forked from lgatto/IntroMachineLearningWithR
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathunsupervised-learning.html
More file actions
904 lines (865 loc) · 63.4 KB
/
Copy pathunsupervised-learning.html
File metadata and controls
904 lines (865 loc) · 63.4 KB
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
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
<!DOCTYPE html>
<html lang="" xml:lang="">
<head>
<meta charset="utf-8" />
<meta http-equiv="X-UA-Compatible" content="IE=edge" />
<title>Chapter 4 Unsupervised Learning | An Introduction to Machine Learning with R</title>
<meta name="description" content="An hands-on introduction to machine learning with R." />
<meta name="generator" content="bookdown 0.17 and GitBook 2.6.7" />
<meta property="og:title" content="Chapter 4 Unsupervised Learning | An Introduction to Machine Learning with R" />
<meta property="og:type" content="book" />
<meta property="og:description" content="An hands-on introduction to machine learning with R." />
<meta name="github-repo" content="lgatto/IntroMachineLearningWithR" />
<meta name="twitter:card" content="summary" />
<meta name="twitter:title" content="Chapter 4 Unsupervised Learning | An Introduction to Machine Learning with R" />
<meta name="twitter:description" content="An hands-on introduction to machine learning with R." />
<meta name="author" content="Laurent Gatto" />
<meta name="date" content="2020-02-28" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<meta name="apple-mobile-web-app-capable" content="yes" />
<meta name="apple-mobile-web-app-status-bar-style" content="black" />
<link rel="prev" href="example-datasets.html"/>
<link rel="next" href="supervised-learning.html"/>
<script src="libs/jquery-2.2.3/jquery.min.js"></script>
<link href="libs/gitbook-2.6.7/css/style.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-table.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-bookdown.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-highlight.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-search.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-fontsettings.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-clipboard.css" rel="stylesheet" />
<script src="libs/htmlwidgets-1.5.1/htmlwidgets.js"></script>
<link href="libs/datatables-css-0.0.0/datatables-crosstalk.css" rel="stylesheet" />
<script src="libs/datatables-binding-0.12/datatables.js"></script>
<link href="libs/dt-core-1.10.20/css/jquery.dataTables.min.css" rel="stylesheet" />
<link href="libs/dt-core-1.10.20/css/jquery.dataTables.extra.css" rel="stylesheet" />
<script src="libs/dt-core-1.10.20/js/jquery.dataTables.min.js"></script>
<link href="libs/crosstalk-1.0.0/css/crosstalk.css" rel="stylesheet" />
<script src="libs/crosstalk-1.0.0/js/crosstalk.min.js"></script>
<style type="text/css">
a.sourceLine { display: inline-block; line-height: 1.25; }
a.sourceLine { pointer-events: none; color: inherit; text-decoration: inherit; }
a.sourceLine:empty { height: 1.2em; }
.sourceCode { overflow: visible; }
code.sourceCode { white-space: pre; position: relative; }
pre.sourceCode { margin: 0; }
@media screen {
div.sourceCode { overflow: auto; }
}
@media print {
code.sourceCode { white-space: pre-wrap; }
a.sourceLine { text-indent: -1em; padding-left: 1em; }
}
pre.numberSource a.sourceLine
{ position: relative; left: -4em; }
pre.numberSource a.sourceLine::before
{ content: attr(data-line-number);
position: relative; left: -1em; text-align: right; vertical-align: baseline;
border: none; pointer-events: all; display: inline-block;
-webkit-touch-callout: none; -webkit-user-select: none;
-khtml-user-select: none; -moz-user-select: none;
-ms-user-select: none; user-select: none;
padding: 0 4px; width: 4em;
color: #aaaaaa;
}
pre.numberSource { margin-left: 3em; border-left: 1px solid #aaaaaa; padding-left: 4px; }
div.sourceCode
{ }
@media screen {
a.sourceLine::before { text-decoration: underline; }
}
code span.al { color: #ff0000; font-weight: bold; } /* Alert */
code span.an { color: #60a0b0; font-weight: bold; font-style: italic; } /* Annotation */
code span.at { color: #7d9029; } /* Attribute */
code span.bn { color: #40a070; } /* BaseN */
code span.bu { } /* BuiltIn */
code span.cf { color: #007020; font-weight: bold; } /* ControlFlow */
code span.ch { color: #4070a0; } /* Char */
code span.cn { color: #880000; } /* Constant */
code span.co { color: #60a0b0; font-style: italic; } /* Comment */
code span.cv { color: #60a0b0; font-weight: bold; font-style: italic; } /* CommentVar */
code span.do { color: #ba2121; font-style: italic; } /* Documentation */
code span.dt { color: #902000; } /* DataType */
code span.dv { color: #40a070; } /* DecVal */
code span.er { color: #ff0000; font-weight: bold; } /* Error */
code span.ex { } /* Extension */
code span.fl { color: #40a070; } /* Float */
code span.fu { color: #06287e; } /* Function */
code span.im { } /* Import */
code span.in { color: #60a0b0; font-weight: bold; font-style: italic; } /* Information */
code span.kw { color: #007020; font-weight: bold; } /* Keyword */
code span.op { color: #666666; } /* Operator */
code span.ot { color: #007020; } /* Other */
code span.pp { color: #bc7a00; } /* Preprocessor */
code span.sc { color: #4070a0; } /* SpecialChar */
code span.ss { color: #bb6688; } /* SpecialString */
code span.st { color: #4070a0; } /* String */
code span.va { color: #19177c; } /* Variable */
code span.vs { color: #4070a0; } /* VerbatimString */
code span.wa { color: #60a0b0; font-weight: bold; font-style: italic; } /* Warning */
</style>
</head>
<body>
<div class="book without-animation with-summary font-size-2 font-family-1" data-basepath=".">
<div class="book-summary">
<nav role="navigation">
<ul class="summary">
<li class="chapter" data-level="1" data-path="index.html"><a href="index.html"><i class="fa fa-check"></i><b>1</b> Preface</a><ul>
<li class="chapter" data-level="1.1" data-path="index.html"><a href="index.html#caution"><i class="fa fa-check"></i><b>1.1</b> Caution</a></li>
<li class="chapter" data-level="1.2" data-path="index.html"><a href="index.html#installation"><i class="fa fa-check"></i><b>1.2</b> Installation</a></li>
<li class="chapter" data-level="1.3" data-path="index.html"><a href="index.html#license"><i class="fa fa-check"></i><b>1.3</b> License</a></li>
<li class="chapter" data-level="1.4" data-path="index.html"><a href="index.html#contact"><i class="fa fa-check"></i><b>1.4</b> Contact</a></li>
</ul></li>
<li class="chapter" data-level="2" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html"><i class="fa fa-check"></i><b>2</b> An Introduction to Machine Learning with R</a><ul>
<li class="chapter" data-level="2.1" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#objectives-and-pre-requisites"><i class="fa fa-check"></i><b>2.1</b> Objectives and pre-requisites</a></li>
<li class="chapter" data-level="2.2" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#why-r"><i class="fa fa-check"></i><b>2.2</b> Why R?</a></li>
<li class="chapter" data-level="2.3" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#overview-of-machine-learning-ml"><i class="fa fa-check"></i><b>2.3</b> Overview of machine learning (ML)</a></li>
<li class="chapter" data-level="2.4" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#material-and-methods"><i class="fa fa-check"></i><b>2.4</b> Material and methods</a><ul>
<li class="chapter" data-level="2.4.1" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#example-data"><i class="fa fa-check"></i><b>2.4.1</b> Example data</a></li>
<li class="chapter" data-level="2.4.2" data-path="an-introduction-to-machine-learning-with-r.html"><a href="an-introduction-to-machine-learning-with-r.html#packages"><i class="fa fa-check"></i><b>2.4.2</b> Packages</a></li>
</ul></li>
</ul></li>
<li class="chapter" data-level="3" data-path="example-datasets.html"><a href="example-datasets.html"><i class="fa fa-check"></i><b>3</b> Example datasets</a><ul>
<li class="chapter" data-level="3.1" data-path="example-datasets.html"><a href="example-datasets.html#edgar-andersons-iris-data"><i class="fa fa-check"></i><b>3.1</b> Edgar Anderson’s Iris Data</a></li>
<li class="chapter" data-level="3.2" data-path="example-datasets.html"><a href="example-datasets.html#motor-trend-car-road-tests"><i class="fa fa-check"></i><b>3.2</b> Motor Trend Car Road Tests</a></li>
<li class="chapter" data-level="3.3" data-path="example-datasets.html"><a href="example-datasets.html#sub-cellular-localisation"><i class="fa fa-check"></i><b>3.3</b> Sub-cellular localisation</a></li>
<li class="chapter" data-level="3.4" data-path="example-datasets.html"><a href="example-datasets.html#the-diamonds-data"><i class="fa fa-check"></i><b>3.4</b> The diamonds data</a></li>
<li class="chapter" data-level="3.5" data-path="example-datasets.html"><a href="example-datasets.html#the-sonar-data"><i class="fa fa-check"></i><b>3.5</b> The Sonar data</a></li>
<li class="chapter" data-level="3.6" data-path="example-datasets.html"><a href="example-datasets.html#housing-values-in-suburbs-of-boston"><i class="fa fa-check"></i><b>3.6</b> Housing Values in Suburbs of Boston</a></li>
<li class="chapter" data-level="3.7" data-path="example-datasets.html"><a href="example-datasets.html#customer-churn"><i class="fa fa-check"></i><b>3.7</b> Customer churn</a></li>
</ul></li>
<li class="chapter" data-level="4" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html"><i class="fa fa-check"></i><b>4</b> Unsupervised Learning</a><ul>
<li class="chapter" data-level="4.1" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#introduction"><i class="fa fa-check"></i><b>4.1</b> Introduction</a></li>
<li class="chapter" data-level="4.2" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#k-means-clustering"><i class="fa fa-check"></i><b>4.2</b> k-means clustering</a><ul>
<li class="chapter" data-level="4.2.1" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#how-does-k-means-work"><i class="fa fa-check"></i><b>4.2.1</b> How does k-means work</a></li>
<li class="chapter" data-level="4.2.2" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#model-selection"><i class="fa fa-check"></i><b>4.2.2</b> Model selection</a></li>
<li class="chapter" data-level="4.2.3" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#how-to-determine-the-number-of-clusters"><i class="fa fa-check"></i><b>4.2.3</b> How to determine the number of clusters</a></li>
</ul></li>
<li class="chapter" data-level="4.3" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#hierarchical-clustering"><i class="fa fa-check"></i><b>4.3</b> Hierarchical clustering</a><ul>
<li class="chapter" data-level="4.3.1" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#how-does-hierarchical-clustering-work"><i class="fa fa-check"></i><b>4.3.1</b> How does hierarchical clustering work</a></li>
<li class="chapter" data-level="4.3.2" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#defining-clusters"><i class="fa fa-check"></i><b>4.3.2</b> Defining clusters</a></li>
</ul></li>
<li class="chapter" data-level="4.4" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#pre-processing"><i class="fa fa-check"></i><b>4.4</b> Pre-processing</a></li>
<li class="chapter" data-level="4.5" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#principal-component-analysis-pca"><i class="fa fa-check"></i><b>4.5</b> Principal component analysis (PCA)</a><ul>
<li class="chapter" data-level="4.5.1" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#how-does-it-work"><i class="fa fa-check"></i><b>4.5.1</b> How does it work</a></li>
<li class="chapter" data-level="4.5.2" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#visualisation"><i class="fa fa-check"></i><b>4.5.2</b> Visualisation</a></li>
<li class="chapter" data-level="4.5.3" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#data-pre-processing"><i class="fa fa-check"></i><b>4.5.3</b> Data pre-processing</a></li>
<li class="chapter" data-level="4.5.4" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#final-comments-on-pca"><i class="fa fa-check"></i><b>4.5.4</b> Final comments on PCA</a></li>
</ul></li>
<li class="chapter" data-level="4.6" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#t-distributed-stochastic-neighbour-embedding"><i class="fa fa-check"></i><b>4.6</b> t-Distributed Stochastic Neighbour Embedding</a><ul>
<li class="chapter" data-level="4.6.1" data-path="unsupervised-learning.html"><a href="unsupervised-learning.html#parameter-tuning"><i class="fa fa-check"></i><b>4.6.1</b> Parameter tuning</a></li>
</ul></li>
</ul></li>
<li class="chapter" data-level="5" data-path="supervised-learning.html"><a href="supervised-learning.html"><i class="fa fa-check"></i><b>5</b> Supervised Learning</a><ul>
<li class="chapter" data-level="5.1" data-path="supervised-learning.html"><a href="supervised-learning.html#introduction-1"><i class="fa fa-check"></i><b>5.1</b> Introduction</a></li>
<li class="chapter" data-level="5.2" data-path="supervised-learning.html"><a href="supervised-learning.html#preview"><i class="fa fa-check"></i><b>5.2</b> Preview</a></li>
<li class="chapter" data-level="5.3" data-path="supervised-learning.html"><a href="supervised-learning.html#model-performance"><i class="fa fa-check"></i><b>5.3</b> Model performance</a><ul>
<li class="chapter" data-level="5.3.1" data-path="supervised-learning.html"><a href="supervised-learning.html#in-sample-and-out-of-sample-error"><i class="fa fa-check"></i><b>5.3.1</b> In-sample and out-of-sample error</a></li>
<li class="chapter" data-level="5.3.2" data-path="supervised-learning.html"><a href="supervised-learning.html#cross-validation"><i class="fa fa-check"></i><b>5.3.2</b> Cross-validation</a></li>
</ul></li>
<li class="chapter" data-level="5.4" data-path="supervised-learning.html"><a href="supervised-learning.html#classification-performance"><i class="fa fa-check"></i><b>5.4</b> Classification performance</a><ul>
<li class="chapter" data-level="5.4.1" data-path="supervised-learning.html"><a href="supervised-learning.html#confusion-matrix"><i class="fa fa-check"></i><b>5.4.1</b> Confusion matrix</a></li>
<li class="chapter" data-level="5.4.2" data-path="supervised-learning.html"><a href="supervised-learning.html#receiver-operating-characteristic-roc-curve"><i class="fa fa-check"></i><b>5.4.2</b> Receiver operating characteristic (ROC) curve</a></li>
<li class="chapter" data-level="5.4.3" data-path="supervised-learning.html"><a href="supervised-learning.html#auc-in-caret"><i class="fa fa-check"></i><b>5.4.3</b> AUC in <code>caret</code></a></li>
</ul></li>
<li class="chapter" data-level="5.5" data-path="supervised-learning.html"><a href="supervised-learning.html#random-forest"><i class="fa fa-check"></i><b>5.5</b> Random forest</a><ul>
<li class="chapter" data-level="5.5.1" data-path="supervised-learning.html"><a href="supervised-learning.html#decision-trees"><i class="fa fa-check"></i><b>5.5.1</b> Decision trees</a></li>
<li class="chapter" data-level="5.5.2" data-path="supervised-learning.html"><a href="supervised-learning.html#training-a-random-forest"><i class="fa fa-check"></i><b>5.5.2</b> Training a random forest</a></li>
</ul></li>
<li class="chapter" data-level="5.6" data-path="supervised-learning.html"><a href="supervised-learning.html#data-pre-processing-1"><i class="fa fa-check"></i><b>5.6</b> Data pre-processing</a><ul>
<li class="chapter" data-level="5.6.1" data-path="supervised-learning.html"><a href="supervised-learning.html#missing-values"><i class="fa fa-check"></i><b>5.6.1</b> Missing values</a></li>
<li class="chapter" data-level="5.6.2" data-path="supervised-learning.html"><a href="supervised-learning.html#median-imputation"><i class="fa fa-check"></i><b>5.6.2</b> Median imputation</a></li>
<li class="chapter" data-level="5.6.3" data-path="supervised-learning.html"><a href="supervised-learning.html#knn-imputation"><i class="fa fa-check"></i><b>5.6.3</b> kNN imputation</a></li>
</ul></li>
<li class="chapter" data-level="5.7" data-path="supervised-learning.html"><a href="supervised-learning.html#scaling-and-centering"><i class="fa fa-check"></i><b>5.7</b> Scaling and centering</a><ul>
<li class="chapter" data-level="5.7.1" data-path="supervised-learning.html"><a href="supervised-learning.html#multiple-pre-processing-methods"><i class="fa fa-check"></i><b>5.7.1</b> Multiple pre-processing methods</a></li>
</ul></li>
<li class="chapter" data-level="5.8" data-path="supervised-learning.html"><a href="supervised-learning.html#model-selection-1"><i class="fa fa-check"></i><b>5.8</b> Model selection</a><ul>
<li class="chapter" data-level="5.8.1" data-path="supervised-learning.html"><a href="supervised-learning.html#glmnet-model"><i class="fa fa-check"></i><b>5.8.1</b> <code>glmnet</code> model</a></li>
<li class="chapter" data-level="5.8.2" data-path="supervised-learning.html"><a href="supervised-learning.html#random-forest-model"><i class="fa fa-check"></i><b>5.8.2</b> random forest model</a></li>
<li class="chapter" data-level="5.8.3" data-path="supervised-learning.html"><a href="supervised-learning.html#knn-model"><i class="fa fa-check"></i><b>5.8.3</b> kNN model</a></li>
<li class="chapter" data-level="5.8.4" data-path="supervised-learning.html"><a href="supervised-learning.html#support-vector-machine-model"><i class="fa fa-check"></i><b>5.8.4</b> Support vector machine model</a></li>
<li class="chapter" data-level="5.8.5" data-path="supervised-learning.html"><a href="supervised-learning.html#naive-bayes"><i class="fa fa-check"></i><b>5.8.5</b> Naive Bayes</a></li>
<li class="chapter" data-level="5.8.6" data-path="supervised-learning.html"><a href="supervised-learning.html#comparing-models"><i class="fa fa-check"></i><b>5.8.6</b> Comparing models</a></li>
<li class="chapter" data-level="5.8.7" data-path="supervised-learning.html"><a href="supervised-learning.html#pre-processing-1"><i class="fa fa-check"></i><b>5.8.7</b> Pre-processing</a></li>
<li class="chapter" data-level="5.8.8" data-path="supervised-learning.html"><a href="supervised-learning.html#predict-using-the-best-model"><i class="fa fa-check"></i><b>5.8.8</b> Predict using the best model</a></li>
</ul></li>
</ul></li>
<li class="chapter" data-level="6" data-path="final-notes.html"><a href="final-notes.html"><i class="fa fa-check"></i><b>6</b> Final notes</a><ul>
<li class="chapter" data-level="6.1" data-path="final-notes.html"><a href="final-notes.html#other-learning-algorithms"><i class="fa fa-check"></i><b>6.1</b> Other learning algorithms</a><ul>
<li class="chapter" data-level="" data-path="final-notes.html"><a href="final-notes.html#semi-supervised-learning"><i class="fa fa-check"></i>Semi-supervised learning</a></li>
<li class="chapter" data-level="" data-path="final-notes.html"><a href="final-notes.html#deep-learning-in-r"><i class="fa fa-check"></i>Deep learning in R</a></li>
</ul></li>
<li class="chapter" data-level="6.2" data-path="final-notes.html"><a href="final-notes.html#model-performance-1"><i class="fa fa-check"></i><b>6.2</b> Model performance</a></li>
<li class="chapter" data-level="6.3" data-path="final-notes.html"><a href="final-notes.html#credit-and-acknowledgements"><i class="fa fa-check"></i><b>6.3</b> Credit and acknowledgements</a></li>
<li class="chapter" data-level="6.4" data-path="final-notes.html"><a href="final-notes.html#references-and-further-reading"><i class="fa fa-check"></i><b>6.4</b> References and further reading</a></li>
<li class="chapter" data-level="6.5" data-path="final-notes.html"><a href="final-notes.html#session-information"><i class="fa fa-check"></i><b>6.5</b> Session information</a></li>
</ul></li>
</ul>
</nav>
</div>
<div class="book-body">
<div class="body-inner">
<div class="book-header" role="navigation">
<h1>
<i class="fa fa-circle-o-notch fa-spin"></i><a href="./">An Introduction to Machine Learning with R</a>
</h1>
</div>
<div class="page-wrapper" tabindex="-1" role="main">
<div class="page-inner">
<section class="normal" id="section-">
<div id="unsupervised-learning" class="section level1">
<h1><span class="header-section-number">Chapter 4</span> Unsupervised Learning</h1>
<div id="introduction" class="section level2">
<h2><span class="header-section-number">4.1</span> Introduction</h2>
<p>In <strong>unsupervised learning</strong> (UML), no labels are provided, and the
learning algorithm focuses solely on detecting structure in unlabelled
input data. One generally differentiates between</p>
<ul>
<li><p><strong>Clustering</strong>, where the goal is to find homogeneous subgroups
within the data; the grouping is based on distance between
observations.</p></li>
<li><p><strong>Dimensionality reduction</strong>, where the goal is to identify patterns in
the features of the data. Dimensionality reduction is often used to
facilitate visualisation of the data, as well as a pre-processing
method before supervised learning.</p></li>
</ul>
<p>UML presents specific challenges and benefits:</p>
<ul>
<li>there is no single goal in UML</li>
<li>there is generally much more unlabelled data available than labelled
data.</li>
</ul>
</div>
<div id="k-means-clustering" class="section level2">
<h2><span class="header-section-number">4.2</span> k-means clustering</h2>
<p>The k-means clustering algorithms aims at partitioning <em>n</em>
observations into a fixed number of <em>k</em> clusters. The algorithm will
find homogeneous clusters.</p>
<p>In R, we use</p>
<div class="sourceCode" id="cb15"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb15-1" data-line-number="1">stats<span class="op">::</span><span class="kw">kmeans</span>(x, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">10</span>)</a></code></pre></div>
<p>where</p>
<ul>
<li><code>x</code> is a numeric data matrix</li>
<li><code>centers</code> is the pre-defined number of clusters</li>
<li>the k-means algorithm has a random component and can be repeated
<code>nstart</code> times to improve the returned model</li>
</ul>
<blockquote>
<p>Challenge:</p>
<ul>
<li>To learn about k-means, let’s use the <code>iris</code> dataset with the sepal and
petal length variables only (to facilitate visualisation). Create
such a data matrix and name it <code>x</code></li>
</ul>
</blockquote>
<blockquote>
<ul>
<li>Run the k-means algorithm on the newly generated data <code>x</code>, save
the results in a new variable <code>cl</code>, and explore its output when
printed.</li>
</ul>
</blockquote>
<blockquote>
<ul>
<li>The actual results of the algorithms, i.e. the cluster membership
can be accessed in the <code>clusters</code> element of the clustering result
output. Use it to colour the inferred clusters to generate a figure
like that shown below.</li>
</ul>
</blockquote>
<div class="figure"><span id="fig:solkmplot"></span>
<img src="IntroMachineLearningWithR_files/figure-html/solkmplot-1.png" alt="k-means algorithm on sepal and petal lengths" width="672" />
<p class="caption">
Figure 4.1: k-means algorithm on sepal and petal lengths
</p>
</div>
<details>
<div class="sourceCode" id="cb16"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb16-1" data-line-number="1">i <-<span class="st"> </span><span class="kw">grep</span>(<span class="st">"Length"</span>, <span class="kw">names</span>(iris))</a>
<a class="sourceLine" id="cb16-2" data-line-number="2">x <-<span class="st"> </span>iris[, i]</a>
<a class="sourceLine" id="cb16-3" data-line-number="3">cl <-<span class="st"> </span><span class="kw">kmeans</span>(x, <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">10</span>)</a>
<a class="sourceLine" id="cb16-4" data-line-number="4"><span class="kw">plot</span>(x, <span class="dt">col =</span> cl<span class="op">$</span>cluster)</a></code></pre></div>
</details>
<div id="how-does-k-means-work" class="section level3">
<h3><span class="header-section-number">4.2.1</span> How does k-means work</h3>
<p><strong>Initialisation</strong>: randomly assign class membership</p>
<div class="sourceCode" id="cb17"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb17-1" data-line-number="1"><span class="kw">set.seed</span>(<span class="dv">12</span>)</a>
<a class="sourceLine" id="cb17-2" data-line-number="2">init <-<span class="st"> </span><span class="kw">sample</span>(<span class="dv">3</span>, <span class="kw">nrow</span>(x), <span class="dt">replace =</span> <span class="ot">TRUE</span>)</a>
<a class="sourceLine" id="cb17-3" data-line-number="3"><span class="kw">plot</span>(x, <span class="dt">col =</span> init)</a></code></pre></div>
<div class="figure"><span id="fig:kmworksinit"></span>
<img src="IntroMachineLearningWithR_files/figure-html/kmworksinit-1.png" alt="k-means random intialisation" width="672" />
<p class="caption">
Figure 4.2: k-means random intialisation
</p>
</div>
<p><strong>Iteration</strong>:</p>
<ol style="list-style-type: decimal">
<li>Calculate the centre of each subgroup as the average position of
all observations is that subgroup.</li>
<li>Each observation is then assigned to the group of its nearest
centre.</li>
</ol>
<p>It’s also possible to stop the algorithm after a certain number of
iterations, or once the centres move less than a certain distance.</p>
<div class="sourceCode" id="cb18"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb18-1" data-line-number="1"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb18-2" data-line-number="2"><span class="kw">plot</span>(x, <span class="dt">col =</span> init)</a>
<a class="sourceLine" id="cb18-3" data-line-number="3">centres <-<span class="st"> </span><span class="kw">sapply</span>(<span class="dv">1</span><span class="op">:</span><span class="dv">3</span>, <span class="cf">function</span>(i) <span class="kw">colMeans</span>(x[init <span class="op">==</span><span class="st"> </span>i, ], ))</a>
<a class="sourceLine" id="cb18-4" data-line-number="4">centres <-<span class="st"> </span><span class="kw">t</span>(centres)</a>
<a class="sourceLine" id="cb18-5" data-line-number="5"><span class="kw">points</span>(centres[, <span class="dv">1</span>], centres[, <span class="dv">2</span>], <span class="dt">pch =</span> <span class="dv">19</span>, <span class="dt">col =</span> <span class="dv">1</span><span class="op">:</span><span class="dv">3</span>)</a>
<a class="sourceLine" id="cb18-6" data-line-number="6"></a>
<a class="sourceLine" id="cb18-7" data-line-number="7">tmp <-<span class="st"> </span><span class="kw">dist</span>(<span class="kw">rbind</span>(centres, x))</a>
<a class="sourceLine" id="cb18-8" data-line-number="8">tmp <-<span class="st"> </span><span class="kw">as.matrix</span>(tmp)[, <span class="dv">1</span><span class="op">:</span><span class="dv">3</span>]</a>
<a class="sourceLine" id="cb18-9" data-line-number="9"></a>
<a class="sourceLine" id="cb18-10" data-line-number="10">ki <-<span class="st"> </span><span class="kw">apply</span>(tmp, <span class="dv">1</span>, which.min)</a>
<a class="sourceLine" id="cb18-11" data-line-number="11">ki <-<span class="st"> </span>ki[<span class="op">-</span>(<span class="dv">1</span><span class="op">:</span><span class="dv">3</span>)]</a>
<a class="sourceLine" id="cb18-12" data-line-number="12"></a>
<a class="sourceLine" id="cb18-13" data-line-number="13"><span class="kw">plot</span>(x, <span class="dt">col =</span> ki)</a>
<a class="sourceLine" id="cb18-14" data-line-number="14"><span class="kw">points</span>(centres[, <span class="dv">1</span>], centres[, <span class="dv">2</span>], <span class="dt">pch =</span> <span class="dv">19</span>, <span class="dt">col =</span> <span class="dv">1</span><span class="op">:</span><span class="dv">3</span>)</a></code></pre></div>
<div class="figure"><span id="fig:kmworksiter"></span>
<img src="IntroMachineLearningWithR_files/figure-html/kmworksiter-1.png" alt="k-means iteration: calculate centers (left) and assign new cluster membership (right)" width="1152" />
<p class="caption">
Figure 4.3: k-means iteration: calculate centers (left) and assign new cluster membership (right)
</p>
</div>
<p><strong>Termination</strong>: Repeat iteration until no point changes its cluster
membership.</p>
<div class="figure">
<img src="https://upload.wikimedia.org/wikipedia/commons/e/ea/K-means_convergence.gif" alt="k-means convergence (credit Wikipedia)" />
<p class="caption">k-means convergence (credit Wikipedia)</p>
</div>
</div>
<div id="model-selection" class="section level3">
<h3><span class="header-section-number">4.2.2</span> Model selection</h3>
<p>Due to the random initialisation, one can obtain different clustering
results. When k-means is run multiple times, the best outcome,
i.e. the one that generates the smallest <em>total within cluster sum of
squares (SS)</em>, is selected. The total within SS is calculated as:</p>
<p>For each cluster results:</p>
<ul>
<li>for each observation, determine the squared euclidean distance from
observation to centre of cluster</li>
<li>sum all distances</li>
</ul>
<p>Note that this is a <strong>local minimum</strong>; there is no guarantee to obtain
a global minimum.</p>
<blockquote>
<p>Challenge:</p>
<p>Repeat k-means on our <code>x</code> data multiple times, setting the number of
iterations to 1 or greater and check whether you repeatedly obtain
the same results. Try the same with random data of identical
dimensions.</p>
</blockquote>
<details>
<div class="sourceCode" id="cb19"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb19-1" data-line-number="1">cl1 <-<span class="st"> </span><span class="kw">kmeans</span>(x, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">10</span>)</a>
<a class="sourceLine" id="cb19-2" data-line-number="2">cl2 <-<span class="st"> </span><span class="kw">kmeans</span>(x, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">10</span>)</a>
<a class="sourceLine" id="cb19-3" data-line-number="3"><span class="kw">table</span>(cl1<span class="op">$</span>cluster, cl2<span class="op">$</span>cluster)</a></code></pre></div>
<pre><code>##
## 1 2 3
## 1 0 41 0
## 2 51 0 0
## 3 0 0 58</code></pre>
<div class="sourceCode" id="cb21"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb21-1" data-line-number="1">cl1 <-<span class="st"> </span><span class="kw">kmeans</span>(x, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">1</span>)</a>
<a class="sourceLine" id="cb21-2" data-line-number="2">cl2 <-<span class="st"> </span><span class="kw">kmeans</span>(x, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">1</span>)</a>
<a class="sourceLine" id="cb21-3" data-line-number="3"><span class="kw">table</span>(cl1<span class="op">$</span>cluster, cl2<span class="op">$</span>cluster)</a></code></pre></div>
<pre><code>##
## 1 2 3
## 1 41 0 0
## 2 0 0 58
## 3 0 51 0</code></pre>
<div class="sourceCode" id="cb23"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb23-1" data-line-number="1"><span class="kw">set.seed</span>(<span class="dv">42</span>)</a>
<a class="sourceLine" id="cb23-2" data-line-number="2">xr <-<span class="st"> </span><span class="kw">matrix</span>(<span class="kw">rnorm</span>(<span class="kw">prod</span>(<span class="kw">dim</span>(x))), <span class="dt">ncol =</span> <span class="kw">ncol</span>(x))</a>
<a class="sourceLine" id="cb23-3" data-line-number="3">cl1 <-<span class="st"> </span><span class="kw">kmeans</span>(xr, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">1</span>)</a>
<a class="sourceLine" id="cb23-4" data-line-number="4">cl2 <-<span class="st"> </span><span class="kw">kmeans</span>(xr, <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">1</span>)</a>
<a class="sourceLine" id="cb23-5" data-line-number="5"><span class="kw">table</span>(cl1<span class="op">$</span>cluster, cl2<span class="op">$</span>cluster)</a></code></pre></div>
<pre><code>##
## 1 2 3
## 1 0 52 0
## 2 0 0 46
## 3 52 0 0</code></pre>
<div class="sourceCode" id="cb25"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb25-1" data-line-number="1">diffres <-<span class="st"> </span>cl1<span class="op">$</span>cluster <span class="op">!=</span><span class="st"> </span>cl2<span class="op">$</span>cluster</a>
<a class="sourceLine" id="cb25-2" data-line-number="2"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb25-3" data-line-number="3"><span class="kw">plot</span>(xr, <span class="dt">col =</span> cl1<span class="op">$</span>cluster, <span class="dt">pch =</span> <span class="kw">ifelse</span>(diffres, <span class="dv">19</span>, <span class="dv">1</span>))</a>
<a class="sourceLine" id="cb25-4" data-line-number="4"><span class="kw">plot</span>(xr, <span class="dt">col =</span> cl2<span class="op">$</span>cluster, <span class="dt">pch =</span> <span class="kw">ifelse</span>(diffres, <span class="dv">19</span>, <span class="dv">1</span>))</a></code></pre></div>
<div class="figure"><span id="fig:selrep"></span>
<img src="IntroMachineLearningWithR_files/figure-html/selrep-1.png" alt="Different k-means results on the same (random) data" width="1152" />
<p class="caption">
Figure 4.4: Different k-means results on the same (random) data
</p>
</div>
</details>
</div>
<div id="how-to-determine-the-number-of-clusters" class="section level3">
<h3><span class="header-section-number">4.2.3</span> How to determine the number of clusters</h3>
<ol style="list-style-type: decimal">
<li>Run k-means with <code>k=1</code>, <code>k=2</code>, …, <code>k=n</code></li>
<li>Record total within SS for each value of k.</li>
<li>Choose k at the <em>elbow</em> position, as illustrated below.</li>
</ol>
<p><img src="IntroMachineLearningWithR_files/figure-html/kmelbow-1.png" width="672" /></p>
<blockquote>
<p>Challenge</p>
<p>Calculate the total within sum of squares for k from 1 to 5 for our
<code>x</code> test data, and reproduce the figure above.</p>
</blockquote>
<details>
<div class="sourceCode" id="cb26"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb26-1" data-line-number="1">ks <-<span class="st"> </span><span class="dv">1</span><span class="op">:</span><span class="dv">5</span></a>
<a class="sourceLine" id="cb26-2" data-line-number="2">tot_within_ss <-<span class="st"> </span><span class="kw">sapply</span>(ks, <span class="cf">function</span>(k) {</a>
<a class="sourceLine" id="cb26-3" data-line-number="3"> cl <-<span class="st"> </span><span class="kw">kmeans</span>(x, k, <span class="dt">nstart =</span> <span class="dv">10</span>)</a>
<a class="sourceLine" id="cb26-4" data-line-number="4"> cl<span class="op">$</span>tot.withinss</a>
<a class="sourceLine" id="cb26-5" data-line-number="5">})</a>
<a class="sourceLine" id="cb26-6" data-line-number="6"><span class="kw">plot</span>(ks, tot_within_ss, <span class="dt">type =</span> <span class="st">"b"</span>)</a></code></pre></div>
<img src="IntroMachineLearningWithR_files/figure-html/solkmelbow-1.png" width="672" />
</details>
</div>
</div>
<div id="hierarchical-clustering" class="section level2">
<h2><span class="header-section-number">4.3</span> Hierarchical clustering</h2>
<div id="how-does-hierarchical-clustering-work" class="section level3">
<h3><span class="header-section-number">4.3.1</span> How does hierarchical clustering work</h3>
<p><strong>Initialisation</strong>: Starts by assigning each of the n points its own cluster</p>
<p><strong>Iteration</strong></p>
<ol style="list-style-type: decimal">
<li>Find the two nearest clusters, and join them together, leading to
n-1 clusters</li>
<li>Continue the cluster merging process until all are grouped into a
single cluster</li>
</ol>
<p><strong>Termination:</strong> All observations are grouped within a single cluster.</p>
<div class="figure"><span id="fig:hcldata"></span>
<img src="IntroMachineLearningWithR_files/figure-html/hcldata-1.png" alt="Hierarchical clustering: initialisation (left) and colour-coded results after iteration (right)." width="1152" />
<p class="caption">
Figure 4.5: Hierarchical clustering: initialisation (left) and colour-coded results after iteration (right).
</p>
</div>
<p>The results of hierarchical clustering are typically visualised along
a <strong>dendrogram</strong>, where the distance between the clusters is
proportional to the branch lengths.</p>
<div class="figure"><span id="fig:hcldendro"></span>
<img src="IntroMachineLearningWithR_files/figure-html/hcldendro-1.png" alt="Visualisation of the hierarchical clustering results on a dendrogram" width="672" />
<p class="caption">
Figure 4.6: Visualisation of the hierarchical clustering results on a dendrogram
</p>
</div>
<p>In R:</p>
<ul>
<li>Calculate the distance using <code>dist</code>, typically the Euclidean
distance.</li>
<li>Hierarchical clustering on this distance matrix using <code>hclust</code></li>
</ul>
<blockquote>
<p>Challenge</p>
<p>Apply hierarchical clustering on the <code>iris</code> data and generate a
dendrogram using the dedicated <code>plot</code> method.</p>
</blockquote>
<details>
<div class="sourceCode" id="cb27"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb27-1" data-line-number="1">d <-<span class="st"> </span><span class="kw">dist</span>(iris[, <span class="dv">1</span><span class="op">:</span><span class="dv">4</span>])</a>
<a class="sourceLine" id="cb27-2" data-line-number="2">hcl <-<span class="st"> </span><span class="kw">hclust</span>(d)</a>
<a class="sourceLine" id="cb27-3" data-line-number="3">hcl</a></code></pre></div>
<pre><code>##
## Call:
## hclust(d = d)
##
## Cluster method : complete
## Distance : euclidean
## Number of objects: 150</code></pre>
<div class="sourceCode" id="cb29"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb29-1" data-line-number="1"><span class="kw">plot</span>(hcl)</a></code></pre></div>
<img src="IntroMachineLearningWithR_files/figure-html/hclsol-1.png" width="672" />
</details>
</div>
<div id="defining-clusters" class="section level3">
<h3><span class="header-section-number">4.3.2</span> Defining clusters</h3>
<p>After producing the hierarchical clustering result, we need to <em>cut
the tree (dendrogram)</em> at a specific height to defined the
clusters. For example, on our test dataset above, we could decide to
cut it at a distance around 1.5, with would produce 2 clusters.</p>
<div class="figure"><span id="fig:cuthcl"></span>
<img src="IntroMachineLearningWithR_files/figure-html/cuthcl-1.png" alt="Cutting the dendrogram at height 1.5." width="672" />
<p class="caption">
Figure 4.7: Cutting the dendrogram at height 1.5.
</p>
</div>
<p>In R we can us the <code>cutree</code> function to</p>
<ul>
<li>cut the tree at a specific height: <code>cutree(hcl, h = 1.5)</code></li>
<li>cut the tree to get a certain number of clusters: <code>cutree(hcl, k = 2)</code></li>
</ul>
<blockquote>
<p>Challenge</p>
<ul>
<li>Cut the iris hierarchical clustering result at a height to obtain
3 clusters by setting <code>h</code>.</li>
<li>Cut the iris hierarchical clustering result at a height to obtain
3 clusters by setting directly <code>k</code>, and verify that both provide
the same results.</li>
</ul>
</blockquote>
<details>
<div class="sourceCode" id="cb30"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb30-1" data-line-number="1"><span class="kw">plot</span>(hcl)</a>
<a class="sourceLine" id="cb30-2" data-line-number="2"><span class="kw">abline</span>(<span class="dt">h =</span> <span class="fl">3.9</span>, <span class="dt">col =</span> <span class="st">"red"</span>)</a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/cuthclsol-1.png" width="672" /></p>
<div class="sourceCode" id="cb31"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb31-1" data-line-number="1"><span class="kw">cutree</span>(hcl, <span class="dt">k =</span> <span class="dv">3</span>)</a></code></pre></div>
<pre><code>## [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [38] 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 3 2 3 2 3 2 3 3 3 3 2 3 2 3 3 2 3 2 3 2 2
## [75] 2 2 2 2 2 3 3 3 3 2 3 2 2 2 3 3 3 2 3 3 3 3 3 2 3 3 2 2 2 2 2 2 3 2 2 2 2
## [112] 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
## [149] 2 2</code></pre>
<div class="sourceCode" id="cb33"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb33-1" data-line-number="1"><span class="kw">cutree</span>(hcl, <span class="dt">h =</span> <span class="fl">3.9</span>)</a></code></pre></div>
<pre><code>## [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [38] 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 3 2 3 2 3 2 3 3 3 3 2 3 2 3 3 2 3 2 3 2 2
## [75] 2 2 2 2 2 3 3 3 3 2 3 2 2 2 3 3 3 2 3 3 3 3 3 2 3 3 2 2 2 2 2 2 3 2 2 2 2
## [112] 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
## [149] 2 2</code></pre>
<div class="sourceCode" id="cb35"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb35-1" data-line-number="1"><span class="kw">identical</span>(<span class="kw">cutree</span>(hcl, <span class="dt">k =</span> <span class="dv">3</span>), <span class="kw">cutree</span>(hcl, <span class="dt">h =</span> <span class="fl">3.9</span>))</a></code></pre></div>
<pre><code>## [1] TRUE</code></pre>
</details>
<blockquote>
<p>Challenge</p>
<p>Using the same value <code>k = 3</code>, verify if k-means and hierarchical
clustering produce the same results on the <code>iris</code> data.</p>
<p>Which one, if any, is correct?</p>
</blockquote>
<details>
<div class="sourceCode" id="cb37"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb37-1" data-line-number="1">km <-<span class="st"> </span><span class="kw">kmeans</span>(iris[, <span class="dv">1</span><span class="op">:</span><span class="dv">4</span>], <span class="dt">centers =</span> <span class="dv">3</span>, <span class="dt">nstart =</span> <span class="dv">10</span>)</a>
<a class="sourceLine" id="cb37-2" data-line-number="2">hcl <-<span class="st"> </span><span class="kw">hclust</span>(<span class="kw">dist</span>(iris[, <span class="dv">1</span><span class="op">:</span><span class="dv">4</span>]))</a>
<a class="sourceLine" id="cb37-3" data-line-number="3"><span class="kw">table</span>(km<span class="op">$</span>cluster, <span class="kw">cutree</span>(hcl, <span class="dt">k =</span> <span class="dv">3</span>))</a></code></pre></div>
<pre><code>##
## 1 2 3
## 1 0 34 28
## 2 0 38 0
## 3 50 0 0</code></pre>
<div class="sourceCode" id="cb39"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb39-1" data-line-number="1"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb39-2" data-line-number="2"><span class="kw">plot</span>(iris<span class="op">$</span>Petal.Length, iris<span class="op">$</span>Sepal.Length, <span class="dt">col =</span> km<span class="op">$</span>cluster, <span class="dt">main =</span> <span class="st">"k-means"</span>)</a>
<a class="sourceLine" id="cb39-3" data-line-number="3"><span class="kw">plot</span>(iris<span class="op">$</span>Petal.Length, iris<span class="op">$</span>Sepal.Length, <span class="dt">col =</span> <span class="kw">cutree</span>(hcl, <span class="dt">k =</span> <span class="dv">3</span>), <span class="dt">main =</span> <span class="st">"Hierarchical clustering"</span>)</a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/iris2algs-1.png" width="1152" /></p>
<div class="sourceCode" id="cb40"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb40-1" data-line-number="1"><span class="co">## Checking with the labels provided with the iris data</span></a>
<a class="sourceLine" id="cb40-2" data-line-number="2"><span class="kw">table</span>(iris<span class="op">$</span>Species, km<span class="op">$</span>cluster)</a></code></pre></div>
<pre><code>##
## 1 2 3
## setosa 0 0 50
## versicolor 48 2 0
## virginica 14 36 0</code></pre>
<div class="sourceCode" id="cb42"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb42-1" data-line-number="1"><span class="kw">table</span>(iris<span class="op">$</span>Species, <span class="kw">cutree</span>(hcl, <span class="dt">k =</span> <span class="dv">3</span>))</a></code></pre></div>
<pre><code>##
## 1 2 3
## setosa 50 0 0
## versicolor 0 23 27
## virginica 0 49 1</code></pre>
</details>
</div>
</div>
<div id="pre-processing" class="section level2">
<h2><span class="header-section-number">4.4</span> Pre-processing</h2>
<p>Many of the machine learning methods that are regularly used are
sensitive to difference scales. This applies to unsupervised methods
as well as supervised methods, as we will see in the next chapter.</p>
<p>A typical way to pre-process the data prior to learning is to scale
the data, or apply principal component analysis (next section). Scaling
assures that all data columns have a mean of 0 and standard deviation of 1.</p>
<p>In R, scaling is done with the <code>scale</code> function.</p>
<blockquote>
<p>Challenge</p>
<p>Using the <code>mtcars</code> data as an example, verify that the variables are
of different scales, then scale the data. To observe the effect
different scales, compare the hierarchical clusters obtained on the
original and scaled data.</p>
</blockquote>
<details>
<div class="sourceCode" id="cb44"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb44-1" data-line-number="1"><span class="kw">colMeans</span>(mtcars)</a></code></pre></div>
<pre><code>## mpg cyl disp hp drat wt qsec
## 20.090625 6.187500 230.721875 146.687500 3.596563 3.217250 17.848750
## vs am gear carb
## 0.437500 0.406250 3.687500 2.812500</code></pre>
<div class="sourceCode" id="cb46"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb46-1" data-line-number="1">hcl1 <-<span class="st"> </span><span class="kw">hclust</span>(<span class="kw">dist</span>(mtcars))</a>
<a class="sourceLine" id="cb46-2" data-line-number="2">hcl2 <-<span class="st"> </span><span class="kw">hclust</span>(<span class="kw">dist</span>(<span class="kw">scale</span>(mtcars)))</a>
<a class="sourceLine" id="cb46-3" data-line-number="3"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb46-4" data-line-number="4"><span class="kw">plot</span>(hcl1, <span class="dt">main =</span> <span class="st">"original data"</span>)</a>
<a class="sourceLine" id="cb46-5" data-line-number="5"><span class="kw">plot</span>(hcl2, <span class="dt">main =</span> <span class="st">"scaled data"</span>)</a></code></pre></div>
<img src="IntroMachineLearningWithR_files/figure-html/scalesol-1.png" width="1152" />
</details>
</div>
<div id="principal-component-analysis-pca" class="section level2">
<h2><span class="header-section-number">4.5</span> Principal component analysis (PCA)</h2>
<p><strong>Dimensionality reduction</strong> techniques are widely used and versatile
techniques that can be used to:</p>
<ul>
<li>find structure in features</li>
<li>pre-processing for other ML algorithms, and</li>
<li>aid in visualisation.</li>
</ul>
<p>The basic principle of dimensionality reduction techniques is to
transform the data into a new space that summarise properties of the
whole data set along a reduced number of dimensions. These are then
ideal candidates used to visualise the data along these reduced number
of informative dimensions.</p>
<div id="how-does-it-work" class="section level3">
<h3><span class="header-section-number">4.5.1</span> How does it work</h3>
<p>Principal Component Analysis (PCA) is a technique that transforms the
original n-dimensional data into a new n-dimensional space.</p>
<ul>
<li>These new dimensions are linear combinations of the original data,
i.e. they are composed of proportions of the original variables.</li>
<li>Along these new dimensions, called principal components, the data
expresses most of its variability along the first PC, then second,
…</li>
<li>Principal components are orthogonal to each other,
i.e. non-correlated.</li>
</ul>
<div class="figure"><span id="fig:pcaex"></span>
<img src="IntroMachineLearningWithR_files/figure-html/pcaex-1.png" alt="Original data (left). PC1 will maximise the variability while minimising the residuals (centre). PC2 is orthogonal to PC1 (right)." width="1152" />
<p class="caption">
Figure 4.8: Original data (left). PC1 will maximise the variability while minimising the residuals (centre). PC2 is orthogonal to PC1 (right).
</p>
</div>
<p>In R, we can use the <code>prcomp</code> function.</p>
<p>Let’s explore PCA on the <code>iris</code> data. While it contains only 4
variables, is already becomes difficult to visualise the 3 groups
along all these dimensions.</p>
<div class="sourceCode" id="cb47"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb47-1" data-line-number="1"><span class="kw">pairs</span>(iris[, <span class="dv">-5</span>], <span class="dt">col =</span> iris[, <span class="dv">5</span>], <span class="dt">pch =</span> <span class="dv">19</span>)</a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/irispairs-1.png" width="672" /></p>
<p>Let’s use PCA to reduce the dimension.</p>
<div class="sourceCode" id="cb48"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb48-1" data-line-number="1">irispca <-<span class="st"> </span><span class="kw">prcomp</span>(iris[, <span class="dv">-5</span>])</a>
<a class="sourceLine" id="cb48-2" data-line-number="2"><span class="kw">summary</span>(irispca)</a></code></pre></div>
<pre><code>## Importance of components:
## PC1 PC2 PC3 PC4
## Standard deviation 2.0563 0.49262 0.2797 0.15439
## Proportion of Variance 0.9246 0.05307 0.0171 0.00521
## Cumulative Proportion 0.9246 0.97769 0.9948 1.00000</code></pre>
<p>A summary of the <code>prcomp</code> output shows that along PC1 along, we are
able to retain over 92% of the total variability in the data.</p>
<div class="figure"><span id="fig:histpc1"></span>
<img src="IntroMachineLearningWithR_files/figure-html/histpc1-1.png" alt="Iris data along PC1." width="672" />
<p class="caption">
Figure 4.9: Iris data along PC1.
</p>
</div>
</div>
<div id="visualisation" class="section level3">
<h3><span class="header-section-number">4.5.2</span> Visualisation</h3>
<p>A <strong>biplot</strong> features all original points re-mapped (rotated) along the
first two PCs as well as the original features as vectors along the
same PCs. Feature vectors that are in the same direction in PC space
are also correlated in the original data space.</p>
<div class="sourceCode" id="cb50"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb50-1" data-line-number="1"><span class="kw">biplot</span>(irispca)</a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/irisbiplot-1.png" width="672" /></p>
<p>One important piece of information when using PCA is the proportion of
variance explained along the PCs, in particular when dealing with high
dimensional data, as PC1 and PC2 (that are generally used for
visualisation), might only account for an insufficient proportion of
variance to be relevant on their own.</p>
<p>In the code chunk below, I extract the standard deviations from the
PCA result to calculate the variances, then obtain the percentage of
and cumulative variance along the PCs.</p>
<div class="sourceCode" id="cb51"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb51-1" data-line-number="1">var <-<span class="st"> </span>irispca<span class="op">$</span>sdev<span class="op">^</span><span class="dv">2</span></a>
<a class="sourceLine" id="cb51-2" data-line-number="2">(pve <-<span class="st"> </span>var<span class="op">/</span><span class="kw">sum</span>(var))</a></code></pre></div>
<pre><code>## [1] 0.924618723 0.053066483 0.017102610 0.005212184</code></pre>
<div class="sourceCode" id="cb53"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb53-1" data-line-number="1"><span class="kw">cumsum</span>(pve)</a></code></pre></div>
<pre><code>## [1] 0.9246187 0.9776852 0.9947878 1.0000000</code></pre>
<blockquote>
<p>Challenge</p>
<ul>
<li>Repeat the PCA analysis on the iris dataset above, reproducing the
biplot and preparing a barplot of the percentage of variance
explained by each PC.</li>
<li>It is often useful to produce custom figures using the data
coordinates in PCA space, which can be accessed as <code>x</code> in the
<code>prcomp</code> object. Reproduce the PCA plots below, along PC1 and PC2
and PC3 and PC4 respectively.</li>
</ul>
</blockquote>
<p><img src="IntroMachineLearningWithR_files/figure-html/irispcax-1.png" width="1152" /></p>
<details>
<div class="sourceCode" id="cb55"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb55-1" data-line-number="1"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb55-2" data-line-number="2"><span class="kw">plot</span>(irispca<span class="op">$</span>x[, <span class="dv">1</span><span class="op">:</span><span class="dv">2</span>], <span class="dt">col =</span> iris<span class="op">$</span>Species)</a>
<a class="sourceLine" id="cb55-3" data-line-number="3"><span class="kw">plot</span>(irispca<span class="op">$</span>x[, <span class="dv">3</span><span class="op">:</span><span class="dv">4</span>], <span class="dt">col =</span> iris<span class="op">$</span>Species)</a></code></pre></div>
</details>
</div>
<div id="data-pre-processing" class="section level3">
<h3><span class="header-section-number">4.5.3</span> Data pre-processing</h3>
<p>We haven’t looked at other <code>prcomp</code> parameters, other that the first
one, <code>x</code>. There are two other ones that are or importance, in
particular in the light of the section on pre-processing above, which
are <code>center</code> and <code>scale.</code>. The former is set to <code>TRUE</code> by default,
while the second one is set the <code>FALSE</code>.</p>
<blockquote>
<p>Challenge</p>
<p>Repeat the analysis comparing the need for scaling on the <code>mtcars</code>
dataset, but using PCA instead of hierarchical clustering. When
comparing the two.</p>
</blockquote>
<details>
<div class="sourceCode" id="cb56"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb56-1" data-line-number="1"><span class="kw">par</span>(<span class="dt">mfrow =</span> <span class="kw">c</span>(<span class="dv">1</span>, <span class="dv">2</span>))</a>
<a class="sourceLine" id="cb56-2" data-line-number="2"><span class="kw">biplot</span>(<span class="kw">prcomp</span>(mtcars, <span class="dt">scale =</span> <span class="ot">FALSE</span>), <span class="dt">main =</span> <span class="st">"No scaling"</span>) <span class="co">## 1</span></a>
<a class="sourceLine" id="cb56-3" data-line-number="3"><span class="kw">biplot</span>(<span class="kw">prcomp</span>(mtcars, <span class="dt">scale =</span> <span class="ot">TRUE</span>), <span class="dt">main =</span> <span class="st">"With scaling"</span>) <span class="co">## 2</span></a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/scalepcasol-1.png" width="672" /></p>
Without scaling, <code>disp</code> and <code>hp</code> are the features with the highest
loadings along PC1 and 2 (all others are negligible), which are also
those with the highest units of measurement. Scaling removes this
effect.
</details>
</div>
<div id="final-comments-on-pca" class="section level3">
<h3><span class="header-section-number">4.5.4</span> Final comments on PCA</h3>
<p>Real datasets often come with <strong>missing values</strong>. In R, these should
be encoded using <code>NA</code>. Unfortunately, PCA cannot deal with missing
values, and observations containing <code>NA</code> values will be dropped
automatically. This is a viable solution only when the proportion of
missing values is low.</p>
<p>It is also possible to impute missing values. This is described in
greater details in the <em>Data pre-processing</em> section in the supervised
machine learning chapter.</p>
<p>Finally, we should be careful when using categorical data in any of
the unsupervised methods described above. Categories are generally
represented as factors, which are encoded as integer levels, and might
give the impression that a distance between levels is a relevant
measure (which it is not, unless the factors are ordered). In such
situations, categorical data can be dropped, or it is possible to
encode categories as binary <strong>dummy variables</strong>. For example, if we
have 3 categories, say <code>A</code>, <code>B</code> and <code>C</code>, we would create two dummy
variables to encode the categories as:</p>
<table>
<thead>
<tr class="header">
<th></th>
<th align="right">x</th>
<th align="right">y</th>
</tr>
</thead>
<tbody>
<tr class="odd">
<td>A</td>
<td align="right">1</td>
<td align="right">0</td>
</tr>
<tr class="even">
<td>B</td>
<td align="right">0</td>
<td align="right">1</td>
</tr>
<tr class="odd">
<td>C</td>
<td align="right">0</td>
<td align="right">0</td>
</tr>
</tbody>
</table>
<p>so that the distance between each category are approximately equal to
1.</p>
</div>
</div>
<div id="t-distributed-stochastic-neighbour-embedding" class="section level2">
<h2><span class="header-section-number">4.6</span> t-Distributed Stochastic Neighbour Embedding</h2>
<p><a href="https://lvdmaaten.github.io/tsne/">t-Distributed Stochastic Neighbour Embedding</a> (t-SNE)
is a <em>non-linear</em> dimensionality reduction technique, i.e. that
different regions of the data space will be subjected to different
transformations. t-SNE will compress small distances, thus bringing
close neighbours together, and will ignore large distances. It is
particularly well suited
for
<a href="https://distill.pub/2016/misread-tsne/">very high dimensional data</a>.</p>
<p>In R, we can use the <code>Rtsne</code> function from the <em><a href="https://CRAN.R-project.org/package=Rtsne">Rtsne</a></em>.
Before, we however need to remove any duplicated entries in the
dataset.</p>
<div class="sourceCode" id="cb57"><pre class="sourceCode r"><code class="sourceCode r"><a class="sourceLine" id="cb57-1" data-line-number="1"><span class="kw">library</span>(<span class="st">"Rtsne"</span>)</a>
<a class="sourceLine" id="cb57-2" data-line-number="2">uiris <-<span class="st"> </span><span class="kw">unique</span>(iris[, <span class="dv">1</span><span class="op">:</span><span class="dv">5</span>])</a>
<a class="sourceLine" id="cb57-3" data-line-number="3">iristsne <-<span class="st"> </span><span class="kw">Rtsne</span>(uiris[, <span class="dv">1</span><span class="op">:</span><span class="dv">4</span>])</a>
<a class="sourceLine" id="cb57-4" data-line-number="4"><span class="kw">plot</span>(iristsne<span class="op">$</span>Y, <span class="dt">col =</span> uiris<span class="op">$</span>Species)</a></code></pre></div>
<p><img src="IntroMachineLearningWithR_files/figure-html/iristsne-1.png" width="672" /></p>
<p>As with PCA, the data can be scaled and centred prior the running
t-SNE (see the <code>pca_center</code> and <code>pca_scale</code> arguments). The algorithm
is stochastic, and will produce different results at each repetition.</p>
<div id="parameter-tuning" class="section level3">
<h3><span class="header-section-number">4.6.1</span> Parameter tuning</h3>
<p>t-SNE (as well as many other methods, in particular classification
algorithms) has two important parameters that can substantially
influence the clustering of the data</p>
<ul>
<li><strong>Perplexity</strong>: balances global and local aspects of the data.</li>
<li><strong>Iterations</strong>: number of iterations before the clustering is
stopped.</li>
</ul>
<p>It is important to adapt these for different data. The figure below
shows a 5032 by 20 dataset that represent protein sub-cellular
localisation.</p>
<div class="figure">
<img src="https://raw.githubusercontent.com/lgatto/visualisation/master/figure/tsnesplots-1.png" alt="Effect of different perplexity and iterations when running t-SNE" />
<p class="caption">Effect of different perplexity and iterations when running t-SNE</p>
</div>
<p>As a comparison, below are the same data with PCA (left) and t-SNE
(right).</p>
<div class="figure">
<img src="https://raw.githubusercontent.com/lgatto/visualisation/master/figure/tsneex-1.png" alt="PCA and t-SNE on hyperLOPIT" />
<p class="caption">PCA and t-SNE on hyperLOPIT</p>
</div>
</div>
</div>
</div>
</section>
</div>
</div>
</div>
<a href="example-datasets.html" class="navigation navigation-prev " aria-label="Previous page"><i class="fa fa-angle-left"></i></a>
<a href="supervised-learning.html" class="navigation navigation-next " aria-label="Next page"><i class="fa fa-angle-right"></i></a>
</div>
</div>
<script src="libs/gitbook-2.6.7/js/app.min.js"></script>
<script src="libs/gitbook-2.6.7/js/lunr.js"></script>
<script src="libs/gitbook-2.6.7/js/clipboard.min.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-search.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-sharing.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-fontsettings.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-bookdown.js"></script>
<script src="libs/gitbook-2.6.7/js/jquery.highlight.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-clipboard.js"></script>
<script>
gitbook.require(["gitbook"], function(gitbook) {
gitbook.start({
"sharing": {
"github": false,
"facebook": true,
"twitter": true,
"linkedin": false,
"weibo": false,
"instapaper": false,
"vk": false,
"all": ["facebook", "twitter", "linkedin", "weibo", "instapaper"]
},
"fontsettings": {
"theme": "white",
"family": "sans",
"size": 2
},
"edit": {
"link": null,
"text": null
},
"history": {
"link": null,
"text": null
},
"view": {
"link": null,
"text": null
},
"download": null,
"toc": {
"collapse": "subsection"
}
});
});
</script>
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement("script");
script.type = "text/javascript";
var src = "true";
if (src === "" || src === "true") src = "https://mathjax.rstudio.com/latest/MathJax.js?config=TeX-MML-AM_CHTML";
if (location.protocol !== "file:")
if (/^https?:/.test(src))
src = src.replace(/^https?:/, '');
script.src = src;
document.getElementsByTagName("head")[0].appendChild(script);
})();
</script>
</body>
</html>