-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmanifold.html
823 lines (741 loc) · 36.1 KB
/
manifold.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Smile - Manifold Learning</title>
<meta name="description" content="Statistical Machine Intelligence and Learning Engine">
<!-- prettify js and CSS -->
<script src="https://cdn.rawgit.com/google/code-prettify/master/loader/run_prettify.js?lang=scala&lang=kotlin&lang=clj"></script>
<style>
.prettyprint ol.linenums > li { list-style-type: decimal; }
</style>
<!-- Bootstrap core CSS -->
<link href="css/cerulean.min.css" rel="stylesheet">
<link href="css/custom.css" rel="stylesheet">
<script src="https://code.jquery.com/jquery.min.js"></script>
<script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.6/js/bootstrap.min.js"></script>
<!-- slider -->
<script src="https://cdnjs.cloudflare.com/ajax/libs/owl-carousel/1.3.3/owl.carousel.min.js"></script>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/owl-carousel/1.3.3/owl.carousel.css" type="text/css" />
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/owl-carousel/1.3.3/owl.transitions.css" type="text/css" />
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/owl-carousel/1.3.3/owl.theme.min.css" type="text/css" />
<!-- table of contents auto generator -->
<script src="js/toc.js" type="text/javascript"></script>
<!-- styles for pager and table of contents -->
<link rel="stylesheet" href="css/pager.css" type="text/css" />
<link rel="stylesheet" href="css/toc.css" type="text/css" />
<!-- Vega-Lite Embed -->
<script src="https://cdn.jsdelivr.net/npm/vega@5"></script>
<script src="https://cdn.jsdelivr.net/npm/vega-lite@5"></script>
<script src="https://cdn.jsdelivr.net/npm/vega-embed@6"></script>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-57GD08QCML"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-57GD08QCML');
</script>
<!-- Sidebar and testimonial-slider -->
<script type="text/javascript">
$(document).ready(function(){
// scroll/follow sidebar
// #sidebar is defined in the content snippet
// This script has to be executed after the snippet loaded.
// $.getScript("js/follow-sidebar.js");
$("#testimonial-slider").owlCarousel({
items: 1,
singleItem: true,
pagination: true,
navigation: false,
loop: true,
autoPlay: 10000,
stopOnHover: true,
transitionStyle: "backSlide",
touchDrag: true
});
});
</script>
</head>
<body>
<div class="container" style="max-width: 1200px;">
<header>
<div class="masthead">
<p class="lead">
<a href="index.html">
<img src="images/smile.jpg" style="height:100px; width:auto; vertical-align: bottom; margin-top: 20px; margin-right: 20px;">
<span class="tagline">Smile — Statistical Machine Intelligence and Learning Engine</span>
</a>
</p>
</div>
<nav class="navbar navbar-default" role="navigation">
<!-- Brand and toggle get grouped for better mobile display -->
<div class="navbar-header">
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target="#navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
</div>
<!-- Collect the nav links, forms, and other content for toggling -->
<div class="collapse navbar-collapse" id="navbar-collapse">
<ul class="nav navbar-nav">
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Overview <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="quickstart.html">Quick Start</a></li>
<li><a href="overview.html">What's Machine Learning</a></li>
<li><a href="data.html">Data Processing</a></li>
<li><a href="visualization.html">Data Visualization</a></li>
<li><a href="vegalite.html">Declarative Visualization</a></li>
<li><a href="gallery.html">Gallery</a></li>
<li><a href="faq.html">FAQ</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Supervised Learning <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="classification.html">Classification</a></li>
<li><a href="regression.html">Regression</a></li>
<li><a href="deep-learning.html">Deep Learning</a></li>
<li><a href="feature.html">Feature Engineering</a></li>
<li><a href="validation.html">Model Validation</a></li>
<li><a href="missing-value-imputation.html">Missing Value Imputation</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Unsupervised Learning <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="clustering.html">Clustering</a></li>
<li><a href="vector-quantization.html">Vector Quantization</a></li>
<li><a href="association-rule.html">Association Rule Mining</a></li>
<li><a href="mds.html">Multi-Dimensional Scaling</a></li>
<li><a href="manifold.html">Manifold Learning</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">LLM & NLP <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="llm.html">Large Language Model (LLM)</a></li>
<li><a href="nlp.html">Natural Language Processing (NLP)</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Math <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="linear-algebra.html">Linear Algebra</a></li>
<li><a href="statistics.html">Statistics</a></li>
<li><a href="wavelet.html">Wavelet</a></li>
<li><a href="interpolation.html">Interpolation</a></li>
<li><a href="graph.html">Graph Data Structure</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">API <b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="api/java/index.html" target="_blank">Java</a></li>
<li><a href="api/scala/index.html" target="_blank">Scala</a></li>
<li><a href="api/kotlin/index.html" target="_blank">Kotlin</a></li>
<li><a href="api/clojure/index.html" target="_blank">Clojure</a></li>
<li><a href="api/json/index.html" target="_blank">JSON</a></li>
</ul>
</li>
<li><a href="https://mybinder.org/v2/gh/haifengl/smile/notebook?urlpath=lab%2Ftree%2Fshell%2Fsrc%2Funiversal%2Fnotebooks%2Findex.ipynb" target="_blank">Try It Online</a></li>
</ul>
</div>
<!-- /.navbar-collapse -->
</nav>
</header>
<div id="content" class="row">
<div class="col-md-3 col-md-push-9 hidden-xs hidden-sm">
<div id="sidebar">
<div class="sidebar-toc" style="margin-bottom: 20px;">
<p class="toc-header">Contents</p>
<div id="toc"></div>
</div>
<div id="search">
<script>
(function() {
var cx = '010264411143030149390:ajvee_ckdzs';
var gcse = document.createElement('script');
gcse.type = 'text/javascript';
gcse.async = true;
gcse.src = (document.location.protocol == 'https:' ? 'https:' : 'http:') +
'//cse.google.com/cse.js?cx=' + cx;
var s = document.getElementsByTagName('script')[0];
s.parentNode.insertBefore(gcse, s);
})();
</script>
<gcse:searchbox-only></gcse:searchbox-only>
</div>
</div>
</div>
<div class="col-md-9 col-md-pull-3">
<h1 id="manifold-top" class="title">Manifold Learning</h1>
<p>Manifold learning finds a low-dimensional basis for describing
high-dimensional data. Manifold learning is a popular approach to nonlinear
dimensionality reduction. Algorithms for this task are based on the idea
that the dimensionality of many data sets is only artificially high; though
each data point consists of perhaps thousands of features, it may be
described as a function of only a few underlying parameters. That is, the
data points are actually samples from a low-dimensional manifold that is
embedded in a high-dimensional space. Manifold learning algorithms attempt
to uncover these parameters in order to find a low-dimensional representation
of the data.</p>
<p>Some prominent approaches are locally linear embedding
(LLE), Hessian LLE, Laplacian eigenmaps, and LTSA. These techniques
construct a low-dimensional data representation using a cost function
that retains local properties of the data, and can be viewed as defining
a graph-based kernel for Kernel PCA. More recently, techniques have been
proposed that, instead of defining a fixed kernel, try to learn the kernel
using semidefinite programming. The most prominent example of such a
technique is maximum variance unfolding (MVU). The central idea of MVU
is to exactly preserve all pairwise distances between nearest neighbors
(in the inner product space), while maximizing the distances between points
that are not nearest neighbors.</p>
<p>An alternative approach to neighborhood preservation is through the
minimization of a cost function that measures differences between
distances in the input and output spaces. Important examples of such
techniques include classical multidimensional scaling (which is identical
to PCA), Isomap (which uses geodesic distances in the data space), diffusion
maps (which uses diffusion distances in the data space), t-SNE (which
minimizes the divergence between distributions over pairs of points),
and curvilinear component analysis.</p>
<h2 id="isomap">Isomap</h2>
<p>Isometric feature mapping (isomap) is a widely used low-dimensional embedding methods,
where geodesic distances on a weighted graph are incorporated with the
classical multidimensional scaling. Isomap is used for computing a
quasi-isometric, low-dimensional embedding of a set of high-dimensional
data points. Isomap is highly efficient and generally applicable to a broad
range of data sources and dimensionalities.</p>
<p>To be specific, the classical MDS performs low-dimensional embedding based
on the pairwise distance between data points, which is generally measured
using straight-line Euclidean distance. Isomap is distinguished by
its use of the geodesic distance induced by a neighborhood graph
embedded in the classical scaling. This is done to incorporate manifold
structure in the resulting embedding. Isomap defines the geodesic distance
to be the sum of edge weights along the shortest path between two nodes.
The top <code>n</code> eigenvectors of the geodesic distance matrix, represent the
coordinates in the new <code>n</code>-dimensional Euclidean space.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_1" data-toggle="tab">Java</a></li>
<li><a href="#scala_1" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_1" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_1">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
def isomap(data: Array[Array[Double]], k: Int, d: Int = 2, CIsomap: Boolean = true): Array[Array[Double]]
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_1">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
public class IsoMap {
public static double[][] fit(double[][] data, Options options);
public static double[][] fit(NearestNeighborGraph nng, Options options);
}
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_1">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
fun isomap(data: Array<DoubleArray>, k: Int, d: Int = 2, CIsomap: Boolean = true): Array<DoubleArray>
</code></pre>
</div>
</div>
</div>
<p>The connectivity of each data point in the neighborhood graph is defined
as its nearest <code>k</code> Euclidean neighbors in the high-dimensional space. This
step is vulnerable to "short-circuit errors" if <code>k</code> is too large with
respect to the manifold structure or if noise in the data moves the
points slightly off the manifold. Even a single short-circuit error
can alter many entries in the geodesic distance matrix, which in turn
can lead to a drastically different (and incorrect) low-dimensional
embedding. Conversely, if <code>k</code> is too small, the neighborhood graph may
become too sparse to approximate geodesic paths accurately.</p>
<p>When the optional parameter <code>CIsomap</code> is true, the method
performs C-Isomap that involves magnifying the regions
of high density and shrink the regions of low density of data points
in the manifold. Edge weights that are maximized in Multi-Dimensional
Scaling(MDS) are modified, with everything else remaining unaffected.</p>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/swissroll.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">Swiss Roll</div>
</div>
<p>In the below example, we apply Isomap to the famous
<a href="https://people.cs.uchicago.edu/~dinoj/manifold/swissroll.html">swiss roll data</a> as shown above.
This data set was created to test out various dimensionality reduction algorithms.
The idea was to create several points in 2d, and then map them to 3d with some smooth
function, and then to see what the algorithm would find when it mapped the points back to 2d.
The data contains 20,000 samples. Because it is computational intensive to calculate
the shortest path for all samples, the example uses only the first 2,000 samples.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_2" data-toggle="tab">Java</a></li>
<li><a href="#scala_2" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_2" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_2">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
val swissroll = read.csv("data/manifold/swissroll.txt", header=false, delimiter="\t").toArray()
show(plot(swissroll, '.', BLUE))
val data = swissroll.slice(0, 2000)
val map = isomap(data, 7)
show(plot(map, '.', BLUE))
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_2">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
var swissroll = Read.csv("data/manifold/swissroll.txt", CSVFormat.DEFAULT.withDelimiter('\t')).toArray();
ScatterPlot.of(swissroll, '.', Color.BLUE).canvas().window();
var data = Arrays.copyOf(swissroll, 2000);
var nng = NearestNeighborGraph.descent(data, MathEx::distance, 7).largest(false);
var map = IsoMap.fit(nng, 2, false);
var graph = nng.graph(false);
var edges = IntStream.range(0, data.length)
.mapToObj(graph::getEdges)
.flatMap(List::stream)
.map(edge -> new int[]{edge.u(), edge.v()})
.toArray(int[][]::new);
Wireframe.of(map, edges).canvas().window();
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_2">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
import java.awt.Color;
import smile.*;
import smile.manifold.*;
import smile.plot.swing.*;
val swissroll = read.csv("data/manifold/swissroll.txt", header=false, delimiter='\t').toArray();
ScatterPlot.of(swissroll, '.', Color.BLUE).canvas().window();
val data = swissroll.slice(0..2000).toTypedArray()
val map = isomap(data, 7);
ScatterPlot.of(map, '.', Color.BLUE).canvas().window();
</code></pre>
</div>
</div>
</div>
<p>In this example, we use <code>k = 7</code> for neighborhood graph. In the mapped
2d space, we also plot the connections between neighbor neighbors.</p>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/isomap.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">Isomap</div>
</div>
<p>Note that Isomap produces strange holes like in a slice of Swiss cheese :).
This issue can be solved by adding to Isomap a vector quantization step.</p>
<h2 id="lle">LLE</h2>
<p>Locally Linear Embedding (LLE) has several advantages over Isomap, including
faster optimization when implemented to take advantage of sparse matrix
algorithms, and better results with many problems. LLE also begins by
finding a set of the nearest neighbors of each point. It then computes
a set of weights for each point that best describe the point as a linear
combination of its neighbors. Finally, it uses an eigenvector-based
optimization technique to find the low-dimensional embedding of points,
such that each point is still described with the same linear combination
of its neighbors. LLE tends to handle non-uniform sample densities poorly
because there is no fixed unit to prevent the weights from drifting as
various regions differ in sample densities.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_3" data-toggle="tab">Java</a></li>
<li><a href="#scala_3" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_3" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_3">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
def lle(data: Array[Array[Double]], k: Int, d: Int = 2): Array[Array[Double]]
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_3">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
public class LLE {
public static double[][] fit(double[][] data, Options options);
public static double[][] fit(double[][] data, NearestNeighborGraph nng, Options options);
}
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_3">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
fun lle(data: Array<DoubleArray>, k: Int, d: Int = 2): Array<DoubleArray>
</code></pre>
</div>
</div>
</div>
<p>The LLE using 8 neighbors on the swiss roll data is as follows.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_4" data-toggle="tab">Java</a></li>
<li><a href="#scala_4" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_4" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_4">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
val map = lle(data, 2)
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_4">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
var map = LLE.fit(data, nng, 2);
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_4">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
val map = lle(data, 8)
</code></pre>
</div>
</div>
</div>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/lle.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">LLE</div>
</div>
<h2 id="laplacian">Laplacian Eigenmap</h2>
<p>Using the notion of the Laplacian of the nearest
neighbor adjacency graph, Laplacian Eigenmap computes a low dimensional
representation of the dataset that optimally preserves local neighborhood
information in a certain sense. The representation map generated by the
algorithm may be viewed as a discrete approximation to a continuous map
that naturally arises from the geometry of the manifold.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_5" data-toggle="tab">Java</a></li>
<li><a href="#scala_5" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_5" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_5">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
def laplacian(data: Array[Array[Double]], k: Int, d: Int = 2, t: Double = -1): Array[Array[Double]]
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_5">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
public class LaplacianEigenmap {
public static double[][] fit(double[][] data, Options options);
public static double[][] fit(NearestNeighborGraph nng, Options options);
}
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_5">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
fun laplacian(data: Array<DoubleArray>, k: Int, d: Int = 2, t: Double = -1): Array<DoubleArray>
</code></pre>
</div>
</div>
</div>
<p>where <code>t</code> is the smooth/width parameter of heat kernel
<code>e<sup>-||x-y||<sup>2</sup>/t</sup></code>.
Non-positive <code>t</code> means discrete weights.</p>
<p>The locality preserving character of the Laplacian Eigenmap algorithm makes
it relatively insensitive to outliers and noise. It is also not prone to
"short-circuiting" as only the local distances are used.</p>
<p>The Laplacian Eigenmap on the swiss roll data is as follows.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_6" data-toggle="tab">Java</a></li>
<li><a href="#scala_6" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_6" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_6">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
val map = laplacian(swissroll.slice(0, 1000), 10, 2, 25.0)
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_6">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
var map = LaplacianEigenmap.fit(nng, 2, 25.0);
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_6">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kolin"><code>
val map = laplacian(swissroll.slice(0..1000).toTypedArray(), 10, 2, 25.0)
</code></pre>
</div>
</div>
</div>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/laplacian.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">Laplacian Eigenmap</div>
</div>
<h2 id="t-sne">t-SNE</h2>
<p>The t-distributed stochastic neighbor embedding (t-SNE) is a nonlinear
dimensionality reduction technique that is particularly well suited
for embedding high-dimensional data into a space of two or three
dimensions, which can then be visualized in a scatter plot. Specifically,
it models each high-dimensional object by a two- or three-dimensional
point in such a way that similar objects are modeled by nearby points
and dissimilar objects are modeled by distant points.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_7" data-toggle="tab">Java</a></li>
<li><a href="#scala_7" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_7" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_7">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
def tsne(X: Array[Array[Double]],
d: Int = 2,
perplexity: Double = 20.0,
eta: Double = 200.0,
iterations: Int = 1000): TSNE
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_7">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
public class TSNE {
public static TSNE fit(double[][] X, Options options);
}
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_7">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
fun tsne(X: Array<DoubleArray>,
d: Int = 2,
perplexity: Double = 20.0,
eta: Double = 200.0,
iterations: Int = 1000): TSNE
</code></pre>
</div>
</div>
</div>
<p>where <code>X</code> is input data, <code>perplexity</code> is the perplexity
of the conditional distribution, <code>eta</code> the learning rate, and <code>iterations</code> is
the number of iterations. If <code>X</code> is a square matrix, it is assumed
to be the distance/dissimilarity matrix.</p>
<p>The t-SNE on the MNIST data is as follows. Note that the input data is preprocessed
using PCA to reduce the dimensionality to 50 before t-SNE is performed.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_8" data-toggle="tab">Java</a></li>
<li><a href="#scala_8" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_8" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_8">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
val mnist = read.csv("data/mnist/mnist2500_X.txt", header=false, delimiter=" ")
val label = read.csv("data/mnist/mnist2500_labels.txt", header=false, delimiter=" ").column(0).toIntArray()
val pc = pca(mnist).getProjection(50)
val x50 = pc(mnist).toArray()
val sne = tsne(x50, 3, 20, 200, 1000)
show(plot(sne.coordinates, label, 'o'))
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_8">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
var format = CSVFormat.DEFAULT.withDelimiter(' ');
var mnist = Read.csv("data/mnist/mnist2500_X.txt", format).toArray();
var label = Read.csv("data/mnist/mnist2500_labels.txt", format).column(0).toIntArray();
var pca = PCA.fit(mnist).getProjection(50);
var X = pca.apply(mnist);
var perplexity = 20;
var tsne = new TSNE(X, 2, perplexity, 200, 1000);
var canvas = ScatterPlot.of(tsne.coordinates(), label, '@').canvas();
canvas.setTitle("t-SNE of MNIST");
canvas.window();
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_8">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
import smile.feature.extraction.*;
val mnist = read.csv("data/mnist/mnist2500_X.txt", header=false, delimiter=' ').toArray();
val label = read.csv("data/mnist/mnist2500_labels.txt", header=false, delimiter=' ').column(0).toIntArray();
val pc = pca(mnist).getProjection(50);
val x50 = pc.apply(mnist);
val sne = tsne(x50, 3, 20.0, 200.0, 1000);
val canvas = ScatterPlot.of(sne.coordinates, label, '@').canvas();
canvas.setTitle("t-SNE of MNIST");
canvas.window();
</code></pre>
</div>
</div>
</div>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/tsne.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">t-SNE</div>
</div>
<h2 id="umap">UMAP</h2>
<p>Uniform Manifold Approximation and Projection can be used for
visualization similarly to t-SNE, but also for general non-linear
dimension reduction. The algorithm is founded on three assumptions
about the data:</p>
<ul>
<li>The data is uniformly distributed on a Riemannian manifold;</li>
<li>The Riemannian metric is locally constant (or can be approximated as such);</li>
<li>The manifold is locally connected.</li>
</ul>
<p>From these assumptions it is possible to model the manifold with a fuzzy
topological structure. The embedding is found by searching for a low
dimensional projection of the data that has the closest possible
equivalent fuzzy topological structure.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_9" data-toggle="tab">Java</a></li>
<li><a href="#scala_9" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_9" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_9">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
def umap(data: Array[Array[Double]],
k: Int = 15,
d: Int = 2,
epochs: Int = 0,
learningRate: Double = 1.0,
minDist: Double = 0.1,
spread: Double = 1.0,
negativeSamples: Int = 5,
repulsionStrength: Double = 1.0,
localConnectivity: Double = 1.0): Array[Array[Double]]
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_9">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
public class UAMP {
public double[][] fit(double[][] data, Options options);
public double[][] fit(T[] data, Metric>T< distance, Options options);
public double[][] fit(T[] data, NearestNeighborGraph nng, Options options);
}
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_9">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
fun umap(data: Array<DoubleArray>,
k: Int = 15,
d: Int = 2,
epochs: Int = 0,
learningRate: Double = 1.0,
minDist: Double = 0.1,
spread: Double = 1.0,
negativeSamples: Int = 5,
repulsionStrength: Double = 1.0,
localConnectivity: Double = 1.0): Array<DoubleArray>
</code></pre>
</div>
</div>
</div>
<p>where <code>data</code> is the input data, <code>k</code> is of
k-nearest neighbors. Larger values result in more global views
of the manifold, while smaller values result in more local data
being preserved. <code>d</code> is the target embedding dimensions.
<code>iterations</code> is the number of iterations to optimize the
low-dimensional representation. Larger values result in more
accurate embedding. Muse be greater than 10. Choose wise value
based on the size of the input data, e.g, 200 for large
data (1000+ samples), 500 for small.
<code>learningRate</code> is the initial learning rate for the
embedding optimization.
<code>minDist</code> is the desired separation between close points
in the embedding space. Smaller values will result in a more
clustered/clumped embedding where nearby points on the manifold are
drawn closer together, while larger values will result on a more even
disperse of points. The value should be set no-greater than
and relative to the spread value, which determines the scale
at which embedded points will be spread out.
<code>spread</code> is the effective scale of embedded points.
In combination with <code>minDist</code>, this determines how
clustered/clumped the embedded points are.
<code>negativeSamples</code> is the number of negative samples to
select per positive sample in the optimization process. Increasing
this value will result in greater repulsive force being applied,
greater optimization cost, but slightly more accuracy.
<code>repulsionStrength</code>plied to negative samples in low
dimensional embedding optimization. Values higher than one will result
in greater weight being given to negative samples.</p>
<p>UMAP on the MNIST data is as follows.</p>
<ul class="nav nav-tabs">
<li class="active"><a href="#java_10" data-toggle="tab">Java</a></li>
<li><a href="#scala_10" data-toggle="tab">Scala</a></li>
<li><a href="#kotlin_10" data-toggle="tab">Kotlin</a></li>
</ul>
<div class="tab-content">
<div class="tab-pane" id="scala_10">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-scala"><code>
val map = umap(mnist.toArray())
show(plot(map, label, 'o'))
</code></pre>
</div>
</div>
<div class="tab-pane active" id="java_10">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-java"><code>
var map = UMAP.fit(mnist, 15);
ScatterPlot.of(map, label, '@').canvas().window();
</code></pre>
</div>
</div>
<div class="tab-pane" id="kotlin_10">
<div class="code" style="text-align: left;">
<pre class="prettyprint lang-kotlin"><code>
val map = umap(mnist);
ScatterPlot.of(map, label, '@').canvas().window();
</code></pre>
</div>
</div>
</div>
<div style="width: 100%; display: inline-block; text-align: center;">
<img src="images/umap.png" class="enlarge" style="width: 480px;" />
<div class="caption" style="min-width: 480px;">UMAP</div>
</div>
<div id="btnv">
<span class="btn-arrow-left">← </span>
<a class="btn-prev-text" href="mds.html" title="Previous Section: Multi-Dimensional Scaling"><span>Multi-Dimensional Scaling</span></a>
<a class="btn-next-text" href="linear-algebra.html" title="Next Section: Linear Algebra"><span>Linear Algebra</span></a>
<span class="btn-arrow-right"> →</span>
</div>
</div>
<script type="text/javascript">
$('#toc').toc({exclude: 'h1, h5, h6', context: '', autoId: true, numerate: false});
</script>
</div>
</div>
<a href=https://github.com/haifengl/smile><img style="position: fixed; top: 0; right: 0; border: 0" src=/images/forkme_right_orange.png alt="Fork me on GitHub"></a>
<!-- Place this tag right after the last button or just before your close body tag. -->
<script async defer id="github-bjs" src="https://buttons.github.io/buttons.js"></script>
</body>
</html>