-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathmixture.html
613 lines (581 loc) · 38.1 KB
/
mixture.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
<!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="2.1. Gaussian mixture models" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/modules/mixture.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="sklearn.mixture is a package which enables one to learn Gaussian Mixture Models (diagonal, spherical, tied and full covariance matrices supported), sample them, and estimate them from data. Facilit..." />
<meta property="og:image" content="https://scikit-learn/stable/_images/sphx_glr_plot_gmm_pdf_001.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="sklearn.mixture is a package which enables one to learn Gaussian Mixture Models (diagonal, spherical, tied and full covariance matrices supported), sample them, and estimate them from data. Facilit..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>2.1. Gaussian mixture models — scikit-learn 1.5.dev0 documentation</title>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/mixture.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="../unsupervised_learning.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2. Unsupervised learning">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="manifold.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2.2. Manifold learning">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="#">2.1. Gaussian mixture models</a><ul>
<li><a class="reference internal" href="#gaussian-mixture">2.1.1. Gaussian Mixture</a></li>
<li><a class="reference internal" href="#variational-bayesian-gaussian-mixture">2.1.2. Variational Bayesian Gaussian Mixture</a><ul>
<li><a class="reference internal" href="#the-dirichlet-process">2.1.2.1. The Dirichlet Process</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="gaussian-mixture-models">
<span id="gmm"></span><span id="mixture"></span><h1><span class="section-number">2.1. </span>Gaussian mixture models<a class="headerlink" href="#gaussian-mixture-models" title="Link to this heading">¶</a></h1>
<p><code class="docutils literal notranslate"><span class="pre">sklearn.mixture</span></code> is a package which enables one to learn
Gaussian Mixture Models (diagonal, spherical, tied and full covariance
matrices supported), sample them, and estimate them from
data. Facilities to help determine the appropriate number of
components are also provided.</p>
<blockquote>
<div><figure class="align-center" id="id2">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm_pdf.html"><img alt="../_images/sphx_glr_plot_gmm_pdf_001.png" src="../_images/sphx_glr_plot_gmm_pdf_001.png" style="width: 320.0px; height: 240.0px;" /></a>
<figcaption>
<p><span class="caption-text"><strong>Two-component Gaussian mixture model:</strong> <em>data points, and equi-probability
surfaces of the model.</em></span><a class="headerlink" href="#id2" title="Link to this image">¶</a></p>
</figcaption>
</figure>
</div></blockquote>
<p>A Gaussian mixture model is a probabilistic model that assumes all the
data points are generated from a mixture of a finite number of
Gaussian distributions with unknown parameters. One can think of
mixture models as generalizing k-means clustering to incorporate
information about the covariance structure of the data as well as the
centers of the latent Gaussians.</p>
<p>Scikit-learn implements different classes to estimate Gaussian
mixture models, that correspond to different estimation strategies,
detailed below.</p>
<section id="gaussian-mixture">
<h2><span class="section-number">2.1.1. </span>Gaussian Mixture<a class="headerlink" href="#gaussian-mixture" title="Link to this heading">¶</a></h2>
<p>The <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture" title="sklearn.mixture.GaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianMixture</span></code></a> object implements the
<a class="reference internal" href="#expectation-maximization"><span class="std std-ref">expectation-maximization</span></a> (EM)
algorithm for fitting mixture-of-Gaussian models. It can also draw
confidence ellipsoids for multivariate models, and compute the
Bayesian Information Criterion to assess the number of clusters in the
data. A <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture.fit" title="sklearn.mixture.GaussianMixture.fit"><code class="xref py py-meth docutils literal notranslate"><span class="pre">GaussianMixture.fit</span></code></a> method is provided that learns a Gaussian
Mixture Model from train data. Given test data, it can assign to each
sample the Gaussian it most probably belongs to using
the <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture.predict" title="sklearn.mixture.GaussianMixture.predict"><code class="xref py py-meth docutils literal notranslate"><span class="pre">GaussianMixture.predict</span></code></a> method.</p>
<p>The <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture" title="sklearn.mixture.GaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianMixture</span></code></a> comes with different options to constrain the
covariance of the difference classes estimated: spherical, diagonal, tied or
full covariance.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm_covariances.html"><img alt="../_images/sphx_glr_plot_gmm_covariances_001.png" src="../_images/sphx_glr_plot_gmm_covariances_001.png" style="width: 450.0px; height: 450.0px;" /></a>
</figure>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_gmm_covariances.html#sphx-glr-auto-examples-mixture-plot-gmm-covariances-py"><span class="std std-ref">GMM covariances</span></a> for an example of
using the Gaussian mixture as clustering on the iris dataset.</p></li>
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_gmm_pdf.html#sphx-glr-auto-examples-mixture-plot-gmm-pdf-py"><span class="std std-ref">Density Estimation for a Gaussian mixture</span></a> for an example on plotting the
density estimation.</p></li>
</ul>
</aside>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Pros and cons of class GaussianMixture</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>
<aside class="topic">
<p class="topic-title">Pros:</p>
<dl class="field-list simple">
<dt class="field-odd">Speed<span class="colon">:</span></dt>
<dd class="field-odd"><p>It is the fastest algorithm for learning mixture models</p>
</dd>
<dt class="field-even">Agnostic<span class="colon">:</span></dt>
<dd class="field-even"><p>As this algorithm maximizes only the likelihood, it
will not bias the means towards zero, or bias the cluster sizes to
have specific structures that might or might not apply.</p>
</dd>
</dl>
</aside>
<aside class="topic">
<p class="topic-title">Cons:</p>
<dl class="field-list simple">
<dt class="field-odd">Singularities<span class="colon">:</span></dt>
<dd class="field-odd"><p>When one has insufficiently many points per
mixture, estimating the covariance matrices becomes difficult,
and the algorithm is known to diverge and find solutions with
infinite likelihood unless one regularizes the covariances artificially.</p>
</dd>
<dt class="field-even">Number of components<span class="colon">:</span></dt>
<dd class="field-even"><p>This algorithm will always use all the
components it has access to, needing held-out data
or information theoretical criteria to decide how many components to use
in the absence of external cues.</p>
</dd>
</dl>
</aside>
<p></div>
</details></p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Selecting the number of components in a classical Gaussian Mixture model</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>The BIC criterion can be used to select the number of components in a Gaussian
Mixture in an efficient way. In theory, it recovers the true number of
components only in the asymptotic regime (i.e. if much data is available and
assuming that the data was actually generated i.i.d. from a mixture of Gaussian
distribution). Note that using a <a class="reference internal" href="#bgmm"><span class="std std-ref">Variational Bayesian Gaussian mixture</span></a>
avoids the specification of the number of components for a Gaussian mixture
model.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm_selection.html"><img alt="../_images/sphx_glr_plot_gmm_selection_002.png" src="../_images/sphx_glr_plot_gmm_selection_002.png" style="width: 320.5px; height: 250.0px;" /></a>
</figure>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_gmm_selection.html#sphx-glr-auto-examples-mixture-plot-gmm-selection-py"><span class="std std-ref">Gaussian Mixture Model Selection</span></a> for an example
of model selection performed with classical Gaussian mixture.</p></li>
</ul>
</aside>
<p></div>
</details></p>
<p id="expectation-maximization"><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Estimation algorithm expectation-maximization</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>The main difficulty in learning Gaussian mixture models from unlabeled
data is that one usually doesn’t know which points came from
which latent component (if one has access to this information it gets
very easy to fit a separate Gaussian distribution to each set of
points). <a class="reference external" href="https://en.wikipedia.org/wiki/Expectation%E2%80%93maximization_algorithm">Expectation-maximization</a>
is a well-founded statistical
algorithm to get around this problem by an iterative process. First
one assumes random components (randomly centered on data points,
learned from k-means, or even just normally distributed around the
origin) and computes for each point a probability of being generated by
each component of the model. Then, one tweaks the
parameters to maximize the likelihood of the data given those
assignments. Repeating this process is guaranteed to always converge
to a local optimum.</p>
<p></div>
</details></p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Choice of the Initialization method</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>There is a choice of four initialization methods (as well as inputting user defined
initial means) to generate the initial centers for the model components:</p>
<dl class="simple">
<dt>k-means (default)</dt><dd><p>This applies a traditional k-means clustering algorithm.
This can be computationally expensive compared to other initialization methods.</p>
</dd>
<dt>k-means++</dt><dd><p>This uses the initialization method of k-means clustering: k-means++.
This will pick the first center at random from the data. Subsequent centers will be
chosen from a weighted distribution of the data favouring points further away from
existing centers. k-means++ is the default initialization for k-means so will be
quicker than running a full k-means but can still take a significant amount of
time for large data sets with many components.</p>
</dd>
<dt>random_from_data</dt><dd><p>This will pick random data points from the input data as the initial
centers. This is a very fast method of initialization but can produce non-convergent
results if the chosen points are too close to each other.</p>
</dd>
<dt>random</dt><dd><p>Centers are chosen as a small perturbation away from the mean of all data.
This method is simple but can lead to the model taking longer to converge.</p>
</dd>
</dl>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm_init.html"><img alt="../_images/sphx_glr_plot_gmm_init_001.png" src="../_images/sphx_glr_plot_gmm_init_001.png" style="width: 400.0px; height: 300.0px;" /></a>
</figure>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_gmm_init.html#sphx-glr-auto-examples-mixture-plot-gmm-init-py"><span class="std std-ref">GMM Initialization Methods</span></a> for an example of
using different initializations in Gaussian Mixture.</p></li>
</ul>
</aside>
<p></div>
</details></p>
</section>
<section id="variational-bayesian-gaussian-mixture">
<span id="bgmm"></span><h2><span class="section-number">2.1.2. </span>Variational Bayesian Gaussian Mixture<a class="headerlink" href="#variational-bayesian-gaussian-mixture" title="Link to this heading">¶</a></h2>
<p>The <a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a> object implements a variant of the
Gaussian mixture model with variational inference algorithms. The API is
similar to the one defined by <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture" title="sklearn.mixture.GaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianMixture</span></code></a>.</p>
<p id="variational-inference"><strong>Estimation algorithm: variational inference</strong></p>
<p>Variational inference is an extension of expectation-maximization that
maximizes a lower bound on model evidence (including
priors) instead of data likelihood. The principle behind
variational methods is the same as expectation-maximization (that is
both are iterative algorithms that alternate between finding the
probabilities for each point to be generated by each mixture and
fitting the mixture to these assigned points), but variational
methods add regularization by integrating information from prior
distributions. This avoids the singularities often found in
expectation-maximization solutions but introduces some subtle biases
to the model. Inference is often notably slower, but not usually as
much so as to render usage unpractical.</p>
<p>Due to its Bayesian nature, the variational algorithm needs more hyperparameters
than expectation-maximization, the most important of these being the
concentration parameter <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code>. Specifying a low value
for the concentration prior will make the model put most of the weight on a few
components and set the remaining components’ weights very close to zero. High
values of the concentration prior will allow a larger number of components to
be active in the mixture.</p>
<p>The parameters implementation of the <a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a> class
proposes two types of prior for the weights distribution: a finite mixture model
with Dirichlet distribution and an infinite mixture model with the Dirichlet
Process. In practice Dirichlet Process inference algorithm is approximated and
uses a truncated distribution with a fixed maximum number of components (called
the Stick-breaking representation). The number of components actually used
almost always depends on the data.</p>
<p>The next figure compares the results obtained for the different type of the
weight concentration prior (parameter <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior_type</span></code>)
for different values of <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code>.
Here, we can see the value of the <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code> parameter
has a strong impact on the effective number of active components obtained. We
can also notice that large values for the concentration weight prior lead to
more uniform weights when the type of prior is ‘dirichlet_distribution’ while
this is not necessarily the case for the ‘dirichlet_process’ type (used by
default).</p>
<p class="centered">
<strong><a class="reference external" href="../auto_examples/mixture/plot_concentration_prior.html"><img alt="plot_bgmm" src="../_images/sphx_glr_plot_concentration_prior_001.png" style="width: 676.8px; height: 384.0px;" /></a> <a class="reference external" href="../auto_examples/mixture/plot_concentration_prior.html"><img alt="plot_dpgmm" src="../_images/sphx_glr_plot_concentration_prior_002.png" style="width: 676.8px; height: 384.0px;" /></a></strong></p><p>The examples below compare Gaussian mixture models with a fixed number of
components, to the variational Gaussian mixture models with a Dirichlet process
prior. Here, a classical Gaussian mixture is fitted with 5 components on a
dataset composed of 2 clusters. We can see that the variational Gaussian mixture
with a Dirichlet process prior is able to limit itself to only 2 components
whereas the Gaussian mixture fits the data with a fixed number of components
that has to be set a priori by the user. In this case the user has selected
<code class="docutils literal notranslate"><span class="pre">n_components=5</span></code> which does not match the true generative distribution of this
toy dataset. Note that with very little observations, the variational Gaussian
mixture models with a Dirichlet process prior can take a conservative stand, and
fit only one component.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm.html"><img alt="../_images/sphx_glr_plot_gmm_001.png" src="../_images/sphx_glr_plot_gmm_001.png" style="width: 448.0px; height: 336.0px;" /></a>
</figure>
<p>On the following figure we are fitting a dataset not well-depicted by a
Gaussian mixture. Adjusting the <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code>, parameter of the
<a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a> controls the number of components used to fit
this data. We also present on the last two plots a random sampling generated
from the two resulting mixtures.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/mixture/plot_gmm_sin.html"><img alt="../_images/sphx_glr_plot_gmm_sin_001.png" src="../_images/sphx_glr_plot_gmm_sin_001.png" style="width: 650.0px; height: 650.0px;" /></a>
</figure>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_gmm.html#sphx-glr-auto-examples-mixture-plot-gmm-py"><span class="std std-ref">Gaussian Mixture Model Ellipsoids</span></a> for an example on
plotting the confidence ellipsoids for both <a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture" title="sklearn.mixture.GaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianMixture</span></code></a>
and <a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a>.</p></li>
<li><p><a class="reference internal" href="../auto_examples/mixture/plot_gmm_sin.html#sphx-glr-auto-examples-mixture-plot-gmm-sin-py"><span class="std std-ref">Gaussian Mixture Model Sine Curve</span></a> shows using
<a class="reference internal" href="generated/sklearn.mixture.GaussianMixture.html#sklearn.mixture.GaussianMixture" title="sklearn.mixture.GaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">GaussianMixture</span></code></a> and <a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a> to fit a
sine wave.</p></li>
<li><p>See <a class="reference internal" href="../auto_examples/mixture/plot_concentration_prior.html#sphx-glr-auto-examples-mixture-plot-concentration-prior-py"><span class="std std-ref">Concentration Prior Type Analysis of Variation Bayesian Gaussian Mixture</span></a>
for an example plotting the confidence ellipsoids for the
<a class="reference internal" href="generated/sklearn.mixture.BayesianGaussianMixture.html#sklearn.mixture.BayesianGaussianMixture" title="sklearn.mixture.BayesianGaussianMixture"><code class="xref py py-class docutils literal notranslate"><span class="pre">BayesianGaussianMixture</span></code></a> with different
<code class="docutils literal notranslate"><span class="pre">weight_concentration_prior_type</span></code> for different values of the parameter
<code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code>.</p></li>
</ul>
</aside>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Pros and cons of variational inference with BayesianGaussianMixture</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>
<aside class="topic">
<p class="topic-title">Pros:</p>
<dl class="field-list simple">
<dt class="field-odd">Automatic selection<span class="colon">:</span></dt>
<dd class="field-odd"><p>when <code class="docutils literal notranslate"><span class="pre">weight_concentration_prior</span></code> is small enough and
<code class="docutils literal notranslate"><span class="pre">n_components</span></code> is larger than what is found necessary by the model, the
Variational Bayesian mixture model has a natural tendency to set some mixture
weights values close to zero. This makes it possible to let the model choose
a suitable number of effective components automatically. Only an upper bound
of this number needs to be provided. Note however that the “ideal” number of
active components is very application specific and is typically ill-defined
in a data exploration setting.</p>
</dd>
<dt class="field-even">Less sensitivity to the number of parameters<span class="colon">:</span></dt>
<dd class="field-even"><p>unlike finite models, which will
almost always use all components as much as they can, and hence will produce
wildly different solutions for different numbers of components, the
variational inference with a Dirichlet process prior
(<code class="docutils literal notranslate"><span class="pre">weight_concentration_prior_type='dirichlet_process'</span></code>) won’t change much
with changes to the parameters, leading to more stability and less tuning.</p>
</dd>
<dt class="field-odd">Regularization<span class="colon">:</span></dt>
<dd class="field-odd"><p>due to the incorporation of prior information,
variational solutions have less pathological special cases than
expectation-maximization solutions.</p>
</dd>
</dl>
</aside>
<aside class="topic">
<p class="topic-title">Cons:</p>
<dl class="field-list simple">
<dt class="field-odd">Speed<span class="colon">:</span></dt>
<dd class="field-odd"><p>the extra parametrization necessary for variational inference makes
inference slower, although not by much.</p>
</dd>
<dt class="field-even">Hyperparameters<span class="colon">:</span></dt>
<dd class="field-even"><p>this algorithm needs an extra hyperparameter
that might need experimental tuning via cross-validation.</p>
</dd>
<dt class="field-odd">Bias<span class="colon">:</span></dt>
<dd class="field-odd"><p>there are many implicit biases in the inference algorithms (and also in
the Dirichlet process if used), and whenever there is a mismatch between
these biases and the data it might be possible to fit better models using a
finite mixture.</p>
</dd>
</dl>
</aside>
<p></div>
</details></p>
<section id="the-dirichlet-process">
<span id="dirichlet-process"></span><h3><span class="section-number">2.1.2.1. </span>The Dirichlet Process<a class="headerlink" href="#the-dirichlet-process" title="Link to this heading">¶</a></h3>
<p>Here we describe variational inference algorithms on Dirichlet process
mixture. The Dirichlet process is a prior probability distribution on
<em>clusterings with an infinite, unbounded, number of partitions</em>.
Variational techniques let us incorporate this prior structure on
Gaussian mixture models at almost no penalty in inference time, comparing
with a finite Gaussian mixture model.</p>
<p>An important question is how can the Dirichlet process use an infinite,
unbounded number of clusters and still be consistent. While a full explanation
doesn’t fit this manual, one can think of its <a class="reference external" href="https://en.wikipedia.org/wiki/Dirichlet_process#The_stick-breaking_process">stick breaking process</a>
analogy to help understanding it. The stick breaking process is a generative
story for the Dirichlet process. We start with a unit-length stick and in each
step we break off a portion of the remaining stick. Each time, we associate the
length of the piece of the stick to the proportion of points that falls into a
group of the mixture. At the end, to represent the infinite mixture, we
associate the last remaining piece of the stick to the proportion of points
that don’t fall into all the other groups. The length of each piece is a random
variable with probability proportional to the concentration parameter. Smaller
values of the concentration will divide the unit-length into larger pieces of
the stick (defining more concentrated distribution). Larger concentration
values will create smaller pieces of the stick (increasing the number of
components with non zero weights).</p>
<p>Variational inference techniques for the Dirichlet process still work
with a finite approximation to this infinite mixture model, but
instead of having to specify a priori how many components one wants to
use, one just specifies the concentration parameter and an upper bound
on the number of mixture components (this upper bound, assuming it is
higher than the “true” number of components, affects only algorithmic
complexity, not the actual number of components used).</p>
</section>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2023, scikit-learn developers (BSD License).
<a href="../_sources/modules/mixture.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>