-
Notifications
You must be signed in to change notification settings - Fork 81
/
Copy pathmodel_persistence.html
603 lines (529 loc) · 41.7 KB
/
model_persistence.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
<!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="9. Model persistence" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://scikit-learn/stable/model_persistence.html" />
<meta property="og:site_name" content="scikit-learn" />
<meta property="og:description" content="After training a scikit-learn model, it is desirable to have a way to persist the model for future use without having to retrain. This can be accomplished using pickle, joblib, skops, ONNX, or PMML..." />
<meta property="og:image" content="https://scikit-learn.org/stable/_static/scikit-learn-logo-small.png" />
<meta property="og:image:alt" content="scikit-learn" />
<meta name="description" content="After training a scikit-learn model, it is desirable to have a way to persist the model for future use without having to retrain. This can be accomplished using pickle, joblib, skops, ONNX, or PMML..." />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>9. Model persistence — scikit-learn 1.6.dev0 documentation</title>
<link rel="canonical" href="https://scikit-learn.org/stable/model_persistence.html" />
<link rel="shortcut icon" href="_static/favicon.ico"/>
<link rel="stylesheet" href="_static/css/vendor/bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="_static/plot_directive.css" type="text/css" />
<link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Vibur" type="text/css" />
<link rel="stylesheet" href="_static/jupyterlite_sphinx.css" type="text/css" />
<link rel="stylesheet" href="_static/sg_gallery.css" type="text/css" />
<link rel="stylesheet" href="_static/sg_gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="_static/sg_gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="_static/sg_gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<script id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/js/vendor/jquery-3.6.3.slim.min.js"></script>
<script src="_static/js/details-permalink.js"></script>
</head>
<body>
<nav id="navbar" class="sk-docs-navbar navbar navbar-expand-md navbar-light bg-light py-0">
<div class="container-fluid sk-docs-container px-0">
<a class="navbar-brand py-0" href="index.html">
<img
class="sk-brand-img"
src="_static/scikit-learn-logo-small.png"
alt="logo"/>
</a>
<button
id="sk-navbar-toggler"
class="navbar-toggler"
type="button"
data-toggle="collapse"
data-target="#navbarSupportedContent"
aria-controls="navbarSupportedContent"
aria-expanded="false"
aria-label="Toggle navigation"
>
<span class="navbar-toggler-icon"></span>
</button>
<div class="sk-navbar-collapse collapse navbar-collapse" id="navbarSupportedContent">
<ul class="navbar-nav mr-auto">
<li class="nav-item">
<a class="sk-nav-link nav-link" href="install.html">Install</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="user_guide.html">User Guide</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="modules/classes.html">API</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" href="auto_examples/index.html">Examples</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link" target="_blank" rel="noopener noreferrer" href="https://blog.scikit-learn.org/">Community</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="getting_started.html" >Getting Started</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="tutorial/index.html" >Tutorial</a>
</li>
<li class="nav-item">
<a class="sk-nav-link nav-link nav-more-item-mobile-items" href="whats_new/v1.6.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/v1.6.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="computing/parallelism.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="8.3. Parallelism, resource management, and configuration">Prev</a><a href="user_guide.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="User Guide">Up</a>
<a href="common_pitfalls.html" role="button" class="btn sk-btn-rellink py-1" sk-rellink-tooltip="10. Common pitfalls and recommended practices">Next</a>
</div>
<div class="alert alert-danger p-1 mb-2" role="alert">
<p class="text-center mb-0">
<strong>scikit-learn 1.6.dev0</strong><br/>
<a href="https://scikit-learn.org/dev/versions.html">Other versions</a>
</p>
</div>
<div class="alert alert-warning p-1 mb-2" role="alert">
<p class="text-center mb-0">
Please <a class="font-weight-bold" href="about.html#citing-scikit-learn"><string>cite us</string></a> if you use the software.
</p>
</div>
<div class="sk-sidebar-toc">
<ul>
<li>
<a href="user_guide.html" class="sk-toc-active">User Guide</a>
</li>
<ul>
<li>
<a href="supervised_learning.html" class="">1. Supervised learning</a>
</li>
<li>
<a href="unsupervised_learning.html" class="">2. Unsupervised learning</a>
</li>
<li>
<a href="model_selection.html" class="">3. Model selection and evaluation</a>
</li>
<li>
<a href="inspection.html" class="">4. Inspection</a>
</li>
<li>
<a href="visualizations.html" class="">5. Visualizations</a>
</li>
<li>
<a href="data_transforms.html" class="">6. Dataset transformations</a>
</li>
<li>
<a href="datasets.html" class="">7. Dataset loading utilities</a>
</li>
<li>
<a href="computing.html" class="">8. Computing with scikit-learn</a>
</li>
<li>
<a href="" class="sk-toc-active">9. Model persistence</a>
</li>
<li>
<a href="common_pitfalls.html" class="">10. Common pitfalls and recommended practices</a>
</li>
<li>
<a href="dispatching.html" class="">11. Dispatching</a>
</li>
</ul>
</ul>
</div>
</div>
</div>
<div id="sk-page-content-wrapper">
<div class="sk-page-content container-fluid body px-md-3" role="main">
<section id="model-persistence">
<span id="id1"></span><h1><span class="section-number">9. </span>Model persistence<a class="headerlink" href="#model-persistence" title="Link to this heading">¶</a></h1>
<p>After training a scikit-learn model, it is desirable to have a way to persist
the model for future use without having to retrain. This can be accomplished
using <a class="reference external" href="https://docs.python.org/3/library/pickle.html">pickle</a>, <a class="reference external" href="https://joblib.readthedocs.io/en/stable/">joblib</a>, <a class="reference external" href="https://skops.readthedocs.io/en/stable/">skops</a>, <a class="reference external" href="https://onnx.ai/">ONNX</a>,
or <a class="reference external" href="https://dmg.org/pmml/v4-4-1/GeneralStructure.html">PMML</a>. In most cases
<code class="docutils literal notranslate"><span class="pre">pickle</span></code> can be used to persist a trained scikit-learn model. Once all
transitive scikit-learn dependencies have been pinned, the trained model can
then be loaded and executed under conditions similar to those in which it was
originally pinned. The following sections will give you some hints on how to
persist a scikit-learn model and will provide details on what each alternative
can offer.</p>
<section id="workflow-overview">
<h2><span class="section-number">9.1. </span>Workflow Overview<a class="headerlink" href="#workflow-overview" title="Link to this heading">¶</a></h2>
<p>In this section we present a general workflow on how to persist a
scikit-learn model. We will demonstrate this with a simple example using
Python’s built-in persistence module, namely <a class="reference external" href="https://docs.python.org/3/library/pickle.html">pickle</a>.</p>
<section id="storing-the-model-in-an-artifact">
<h3><span class="section-number">9.1.1. </span>Storing the model in an artifact<a class="headerlink" href="#storing-the-model-in-an-artifact" title="Link to this heading">¶</a></h3>
<p>Once the model training process in completed, the trained model can be stored
as an artifact with the help of <code class="docutils literal notranslate"><span class="pre">pickle</span></code>. The model can be saved using the
process of serialization, where the Python object hierarchy is converted into
a byte stream. We can persist a trained model in the following manner:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">svm</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">datasets</span>
<span class="gp">>>> </span><span class="kn">import</span> <span class="nn">pickle</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">svm</span><span class="o">.</span><span class="n">SVC</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">X</span><span class="p">,</span> <span class="n">y</span> <span class="o">=</span> <span class="n">datasets</span><span class="o">.</span><span class="n">load_iris</span><span class="p">(</span><span class="n">return_X_y</span><span class="o">=</span><span class="kc">True</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="go">SVC()</span>
<span class="gp">>>> </span><span class="n">s</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">clf</span><span class="p">)</span>
</pre></div>
</div>
</section>
<section id="replicating-the-training-environment-in-production">
<h3><span class="section-number">9.1.2. </span>Replicating the training environment in production<a class="headerlink" href="#replicating-the-training-environment-in-production" title="Link to this heading">¶</a></h3>
<p>The versions of the dependencies used may differ from training to production.
This may result in unexpected behaviour and errors while using the trained
model. To prevent such situations it is recommended to use the same
dependencies and versions in both the training and production environment.
These transitive dependencies can be pinned with the help of <code class="docutils literal notranslate"><span class="pre">pip</span></code>, <code class="docutils literal notranslate"><span class="pre">conda</span></code>,
<code class="docutils literal notranslate"><span class="pre">poetry</span></code>, <code class="docutils literal notranslate"><span class="pre">conda-lock</span></code>, <code class="docutils literal notranslate"><span class="pre">pixi</span></code>, etc.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>To execute a pickled scikit-learn model in a reproducible environment it is
advisable to pin all transitive scikit-learn dependencies. This prevents
any incompatibility issues that may arise while trying to load the pickled
model. You can read more about persisting models with <code class="docutils literal notranslate"><span class="pre">pickle</span></code> over
<a class="reference internal" href="#persisting-models-with-pickle"><span class="std std-ref">here</span></a>.</p>
</div>
</section>
<section id="loading-the-model-artifact">
<h3><span class="section-number">9.1.3. </span>Loading the model artifact<a class="headerlink" href="#loading-the-model-artifact" title="Link to this heading">¶</a></h3>
<p>The saved scikit-learn model can be loaded using <code class="docutils literal notranslate"><span class="pre">pickle</span></code> for future use
without having to re-train the entire model from scratch. The saved model
artifact can be unpickled by converting the byte stream into an object
hierarchy. This can be done with the help of <code class="docutils literal notranslate"><span class="pre">pickle</span></code> as follows:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf2</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="n">s</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf2</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X</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="go">array([0])</span>
<span class="gp">>>> </span><span class="n">y</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="go">0</span>
</pre></div>
</div>
</section>
<section id="serving-the-model-artifact">
<h3><span class="section-number">9.1.4. </span>Serving the model artifact<a class="headerlink" href="#serving-the-model-artifact" title="Link to this heading">¶</a></h3>
<p>The last step after training a scikit-learn model is serving the model.
Once the trained model is successfully loaded it can be served to manage
different prediction requests. This can involve deploying the model as a
web service using containerization, or other model deployment strategies,
according to the specifications. In the next sections, we will explore
different approaches to persist a trained scikit-learn model.</p>
</section>
</section>
<section id="persisting-models-with-pickle">
<span id="id3"></span><h2><span class="section-number">9.2. </span>Persisting models with pickle<a class="headerlink" href="#persisting-models-with-pickle" title="Link to this heading">¶</a></h2>
<p>As demonstrated in the previous section, <code class="docutils literal notranslate"><span class="pre">pickle</span></code> uses serialization and
deserialization to persist scikit-learn models. Instead of using <code class="docutils literal notranslate"><span class="pre">dumps</span></code> and
<code class="docutils literal notranslate"><span class="pre">loads</span></code>, <code class="docutils literal notranslate"><span class="pre">dump</span></code> and <code class="docutils literal notranslate"><span class="pre">load</span></code> can also be used in the following way:</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.tree</span> <span class="kn">import</span> <span class="n">DecisionTreeClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="kn">import</span> <span class="n">datasets</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">DecisionTreeClassifier</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">X</span><span class="p">,</span> <span class="n">y</span> <span class="o">=</span> <span class="n">datasets</span><span class="o">.</span><span class="n">load_iris</span><span class="p">(</span><span class="n">return_X_y</span><span class="o">=</span><span class="kc">True</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="go">DecisionTreeClassifier()</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">pickle</span> <span class="kn">import</span> <span class="n">dump</span><span class="p">,</span> <span class="n">load</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="s1">'filename.pkl'</span><span class="p">,</span> <span class="s1">'wb'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span> <span class="n">dump</span><span class="p">(</span><span class="n">clf</span><span class="p">,</span> <span class="n">f</span><span class="p">)</span>
<span class="gp">>>> </span><span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="s1">'filename.pkl'</span><span class="p">,</span> <span class="s1">'rb'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span> <span class="n">clf2</span> <span class="o">=</span> <span class="n">load</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf2</span><span class="o">.</span><span class="n">predict</span><span class="p">(</span><span class="n">X</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="go">array([0])</span>
<span class="gp">>>> </span><span class="n">y</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="go">0</span>
</pre></div>
</div>
<p>For applications that involve writing and loading the serialized object to or
from a file, <code class="docutils literal notranslate"><span class="pre">dump</span></code> and <code class="docutils literal notranslate"><span class="pre">load</span></code> can be used instead of <code class="docutils literal notranslate"><span class="pre">dumps</span></code> and <code class="docutils literal notranslate"><span class="pre">loads</span></code>. When
file operations are not required the pickled representation of the object can
be returned as a bytes object with the help of the <code class="docutils literal notranslate"><span class="pre">dumps</span></code> function. The
reconstituted object hierarchy of the pickled data can then be returned using
the <code class="docutils literal notranslate"><span class="pre">loads</span></code> function.</p>
</section>
<section id="persisting-models-with-joblib">
<h2><span class="section-number">9.3. </span>Persisting models with joblib<a class="headerlink" href="#persisting-models-with-joblib" title="Link to this heading">¶</a></h2>
<p>In the specific case of scikit-learn, it may be better to use joblib’s
replacement of pickle (<code class="docutils literal notranslate"><span class="pre">dump</span></code> & <code class="docutils literal notranslate"><span class="pre">load</span></code>), which is more efficient on
objects that carry large numpy arrays internally as is often the case for
fitted scikit-learn estimators, but can only pickle to the disk and not to a
string:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">joblib</span> <span class="kn">import</span> <span class="n">dump</span><span class="p">,</span> <span class="n">load</span>
<span class="gp">>>> </span><span class="n">dump</span><span class="p">(</span><span class="n">clf</span><span class="p">,</span> <span class="s1">'filename.joblib'</span><span class="p">)</span>
</pre></div>
</div>
<p>Later you can load back the pickled model (possibly in another Python process)
with:</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">load</span><span class="p">(</span><span class="s1">'filename.joblib'</span><span class="p">)</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><code class="docutils literal notranslate"><span class="pre">dump</span></code> and <code class="docutils literal notranslate"><span class="pre">load</span></code> functions also accept file-like object
instead of filenames. More information on data persistence with Joblib is
available <a class="reference external" href="https://joblib.readthedocs.io/en/latest/persistence.html">here</a>.</p>
</div>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>InconsistentVersionWarning</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>When an estimator is unpickled with a scikit-learn version that is inconsistent
with the version the estimator was pickled with, a
<a class="reference internal" href="modules/generated/sklearn.exceptions.InconsistentVersionWarning.html#sklearn.exceptions.InconsistentVersionWarning" title="sklearn.exceptions.InconsistentVersionWarning"><code class="xref py py-class docutils literal notranslate"><span class="pre">InconsistentVersionWarning</span></code></a> is raised. This warning
can be caught to obtain the original version the estimator was pickled with:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">sklearn.exceptions</span> <span class="kn">import</span> <span class="n">InconsistentVersionWarning</span>
<span class="n">warnings</span><span class="o">.</span><span class="n">simplefilter</span><span class="p">(</span><span class="s2">"error"</span><span class="p">,</span> <span class="n">InconsistentVersionWarning</span><span class="p">)</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">est</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="s2">"model_from_prevision_version.pickle"</span><span class="p">)</span>
<span class="k">except</span> <span class="n">InconsistentVersionWarning</span> <span class="k">as</span> <span class="n">w</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="n">w</span><span class="o">.</span><span class="n">original_sklearn_version</span><span class="p">)</span>
</pre></div>
</div>
<p></div>
</details></p>
</section>
<section id="security-maintainability-limitations-for-pickle-and-joblib">
<span id="persistence-limitations"></span><h2><span class="section-number">9.4. </span>Security & maintainability limitations for pickle and joblib<a class="headerlink" href="#security-maintainability-limitations-for-pickle-and-joblib" title="Link to this heading">¶</a></h2>
<p>pickle (and joblib by extension), has some issues regarding maintainability
and security. Because of this,</p>
<ul class="simple">
<li><p>Never unpickle untrusted data as it could lead to malicious code being
executed upon loading.</p></li>
<li><p>While models saved using one version of scikit-learn might load in
other versions, this is entirely unsupported and inadvisable. It should
also be kept in mind that operations performed on such data could give
different and unexpected results.</p></li>
</ul>
<p>In order to rebuild a similar model with future versions of scikit-learn,
additional metadata should be saved along the pickled model:</p>
<ul class="simple">
<li><p>The training data, e.g. a reference to an immutable snapshot</p></li>
<li><p>The python source code used to generate the model</p></li>
<li><p>The versions of scikit-learn and its dependencies</p></li>
<li><p>The cross validation score obtained on the training data</p></li>
</ul>
<p>This should make it possible to check that the cross-validation score is in the
same range as before.</p>
<p>Aside for a few exceptions, pickled models should be portable across
architectures assuming the same versions of dependencies and Python are used.
If you encounter an estimator that is not portable please open an issue on
GitHub. Pickled models are often deployed in production using containers, like
Docker, in order to freeze the environment and dependencies.</p>
<p>If you want to know more about these issues and explore other possible
serialization methods, please refer to this
<a class="reference external" href="https://pyvideo.org/video/2566/pickles-are-for-delis-not-software">talk by Alex Gaynor</a>.</p>
</section>
<section id="persisting-models-with-a-more-secure-format-using-skops">
<h2><span class="section-number">9.5. </span>Persisting models with a more secure format using skops<a class="headerlink" href="#persisting-models-with-a-more-secure-format-using-skops" title="Link to this heading">¶</a></h2>
<p><a class="reference external" href="https://skops.readthedocs.io/en/stable/">skops</a> provides a more secure
format via the <a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#module-skops.io" title="(in skops)"><code class="xref py py-mod docutils literal notranslate"><span class="pre">skops.io</span></code></a> module. It avoids using <a class="reference external" href="https://docs.python.org/3/library/pickle.html#module-pickle" title="(in Python v3.12)"><code class="xref py py-mod docutils literal notranslate"><span class="pre">pickle</span></code></a> and only
loads files which have types and references to functions which are trusted
either by default or by the user.</p>
<p><details id="summary-anchor">
<summary class="btn btn-light">
<strong>Using skops</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 API is very similar to <code class="docutils literal notranslate"><span class="pre">pickle</span></code>, and
you can persist your models as explain in the <a class="reference external" href="https://skops.readthedocs.io/en/stable/persistence.html">docs</a> using
<a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#skops.io.dump" title="(in skops)"><code class="xref py py-func docutils literal notranslate"><span class="pre">skops.io.dump</span></code></a> and <a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#skops.io.dumps" title="(in skops)"><code class="xref py py-func docutils literal notranslate"><span class="pre">skops.io.dumps</span></code></a>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">skops.io</span> <span class="k">as</span> <span class="nn">sio</span>
<span class="n">obj</span> <span class="o">=</span> <span class="n">sio</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">clf</span><span class="p">)</span>
</pre></div>
</div>
<p>And you can load them back using <a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#skops.io.load" title="(in skops)"><code class="xref py py-func docutils literal notranslate"><span class="pre">skops.io.load</span></code></a> and
<a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#skops.io.loads" title="(in skops)"><code class="xref py py-func docutils literal notranslate"><span class="pre">skops.io.loads</span></code></a>. However, you need to specify the types which are
trusted by you. You can get existing unknown types in a dumped object / file
using <a class="reference external" href="https://skops.readthedocs.io/en/stable/modules/classes.html#skops.io.get_untrusted_types" title="(in skops)"><code class="xref py py-func docutils literal notranslate"><span class="pre">skops.io.get_untrusted_types</span></code></a>, and after checking its contents,
pass it to the load function:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">unknown_types</span> <span class="o">=</span> <span class="n">sio</span><span class="o">.</span><span class="n">get_untrusted_types</span><span class="p">(</span><span class="n">data</span><span class="o">=</span><span class="n">obj</span><span class="p">)</span>
<span class="n">clf</span> <span class="o">=</span> <span class="n">sio</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">trusted</span><span class="o">=</span><span class="n">unknown_types</span><span class="p">)</span>
</pre></div>
</div>
<p>If you trust the source of the file / object, you can pass <code class="docutils literal notranslate"><span class="pre">trusted=True</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">clf</span> <span class="o">=</span> <span class="n">sio</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">trusted</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
</pre></div>
</div>
<p>Please report issues and feature requests related to this format on the <a class="reference external" href="https://github.com/skops-dev/skops/issues">skops
issue tracker</a>.</p>
<p></div>
</details></p>
</section>
<section id="persisting-models-with-interoperable-formats">
<h2><span class="section-number">9.6. </span>Persisting models with interoperable formats<a class="headerlink" href="#persisting-models-with-interoperable-formats" title="Link to this heading">¶</a></h2>
<p>For reproducibility and quality control needs, when different architectures
and environments should be taken into account, exporting the model in
<a class="reference external" href="https://onnx.ai/">Open Neural Network
Exchange</a> format or <a class="reference external" href="https://dmg.org/pmml/v4-4-1/GeneralStructure.html">Predictive Model Markup Language
(PMML)</a> format
might be a better approach than using <code class="docutils literal notranslate"><span class="pre">pickle</span></code> alone.
These are helpful where you may want to use your model for prediction in a
different environment from where the model was trained.</p>
<p>ONNX is a binary serialization of the model. It has been developed to improve
the usability of the interoperable representation of data models.
It aims to facilitate the conversion of the data
models between different machine learning frameworks, and to improve their
portability on different computing architectures. More details are available
from the <a class="reference external" href="https://onnx.ai/get-started.html">ONNX tutorial</a>.
To convert scikit-learn model to ONNX a specific tool <a class="reference external" href="http://onnx.ai/sklearn-onnx/">sklearn-onnx</a> has been developed.</p>
<p>PMML is an implementation of the <a class="reference external" href="https://en.wikipedia.org/wiki/XML">XML</a> document standard
defined to represent data models together with the data used to generate them.
Being human and machine readable,
PMML is a good option for model validation on different platforms and
long term archiving. On the other hand, as XML in general, its verbosity does
not help in production when performance is critical.
To convert scikit-learn model to PMML you can use for example <a class="reference external" href="https://github.com/jpmml/sklearn2pmml">sklearn2pmml</a> distributed under the Affero GPLv3
license.</p>
</section>
<section id="summarizing-the-keypoints">
<h2><span class="section-number">9.7. </span>Summarizing the keypoints<a class="headerlink" href="#summarizing-the-keypoints" title="Link to this heading">¶</a></h2>
<p>Based on the different approaches for model persistence, the keypoints for each
approach can be summarized as follows:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">pickle</span></code>: It is native to Python and any Python object can be serialized and
deserialized using <code class="docutils literal notranslate"><span class="pre">pickle</span></code>, including custom Python classes and objects.
While <code class="docutils literal notranslate"><span class="pre">pickle</span></code> can be used to easily save and load scikit-learn models,
unpickling of untrusted data might lead to security issues.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">joblib</span></code>: Efficient storage and memory mapping techniques make it faster
when working with large machine learning models or large numpy arrays. However,
it may trigger the execution of malicious code while loading untrusted data.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">skops</span></code>: Trained scikit-learn models can be easily shared and put into
production using <code class="docutils literal notranslate"><span class="pre">skops</span></code>. It is more secure compared to alternate approaches
as it allows users to load data from trusted sources. It however, does not
allow for persistence of arbitrary Python code.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">ONNX</span></code>: It provides a uniform format for persisting any machine learning
or deep learning model (other than scikit-learn) and is useful
for model inference. It can however, result in compatibility issues with
different frameworks.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">PMML</span></code>: Platform independent format that can be used to persist models
and reduce the risk of vendor lock-ins. The complexity and verbosity of
this format might make it harder to use for larger models.</p></li>
</ul>
</section>
</section>
</div>
<div class="container">
<footer class="sk-content-footer">
© 2007 - 2024, scikit-learn developers (BSD License).
<a href="_sources/model_persistence.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>