-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathneural_networks_supervised.html
601 lines (569 loc) · 46.7 KB
/
neural_networks_supervised.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
<!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="1.17. Neural network models (supervised)" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/modules/neural_networks_supervised.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="Multi-layer Perceptron: Multi-layer Perceptron (MLP) is a supervised learning algorithm that learns a function f: R^m \rightarrow R^o by training on a dataset, where m is the number of dimensions f..." />
<meta property="og:image" content="https://scikit-learn/stable/_images/multilayerperceptron_network.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="Multi-layer Perceptron: Multi-layer Perceptron (MLP) is a supervised learning algorithm that learns a function f: R^m \rightarrow R^o by training on a dataset, where m is the number of dimensions f..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>1.17. Neural network models (supervised) — scikit-learn 1.5.dev0 documentation</title>
<link rel="canonical" href="http://scikit-learn.org/stable/modules/neural_networks_supervised.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="calibration.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="1.16. Probability calibration">Prev</a><a href="../supervised_learning.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="1. Supervised learning">Up</a>
<a href="../unsupervised_learning.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="2. Unsupervised 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="#">1.17. Neural network models (supervised)</a><ul>
<li><a class="reference internal" href="#multi-layer-perceptron">1.17.1. Multi-layer Perceptron</a></li>
<li><a class="reference internal" href="#classification">1.17.2. Classification</a></li>
<li><a class="reference internal" href="#regression">1.17.3. Regression</a></li>
<li><a class="reference internal" href="#regularization">1.17.4. Regularization</a></li>
<li><a class="reference internal" href="#algorithms">1.17.5. Algorithms</a></li>
<li><a class="reference internal" href="#complexity">1.17.6. Complexity</a></li>
<li><a class="reference internal" href="#tips-on-practical-use">1.17.7. Tips on Practical Use</a></li>
<li><a class="reference internal" href="#more-control-with-warm-start">1.17.8. More control with warm_start</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
<div id="sk-page-content-wrapper">
<div class="sk-page-content container-fluid body px-md-3" role="main">
<section id="neural-network-models-supervised">
<span id="neural-networks-supervised"></span><h1><span class="section-number">1.17. </span>Neural network models (supervised)<a class="headerlink" href="#neural-network-models-supervised" title="Link to this heading">¶</a></h1>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>This implementation is not intended for large-scale applications. In particular,
scikit-learn offers no GPU support. For much faster, GPU-based implementations,
as well as frameworks offering much more flexibility to build deep learning
architectures, see <a class="reference internal" href="../related_projects.html#related-projects"><span class="std std-ref">Related Projects</span></a>.</p>
</div>
<section id="multi-layer-perceptron">
<span id="multilayer-perceptron"></span><h2><span class="section-number">1.17.1. </span>Multi-layer Perceptron<a class="headerlink" href="#multi-layer-perceptron" title="Link to this heading">¶</a></h2>
<p><strong>Multi-layer Perceptron (MLP)</strong> is a supervised learning algorithm that learns
a function <span class="math notranslate nohighlight">\(f: R^m \rightarrow R^o\)</span> by training on a dataset,
where <span class="math notranslate nohighlight">\(m\)</span> is the number of dimensions for input and <span class="math notranslate nohighlight">\(o\)</span> is the
number of dimensions for output. Given a set of features <span class="math notranslate nohighlight">\(X = {x_1, x_2, ..., x_m}\)</span>
and a target <span class="math notranslate nohighlight">\(y\)</span>, it can learn a non-linear function approximator for either
classification or regression. It is different from logistic regression, in that
between the input and the output layer, there can be one or more non-linear
layers, called hidden layers. Figure 1 shows a one hidden layer MLP with scalar
output.</p>
<figure class="align-center" id="id3">
<a class="reference internal image-reference" href="../_images/multilayerperceptron_network.png"><img alt="../_images/multilayerperceptron_network.png" src="../_images/multilayerperceptron_network.png" style="width: 469.79999999999995px; height: 510.59999999999997px;" /></a>
<figcaption>
<p><span class="caption-text"><strong>Figure 1 : One hidden layer MLP.</strong></span><a class="headerlink" href="#id3" title="Link to this image">¶</a></p>
</figcaption>
</figure>
<p>The leftmost layer, known as the input layer, consists of a set of neurons
<span class="math notranslate nohighlight">\(\{x_i | x_1, x_2, ..., x_m\}\)</span> representing the input features. Each
neuron in the hidden layer transforms the values from the previous layer with
a weighted linear summation <span class="math notranslate nohighlight">\(w_1x_1 + w_2x_2 + ... + w_mx_m\)</span>, followed
by a non-linear activation function <span class="math notranslate nohighlight">\(g(\cdot):R \rightarrow R\)</span> - like
the hyperbolic tan function. The output layer receives the values from the
last hidden layer and transforms them into output values.</p>
<p>The module contains the public attributes <code class="docutils literal notranslate"><span class="pre">coefs_</span></code> and <code class="docutils literal notranslate"><span class="pre">intercepts_</span></code>.
<code class="docutils literal notranslate"><span class="pre">coefs_</span></code> is a list of weight matrices, where weight matrix at index
<span class="math notranslate nohighlight">\(i\)</span> represents the weights between layer <span class="math notranslate nohighlight">\(i\)</span> and layer
<span class="math notranslate nohighlight">\(i+1\)</span>. <code class="docutils literal notranslate"><span class="pre">intercepts_</span></code> is a list of bias vectors, where the vector
at index <span class="math notranslate nohighlight">\(i\)</span> represents the bias values added to layer <span class="math notranslate nohighlight">\(i+1\)</span>.</p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Advantages and disadvantages of Multi-layer Perceptron</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 advantages of Multi-layer Perceptron are:</p>
<ul class="simple">
<li><p>Capability to learn non-linear models.</p></li>
<li><p>Capability to learn models in real-time (on-line learning)
using <code class="docutils literal notranslate"><span class="pre">partial_fit</span></code>.</p></li>
</ul>
<p>The disadvantages of Multi-layer Perceptron (MLP) include:</p>
<ul class="simple">
<li><p>MLP with hidden layers have a non-convex loss function where there exists
more than one local minimum. Therefore different random weight
initializations can lead to different validation accuracy.</p></li>
<li><p>MLP requires tuning a number of hyperparameters such as the number of
hidden neurons, layers, and iterations.</p></li>
<li><p>MLP is sensitive to feature scaling.</p></li>
</ul>
<p>Please see <a class="reference internal" href="#mlp-tips"><span class="std std-ref">Tips on Practical Use</span></a> section that addresses
some of these disadvantages.</p>
<p></div>
</details></p>
</section>
<section id="classification">
<h2><span class="section-number">1.17.2. </span>Classification<a class="headerlink" href="#classification" title="Link to this heading">¶</a></h2>
<p>Class <a class="reference internal" href="generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier" title="sklearn.neural_network.MLPClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPClassifier</span></code></a> implements a multi-layer perceptron (MLP) algorithm
that trains using <a class="reference external" href="http://ufldl.stanford.edu/wiki/index.php/Backpropagation_Algorithm">Backpropagation</a>.</p>
<p>MLP trains on two arrays: array X of size (n_samples, n_features), which holds
the training samples represented as floating point feature vectors; and array
y of size (n_samples,), which holds the target values (class labels) for the
training samples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.neural_network</span> <span class="kn">import</span> <span class="n">MLPClassifier</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mf">0.</span><span class="p">,</span> <span class="mf">0.</span><span class="p">],</span> <span class="p">[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">MLPClassifier</span><span class="p">(</span><span class="n">solver</span><span class="o">=</span><span class="s1">'lbfgs'</span><span class="p">,</span> <span class="n">alpha</span><span class="o">=</span><span class="mf">1e-5</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">hidden_layer_sizes</span><span class="o">=</span><span class="p">(</span><span class="mi">5</span><span class="p">,</span> <span class="mi">2</span><span class="p">),</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">...</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">MLPClassifier(alpha=1e-05, hidden_layer_sizes=(5, 2), random_state=1,</span>
<span class="go"> solver='lbfgs')</span>
</pre></div>
</div>
<p>After fitting (training), the model can predict labels for new samples:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mf">2.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">],</span> <span class="p">[</span><span class="o">-</span><span class="mf">1.</span><span class="p">,</span> <span class="o">-</span><span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([1, 0])</span>
</pre></div>
</div>
<p>MLP can fit a non-linear model to the training data. <code class="docutils literal notranslate"><span class="pre">clf.coefs_</span></code>
contains the weight matrices that constitute the model parameters:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="p">[</span><span class="n">coef</span><span class="o">.</span><span class="n">shape</span> <span class="k">for</span> <span class="n">coef</span> <span class="ow">in</span> <span class="n">clf</span><span class="o">.</span><span class="n">coefs_</span><span class="p">]</span>
<span class="go">[(2, 5), (5, 2), (2, 1)]</span>
</pre></div>
</div>
<p>Currently, <a class="reference internal" href="generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier" title="sklearn.neural_network.MLPClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPClassifier</span></code></a> supports only the
Cross-Entropy loss function, which allows probability estimates by running the
<code class="docutils literal notranslate"><span class="pre">predict_proba</span></code> method.</p>
<p>MLP trains using Backpropagation. More precisely, it trains using some form of
gradient descent and the gradients are calculated using Backpropagation. For
classification, it minimizes the Cross-Entropy loss function, giving a vector
of probability estimates <span class="math notranslate nohighlight">\(P(y|x)\)</span> per sample <span class="math notranslate nohighlight">\(x\)</span>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict_proba</span><span class="p">([[</span><span class="mf">2.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">],</span> <span class="p">[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([[1.967...e-04, 9.998...-01],</span>
<span class="go"> [1.967...e-04, 9.998...-01]])</span>
</pre></div>
</div>
<p><a class="reference internal" href="generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier" title="sklearn.neural_network.MLPClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPClassifier</span></code></a> supports multi-class classification by
applying <a class="reference external" href="https://en.wikipedia.org/wiki/Softmax_activation_function">Softmax</a>
as the output function.</p>
<p>Further, the model supports <a class="reference internal" href="multiclass.html#multiclass"><span class="std std-ref">multi-label classification</span></a>
in which a sample can belong to more than one class. For each class, the raw
output passes through the logistic function. Values larger or equal to <code class="docutils literal notranslate"><span class="pre">0.5</span></code>
are rounded to <code class="docutils literal notranslate"><span class="pre">1</span></code>, otherwise to <code class="docutils literal notranslate"><span class="pre">0</span></code>. For a predicted output of a sample, the
indices where the value is <code class="docutils literal notranslate"><span class="pre">1</span></code> represents the assigned classes of that sample:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mf">0.</span><span class="p">,</span> <span class="mf">0.</span><span class="p">],</span> <span class="p">[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">],</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">MLPClassifier</span><span class="p">(</span><span class="n">solver</span><span class="o">=</span><span class="s1">'lbfgs'</span><span class="p">,</span> <span class="n">alpha</span><span class="o">=</span><span class="mf">1e-5</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">hidden_layer_sizes</span><span class="o">=</span><span class="p">(</span><span class="mi">15</span><span class="p">,),</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
<span class="gp">...</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="go">MLPClassifier(alpha=1e-05, hidden_layer_sizes=(15,), random_state=1,</span>
<span class="go"> solver='lbfgs')</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([[1, 1]])</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mf">0.</span><span class="p">,</span> <span class="mf">0.</span><span class="p">]])</span>
<span class="go">array([[0, 1]])</span>
</pre></div>
</div>
<p>See the examples below and the docstring of
<a class="reference internal" href="generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier.fit" title="sklearn.neural_network.MLPClassifier.fit"><code class="xref py py-meth docutils literal notranslate"><span class="pre">MLPClassifier.fit</span></code></a> for further information.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/neural_networks/plot_mlp_training_curves.html#sphx-glr-auto-examples-neural-networks-plot-mlp-training-curves-py"><span class="std std-ref">Compare Stochastic learning strategies for MLPClassifier</span></a></p></li>
<li><p><a class="reference internal" href="../auto_examples/neural_networks/plot_mnist_filters.html#sphx-glr-auto-examples-neural-networks-plot-mnist-filters-py"><span class="std std-ref">Visualization of MLP weights on MNIST</span></a></p></li>
</ul>
</aside>
</section>
<section id="regression">
<h2><span class="section-number">1.17.3. </span>Regression<a class="headerlink" href="#regression" title="Link to this heading">¶</a></h2>
<p>Class <a class="reference internal" href="generated/sklearn.neural_network.MLPRegressor.html#sklearn.neural_network.MLPRegressor" title="sklearn.neural_network.MLPRegressor"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPRegressor</span></code></a> implements a multi-layer perceptron (MLP) that
trains using backpropagation with no activation function in the output layer,
which can also be seen as using the identity function as activation function.
Therefore, it uses the square error as the loss function, and the output is a
set of continuous values.</p>
<p><a class="reference internal" href="generated/sklearn.neural_network.MLPRegressor.html#sklearn.neural_network.MLPRegressor" title="sklearn.neural_network.MLPRegressor"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPRegressor</span></code></a> also supports multi-output regression, in
which a sample can have more than one target.</p>
</section>
<section id="regularization">
<h2><span class="section-number">1.17.4. </span>Regularization<a class="headerlink" href="#regularization" title="Link to this heading">¶</a></h2>
<p>Both <a class="reference internal" href="generated/sklearn.neural_network.MLPRegressor.html#sklearn.neural_network.MLPRegressor" title="sklearn.neural_network.MLPRegressor"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPRegressor</span></code></a> and <a class="reference internal" href="generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier" title="sklearn.neural_network.MLPClassifier"><code class="xref py py-class docutils literal notranslate"><span class="pre">MLPClassifier</span></code></a> use parameter <code class="docutils literal notranslate"><span class="pre">alpha</span></code>
for regularization (L2 regularization) term which helps in avoiding overfitting
by penalizing weights with large magnitudes. Following plot displays varying
decision function with value of alpha.</p>
<figure class="align-center">
<a class="reference external image-reference" href="../auto_examples/neural_networks/plot_mlp_alpha.html"><img alt="../_images/sphx_glr_plot_mlp_alpha_001.png" src="../_images/sphx_glr_plot_mlp_alpha_001.png" style="width: 1275.0px; height: 675.0px;" /></a>
</figure>
<p>See the examples below for further information.</p>
<aside class="topic">
<p class="topic-title">Examples:</p>
<ul class="simple">
<li><p><a class="reference internal" href="../auto_examples/neural_networks/plot_mlp_alpha.html#sphx-glr-auto-examples-neural-networks-plot-mlp-alpha-py"><span class="std std-ref">Varying regularization in Multi-layer Perceptron</span></a></p></li>
</ul>
</aside>
</section>
<section id="algorithms">
<h2><span class="section-number">1.17.5. </span>Algorithms<a class="headerlink" href="#algorithms" title="Link to this heading">¶</a></h2>
<p>MLP trains using <a class="reference external" href="https://en.wikipedia.org/wiki/Stochastic_gradient_descent">Stochastic Gradient Descent</a>,
<a class="reference external" href="https://arxiv.org/abs/1412.6980">Adam</a>, or
<a class="reference external" href="https://en.wikipedia.org/wiki/Limited-memory_BFGS">L-BFGS</a>.
Stochastic Gradient Descent (SGD) updates parameters using the gradient of the
loss function with respect to a parameter that needs adaptation, i.e.</p>
<div class="math notranslate nohighlight">
\[w \leftarrow w - \eta (\alpha \frac{\partial R(w)}{\partial w}
+ \frac{\partial Loss}{\partial w})\]</div>
<p>where <span class="math notranslate nohighlight">\(\eta\)</span> is the learning rate which controls the step-size in
the parameter space search. <span class="math notranslate nohighlight">\(Loss\)</span> is the loss function used
for the network.</p>
<p>More details can be found in the documentation of
<a class="reference external" href="https://scikit-learn.org/stable/modules/sgd.html">SGD</a></p>
<p>Adam is similar to SGD in a sense that it is a stochastic optimizer, but it can
automatically adjust the amount to update parameters based on adaptive estimates
of lower-order moments.</p>
<p>With SGD or Adam, training supports online and mini-batch learning.</p>
<p>L-BFGS is a solver that approximates the Hessian matrix which represents the
second-order partial derivative of a function. Further it approximates the
inverse of the Hessian matrix to perform parameter updates. The implementation
uses the Scipy version of <a class="reference external" href="https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.fmin_l_bfgs_b.html">L-BFGS</a>.</p>
<p>If the selected solver is ‘L-BFGS’, training does not support online nor
mini-batch learning.</p>
</section>
<section id="complexity">
<h2><span class="section-number">1.17.6. </span>Complexity<a class="headerlink" href="#complexity" title="Link to this heading">¶</a></h2>
<p>Suppose there are <span class="math notranslate nohighlight">\(n\)</span> training samples, <span class="math notranslate nohighlight">\(m\)</span> features, <span class="math notranslate nohighlight">\(k\)</span>
hidden layers, each containing <span class="math notranslate nohighlight">\(h\)</span> neurons - for simplicity, and <span class="math notranslate nohighlight">\(o\)</span>
output neurons. The time complexity of backpropagation is
<span class="math notranslate nohighlight">\(O(n\cdot m \cdot h^k \cdot o \cdot i)\)</span>, where <span class="math notranslate nohighlight">\(i\)</span> is the number
of iterations. Since backpropagation has a high time complexity, it is advisable
to start with smaller number of hidden neurons and few hidden layers for
training.</p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
Mathematical formulation
<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>Given a set of training examples <span class="math notranslate nohighlight">\((x_1, y_1), (x_2, y_2), \ldots, (x_n, y_n)\)</span>
where <span class="math notranslate nohighlight">\(x_i \in \mathbf{R}^n\)</span> and <span class="math notranslate nohighlight">\(y_i \in \{0, 1\}\)</span>, a one hidden
layer one hidden neuron MLP learns the function <span class="math notranslate nohighlight">\(f(x) = W_2 g(W_1^T x + b_1) + b_2\)</span>
where <span class="math notranslate nohighlight">\(W_1 \in \mathbf{R}^m\)</span> and <span class="math notranslate nohighlight">\(W_2, b_1, b_2 \in \mathbf{R}\)</span> are
model parameters. <span class="math notranslate nohighlight">\(W_1, W_2\)</span> represent the weights of the input layer and
hidden layer, respectively; and <span class="math notranslate nohighlight">\(b_1, b_2\)</span> represent the bias added to
the hidden layer and the output layer, respectively.
<span class="math notranslate nohighlight">\(g(\cdot) : R \rightarrow R\)</span> is the activation function, set by default as
the hyperbolic tan. It is given as,</p>
<div class="math notranslate nohighlight">
\[g(z)= \frac{e^z-e^{-z}}{e^z+e^{-z}}\]</div>
<p>For binary classification, <span class="math notranslate nohighlight">\(f(x)\)</span> passes through the logistic function
<span class="math notranslate nohighlight">\(g(z)=1/(1+e^{-z})\)</span> to obtain output values between zero and one. A
threshold, set to 0.5, would assign samples of outputs larger or equal 0.5
to the positive class, and the rest to the negative class.</p>
<p>If there are more than two classes, <span class="math notranslate nohighlight">\(f(x)\)</span> itself would be a vector of
size (n_classes,). Instead of passing through logistic function, it passes
through the softmax function, which is written as,</p>
<div class="math notranslate nohighlight">
\[\text{softmax}(z)_i = \frac{\exp(z_i)}{\sum_{l=1}^k\exp(z_l)}\]</div>
<p>where <span class="math notranslate nohighlight">\(z_i\)</span> represents the <span class="math notranslate nohighlight">\(i\)</span> th element of the input to softmax,
which corresponds to class <span class="math notranslate nohighlight">\(i\)</span>, and <span class="math notranslate nohighlight">\(K\)</span> is the number of classes.
The result is a vector containing the probabilities that sample <span class="math notranslate nohighlight">\(x\)</span>
belong to each class. The output is the class with the highest probability.</p>
<p>In regression, the output remains as <span class="math notranslate nohighlight">\(f(x)\)</span>; therefore, output activation
function is just the identity function.</p>
<p>MLP uses different loss functions depending on the problem type. The loss
function for classification is Average Cross-Entropy, which in binary case is
given as,</p>
<div class="math notranslate nohighlight">
\[Loss(\hat{y},y,W) = -\dfrac{1}{n}\sum_{i=0}^n(y_i \ln {\hat{y_i}} + (1-y_i) \ln{(1-\hat{y_i})}) + \dfrac{\alpha}{2n} ||W||_2^2\]</div>
<p>where <span class="math notranslate nohighlight">\(\alpha ||W||_2^2\)</span> is an L2-regularization term (aka penalty)
that penalizes complex models; and <span class="math notranslate nohighlight">\(\alpha > 0\)</span> is a non-negative
hyperparameter that controls the magnitude of the penalty.</p>
<p>For regression, MLP uses the Mean Square Error loss function; written as,</p>
<div class="math notranslate nohighlight">
\[Loss(\hat{y},y,W) = \frac{1}{2n}\sum_{i=0}^n||\hat{y}_i - y_i ||_2^2 + \frac{\alpha}{2n} ||W||_2^2\]</div>
<p>Starting from initial random weights, multi-layer perceptron (MLP) minimizes
the loss function by repeatedly updating these weights. After computing the
loss, a backward pass propagates it from the output layer to the previous
layers, providing each weight parameter with an update value meant to decrease
the loss.</p>
<p>In gradient descent, the gradient <span class="math notranslate nohighlight">\(\nabla Loss_{W}\)</span> of the loss with respect
to the weights is computed and deducted from <span class="math notranslate nohighlight">\(W\)</span>.
More formally, this is expressed as,</p>
<div class="math notranslate nohighlight">
\[W^{i+1} = W^i - \epsilon \nabla {Loss}_{W}^{i}\]</div>
<p>where <span class="math notranslate nohighlight">\(i\)</span> is the iteration step, and <span class="math notranslate nohighlight">\(\epsilon\)</span> is the learning rate
with a value larger than 0.</p>
<p>The algorithm stops when it reaches a preset maximum number of iterations; or
when the improvement in loss is below a certain, small number.</p>
<p></div>
</details></p>
</section>
<section id="tips-on-practical-use">
<span id="mlp-tips"></span><h2><span class="section-number">1.17.7. </span>Tips on Practical Use<a class="headerlink" href="#tips-on-practical-use" title="Link to this heading">¶</a></h2>
<ul>
<li><p>Multi-layer Perceptron is sensitive to feature scaling, so it
is highly recommended to scale your data. For example, scale each
attribute on the input vector X to [0, 1] or [-1, +1], or standardize
it to have mean 0 and variance 1. Note that you must apply the <em>same</em>
scaling to the test set for meaningful results.
You can use <a class="reference internal" href="generated/sklearn.preprocessing.StandardScaler.html#sklearn.preprocessing.StandardScaler" title="sklearn.preprocessing.StandardScaler"><code class="xref py py-class docutils literal notranslate"><span class="pre">StandardScaler</span></code></a> for standardization.</p>
<div class="doctest highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.preprocessing</span> <span class="kn">import</span> <span class="n">StandardScaler</span>
<span class="gp">>>> </span><span class="n">scaler</span> <span class="o">=</span> <span class="n">StandardScaler</span><span class="p">()</span>
<span class="gp">>>> </span><span class="c1"># Don't cheat - fit only on training data</span>
<span class="gp">>>> </span><span class="n">scaler</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X_train</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">X_train</span> <span class="o">=</span> <span class="n">scaler</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">X_train</span><span class="p">)</span>
<span class="gp">>>> </span><span class="c1"># apply same transformation to test data</span>
<span class="gp">>>> </span><span class="n">X_test</span> <span class="o">=</span> <span class="n">scaler</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">X_test</span><span class="p">)</span>
</pre></div>
</div>
<p>An alternative and recommended approach is to use
<a class="reference internal" href="generated/sklearn.preprocessing.StandardScaler.html#sklearn.preprocessing.StandardScaler" title="sklearn.preprocessing.StandardScaler"><code class="xref py py-class docutils literal notranslate"><span class="pre">StandardScaler</span></code></a> in a
<a class="reference internal" href="generated/sklearn.pipeline.Pipeline.html#sklearn.pipeline.Pipeline" title="sklearn.pipeline.Pipeline"><code class="xref py py-class docutils literal notranslate"><span class="pre">Pipeline</span></code></a></p>
</li>
<li><p>Finding a reasonable regularization parameter <span class="math notranslate nohighlight">\(\alpha\)</span> is best done
using <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 notranslate"><span class="pre">GridSearchCV</span></code></a>, usually in the range
<code class="docutils literal notranslate"><span class="pre">10.0</span> <span class="pre">**</span> <span class="pre">-np.arange(1,</span> <span class="pre">7)</span></code>.</p></li>
<li><p>Empirically, we observed that <code class="docutils literal notranslate"><span class="pre">L-BFGS</span></code> converges faster and
with better solutions on small datasets. For relatively large
datasets, however, <code class="docutils literal notranslate"><span class="pre">Adam</span></code> is very robust. It usually converges
quickly and gives pretty good performance. <code class="docutils literal notranslate"><span class="pre">SGD</span></code> with momentum or
nesterov’s momentum, on the other hand, can perform better than
those two algorithms if learning rate is correctly tuned.</p></li>
</ul>
</section>
<section id="more-control-with-warm-start">
<h2><span class="section-number">1.17.8. </span>More control with warm_start<a class="headerlink" href="#more-control-with-warm-start" title="Link to this heading">¶</a></h2>
<p>If you want more control over stopping criteria or learning rate in SGD,
or want to do additional monitoring, using <code class="docutils literal notranslate"><span class="pre">warm_start=True</span></code> and
<code class="docutils literal notranslate"><span class="pre">max_iter=1</span></code> and iterating yourself can be helpful:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mf">0.</span><span class="p">,</span> <span class="mf">0.</span><span class="p">],</span> <span class="p">[</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">1.</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">MLPClassifier</span><span class="p">(</span><span class="n">hidden_layer_sizes</span><span class="o">=</span><span class="p">(</span><span class="mi">15</span><span class="p">,),</span> <span class="n">random_state</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">max_iter</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">warm_start</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">10</span><span class="p">):</span>
<span class="gp">... </span> <span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="gp">... </span> <span class="c1"># additional monitoring / inspection</span>
<span class="go">MLPClassifier(...</span>
</pre></div>
</div>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>References</strong>
<span class="tooltiptext">Click for more details</span>
<a class="headerlink" href="#summary-anchor" title="Permalink to this heading">¶</a>
</summary>
<div class="card"></p>
<blockquote>
<div><ul class="simple">
<li><p><a class="reference external" href="https://www.iro.umontreal.ca/~pift6266/A06/refs/backprop_old.pdf">“Learning representations by back-propagating errors.”</a>
Rumelhart, David E., Geoffrey E. Hinton, and Ronald J. Williams.</p></li>
<li><p><a class="reference external" href="https://leon.bottou.org/projects/sgd">“Stochastic Gradient Descent”</a> L. Bottou - Website, 2010.</p></li>
<li><p><a class="reference external" href="http://ufldl.stanford.edu/wiki/index.php/Backpropagation_Algorithm">“Backpropagation”</a>
Andrew Ng, Jiquan Ngiam, Chuan Yu Foo, Yifan Mai, Caroline Suen - Website, 2011.</p></li>
<li><p><a class="reference external" href="http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf">“Efficient BackProp”</a>
Y. LeCun, L. Bottou, G. Orr, K. Müller - In Neural Networks: Tricks
of the Trade 1998.</p></li>
<li><p><a class="reference external" href="https://arxiv.org/abs/1412.6980">“Adam: A method for stochastic optimization.”</a>
Kingma, Diederik, and Jimmy Ba (2014)</p></li>
</ul>
</div></blockquote>
<p></div>
</details></p>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2024, scikit-learn developers (BSD License).
<a href="../_sources/modules/neural_networks_supervised.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>