><span class="pre">sklearn.covariance</span></code></a>
package, or it can be otherwise obtained by fitting an <a class="reference internal" href="generated/sklearn.covariance.OAS.html#sklearn.covariance.OAS" title="sklearn.covariance.OAS"><code class="xref py py-class docutils literal notranslate"><span class="pre">OAS</span></code></a>
object to the same sample.</p>
<figure class="align-center" id="id13">
<a class="reference external image-reference" href="../auto_examples/covariance/plot_covariance_estimation.html"><img alt="../_images/sphx_glr_plot_covariance_estimation_001.png" src="../_images/sphx_glr_plot_covariance_estimation_001.png" style="width: 416.0px; height: 312.0px;" /></a>
<figcaption>
<p><span class="caption-text">Bias-variance trade-off when setting the shrinkage: comparing the
choices of Ledoit-Wolf and OAS estimators</span><a class="headerlink" href="#id13" title="Link to this image">¶</a></p>
</figcaption>
</figure>
<aside class="topic">
<p class="topic-title">References:</p>
<aside class="footnote-list brackets">
<aside class="footnote brackets" id="id6" role="doc-footnote">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id5">2</a><span class="fn-bracket">]</span></span>
<p><a class="reference external" href="https://arxiv.org/abs/0907.4698">“Shrinkage algorithms for MMSE covariance estimation.”,
Chen, Y., Wiesel, A., Eldar, Y. C., & Hero, A. O.
IEEE Transactions on Signal Processing, 58(10), 5016-5029, 2010.</a></p>
</aside>
</aside>
</aside>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_covariance_estimation.html#sphx-glr-auto-examples-covariance-plot-covariance-estimation-py"><span class="std std-ref">Shrinkage covariance estimation: LedoitWolf vs OAS and max-likelihood</span></a> for
an example on how to fit an <a class="reference internal" href="generated/sklearn.covariance.OAS.html#sklearn.covariance.OAS" title="sklearn.covariance.OAS"><code class="xref py py-class docutils literal notranslate"><span class="pre">OAS</span></code></a> object
to data.</p></li>
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_lw_vs_oas.html#sphx-glr-auto-examples-covariance-plot-lw-vs-oas-py"><span class="std std-ref">Ledoit-Wolf vs OAS estimation</span></a> to visualize the
Mean Squared Error difference between a <a class="reference internal" href="generated/sklearn.covariance.LedoitWolf.html#sklearn.covariance.LedoitWolf" title="sklearn.covariance.LedoitWolf"><code class="xref py py-class docutils literal notranslate"><span class="pre">LedoitWolf</span></code></a> and
an <a class="reference internal" href="generated/sklearn.covariance.OAS.html#sklearn.covariance.OAS" title="sklearn.covariance.OAS"><code class="xref py py-class docutils literal notranslate"><span class="pre">OAS</span></code></a> estimator of the covariance.</p></li>
</ul>
</aside>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/covariance/plot_lw_vs_oas.html"><img alt="../_images/sphx_glr_plot_lw_vs_oas_001.png" src="../_images/sphx_glr_plot_lw_vs_oas_001.png" style="width: 480.0px; height: 360.0px;" /></a>
</figure>
</section>
</section>
<section id="sparse-inverse-covariance">
<span id="id7"></span><h2><span class="section-number">2.6.3. </span>Sparse inverse covariance<a class="headerlink" href="#sparse-inverse-covariance" title="Link to this heading">¶</a></h2>
<p>The matrix inverse of the covariance matrix, often called the precision
matrix, is proportional to the partial correlation matrix. It gives the
partial independence relationship. In other words, if two features are
independent conditionally on the others, the corresponding coefficient in
the precision matrix will be zero. This is why it makes sense to
estimate a sparse precision matrix: the estimation of the covariance
matrix is better conditioned by learning independence relations from
the data. This is known as <em>covariance selection</em>.</p>
<p>In the small-samples situation, in which <code class="docutils literal notranslate"><span class="pre">n_samples</span></code> is on the order
of <code class="docutils literal notranslate"><span class="pre">n_features</span></code> or smaller, sparse inverse covariance estimators tend to work
better than shrunk covariance estimators. However, in the opposite
situation, or for very correlated data, they can be numerically unstable.
In addition, unlike shrinkage estimators, sparse estimators are able to
recover off-diagonal structure.</p>
<p>The <a class="reference internal" href="generated/sklearn.covariance.GraphicalLasso.html#sklearn.covariance.GraphicalLasso" title="sklearn.covariance.GraphicalLasso"><code class="xref py py-class docutils literal notranslate"><span class="pre">GraphicalLasso</span></code></a> estimator uses an l1 penalty to enforce sparsity on
the precision matrix: the higher its <code class="docutils literal notranslate"><span class="pre">alpha</span></code> parameter, the more sparse
the precision matrix. The corresponding <a class="reference internal" href="generated/sklearn.covariance.GraphicalLassoCV.html#sklearn.covariance.GraphicalLassoCV" title="sklearn.covariance.GraphicalLassoCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">GraphicalLassoCV</span></code></a> object uses
cross-validation to automatically set the <code class="docutils literal notranslate"><span class="pre">alpha</span></code> parameter.</p>
<figure class="align-center" id="id14">
<a class="reference external image-reference" href="../auto_examples/covariance/plot_sparse_cov.html"><img alt="../_images/sphx_glr_plot_sparse_cov_001.png" src="../_images/sphx_glr_plot_sparse_cov_001.png" style="width: 600.0px; height: 360.0px;" /></a>
<figcaption>
<p><span class="caption-text"><em>A comparison of maximum likelihood, shrinkage and sparse estimates of
the covariance and precision matrix in the very small samples
settings.</em></span><a class="headerlink" href="#id14" title="Link to this image">¶</a></p>
</figcaption>
</figure>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><strong>Structure recovery</strong></p>
<p>Recovering a graphical structure from correlations in the data is a
challenging thing. If you are interested in such recovery keep in mind
that:</p>
<ul class="simple">
<li><p>Recovery is easier from a correlation matrix than a covariance
matrix: standardize your observations before running <a class="reference internal" href="generated/sklearn.covariance.GraphicalLasso.html#sklearn.covariance.GraphicalLasso" title="sklearn.covariance.GraphicalLasso"><code class="xref py py-class docutils literal notranslate"><span class="pre">GraphicalLasso</span></code></a></p></li>
<li><p>If the underlying graph has nodes with much more connections than
the average node, the algorithm will miss some of these connections.</p></li>
<li><p>If your number of observations is not large compared to the number
of edges in your underlying graph, you will not recover it.</p></li>
<li><p>Even if you are in favorable recovery conditions, the alpha
parameter chosen by cross-validation (e.g. using the
<a class="reference internal" href="generated/sklearn.covariance.GraphicalLassoCV.html#sklearn.covariance.GraphicalLassoCV" title="sklearn.covariance.GraphicalLassoCV"><code class="xref py py-class docutils literal notranslate"><span class="pre">GraphicalLassoCV</span></code></a> object) will lead to selecting too many edges.
However, the relevant edges will have heavier weights than the
irrelevant ones.</p></li>
</ul>
</div>
<p>The mathematical formulation is the following:</p>
<div class="math notranslate nohighlight">
\[\hat{K} = \mathrm{argmin}_K \big(
\mathrm{tr} S K - \mathrm{log} \mathrm{det} K
+ \alpha \|K\|_1
\big)\]</div>
<p>Where <span class="math notranslate nohighlight">\(K\)</span> is the precision matrix to be estimated, and <span class="math notranslate nohighlight">\(S\)</span> is the
sample covariance matrix. <span class="math notranslate nohighlight">\(\|K\|_1\)</span> is the sum of the absolute values of
off-diagonal coefficients of <span class="math notranslate nohighlight">\(K\)</span>. The algorithm employed to solve this
problem is the GLasso algorithm, from the Friedman 2008 Biostatistics
paper. It is the same algorithm as in the R <code class="docutils literal notranslate"><span class="pre">glasso</span></code> package.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/covariance/plot_sparse_cov.html#sphx-glr-auto-examples-covariance-plot-sparse-cov-py"><span class="std std-ref">Sparse inverse covariance estimation</span></a>: example on synthetic
data showing some recovery of a structure, and comparing to other
covariance estimators.</p></li>
<li><p><a class="reference internal" href="../auto_examples/applications/plot_stock_market.html#sphx-glr-auto-examples-applications-plot-stock-market-py"><span class="std std-ref">Visualizing the stock market structure</span></a>: example on real
stock market data, finding which symbols are most linked.</p></li>
</ul>
</aside>
<aside class="topic">
<p class="topic-title">References:</p>
<ul class="simple">
<li><p>Friedman et al, <a class="reference external" href="https://biostatistics.oxfordjournals.org/content/9/3/432.short">“Sparse inverse covariance estimation with the
graphical lasso”</a>,
Biostatistics 9, pp 432, 2008</p></li>
</ul>
</aside>
</section>
<section id="robust-covariance-estimation">
<span id="robust-covariance"></span><h2><span class="section-number">2.6.4. </span>Robust Covariance Estimation<a class="headerlink" href="#robust-covariance-estimation" title="Link to this heading">¶</a></h2>
<p>Real data sets are often subject to measurement or recording
errors. Regular but uncommon observations may also appear for a variety
of reasons. Observations which are very uncommon are called
outliers.
The empirical covariance estimator and the shrunk covariance
estimators presented above are very sensitive to the presence of
outliers in the data. Therefore, one should use robust
covariance estimators to estimate the covariance of its real data
sets. Alternatively, robust covariance estimators can be used to
perform outlier detection and discard/downweight some observations
according to further processing of the data.</p>
<p>The <code class="docutils literal notranslate"><span class="pre">sklearn.covariance</span></code> package implements a robust estimator of covariance,
the Minimum Covariance Determinant <a class="footnote-reference brackets" href="#id11" id="id8" role="doc-noteref"><span class="fn-bracket">[</span>3<span class="fn-bracket">]</span></a>.</p>
<section id="minimum-covariance-determinant">
<h3><span class="section-number">2.6.4.1. </span>Minimum Covariance Determinant<a class="headerlink" href="#minimum-covariance-determinant" title="Link to this heading">¶</a></h3>
<p>The Minimum Covariance Determinant estimator is a robust estimator of
a data set’s covariance introduced by P.J. Rousseeuw in <a class="footnote-reference brackets" href="#id11" id="id9" role="doc-noteref"><span class="fn-bracket">[</span>3<span class="fn-bracket">]</span></a>. The idea
is to find a given proportion (h) of “good” observations which are not
outliers and compute their empirical covariance matrix. This
empirical covariance matrix is then rescaled to compensate the
performed selection of observations (“consistency step”). Having
computed the Minimum Covariance Determinant estimator, one can give
weights to observations according to their Mahalanobis distance,
leading to a reweighted estimate of the covariance matrix of the data
set (“reweighting step”).</p>
<p>Rousseeuw and Van Driessen <a class="footnote-reference brackets" href="#id12" id="id10" role="doc-noteref"><span class="fn-bracket">[</span>4<span class="fn-bracket">]</span></a> developed the FastMCD algorithm in order
to compute the Minimum Covariance Determinant. This algorithm is used
in scikit-learn when fitting an MCD object to data. The FastMCD
algorithm also computes a robust estimate of the data set location at
the same time.</p>
<p>Raw estimates can be accessed as <code class="docutils literal notranslate"><span class="pre">raw_location_</span></code> and <code class="docutils literal notranslate"><span class="pre">raw_covariance_</span></code>
attributes of a <a class="reference internal" href="generated/sklearn.covariance.MinCovDet.html#sklearn.covariance.MinCovDet" title="sklearn.covariance.MinCovDet"><code class="xref py py-class docutils literal notranslate"><span class="pre">MinCovDet</span></code></a> robust covariance estimator object.</p>
<aside class="topic">
<p class="topic-title">References:</p>
<aside class="footnote-list brackets">
<aside class="footnote brackets" id="id11" role="doc-footnote">
<span class="label"><span class="fn-bracket">[</span>3<span class="fn-bracket">]</span></span>
<span class="backrefs">(<a role="doc-backlink" href="#id8">1</a>,<a role="doc-backlink" href="#id9">2</a>)</span>
<p>P. J. Rousseeuw. Least median of squares regression.
J. Am Stat Ass, 79:871, 1984.</p>
</aside>
<aside class="footnote brackets" id="id12" role="doc-footnote">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id10">4</a><span class="fn-bracket">]</span></span>
<p>A Fast Algorithm for the Minimum Covariance Determinant Estimator,
1999, American Statistical Association and the American Society
for Quality, TECHNOMETRICS.</p>
</aside>
</aside>
</aside>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_robust_vs_empirical_covariance.html#sphx-glr-auto-examples-covariance-plot-robust-vs-empirical-covariance-py"><span class="std std-ref">Robust vs Empirical covariance estimate</span></a> for
an example on how to fit a <a class="reference internal" href="generated/sklearn.covariance.MinCovDet.html#sklearn.covariance.MinCovDet" title="sklearn.covariance.MinCovDet"><code class="xref py py-class docutils literal notranslate"><span class="pre">MinCovDet</span></code></a> object to data and see how
the estimate remains accurate despite the presence of outliers.</p></li>
<li><p>See <a class="reference internal" href="../auto_examples/covariance/plot_mahalanobis_distances.html#sphx-glr-auto-examples-covariance-plot-mahalanobis-distances-py"><span class="std std-ref">Robust covariance estimation and Mahalanobis distances relevance</span></a> to
visualize the difference between <a class="reference internal" href="generated/sklearn.covariance.EmpiricalCovariance.html#sklearn.covariance.EmpiricalCovariance" title="sklearn.covariance.EmpiricalCovariance"><code class="xref py py-class docutils literal notranslate"><span class="pre">EmpiricalCovariance</span></code></a> and
<a class="reference internal" href="generated/sklearn.covariance.MinCovDet.html#sklearn.covariance.MinCovDet" title="sklearn.covariance.MinCovDet"><code class="xref py py-class docutils literal notranslate"><span class="pre">MinCovDet</span></code></a> covariance estimators in terms of Mahalanobis distance
(so we get a better estimate of the precision matrix too).</p></li>
</ul>
</aside>
<hr class="docutils" />
<table class="docutils align-default">
<thead>
<tr class="row-odd"><th class="head"><p>Influence of outliers on location and covariance estimates</p></th>
<th class="head"><p>Separating inliers from outliers using a Mahalanobis distance</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p><a class="reference external" href="../auto_examples/covariance/plot_robust_vs_empirical_covariance.html"><img alt="robust_vs_emp" src="../_images/sphx_glr_plot_robust_vs_empirical_covariance_001.png" style="width: 313.6px; height: 235.2px;" /></a></p></td>
<td><p><a class="reference external" href="../auto_examples/covariance/plot_mahalanobis_distances.html"><img alt="mahalanobis" src="../_images/sphx_glr_plot_mahalanobis_distances_001.png" style="width: 490.0px; height: 245.0px;" /></a></p></td>
</tr>
</tbody>
</table>
</section>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2023, scikit-learn developers (BSD License).
<a href="../_sources/modules/covariance.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>