1
|
|
|
<?php |
2
|
|
|
/** |
3
|
|
|
* |
4
|
|
|
* (c) Ruben Dorado <[email protected]> |
5
|
|
|
* |
6
|
|
|
* For the full copyright and license information, please view the LICENSE |
7
|
|
|
* file that was distributed with this source code. |
8
|
|
|
*/ |
9
|
|
|
namespace SiteAnalyzer; |
10
|
|
|
|
11
|
|
|
use Exception; |
12
|
|
|
|
13
|
|
|
/** |
14
|
|
|
* class ML |
15
|
|
|
* |
16
|
|
|
* @package SiteAnalyzer |
17
|
|
|
* @author Ruben Dorado <[email protected]> |
18
|
|
|
* @copyright 2018 Ruben Dorado |
19
|
|
|
* @license http://www.opensource.org/licenses/MIT The MIT License |
20
|
|
|
*/ |
21
|
|
|
class ML |
22
|
|
|
{ |
23
|
|
|
|
24
|
|
|
/* |
25
|
|
|
* @param |
26
|
|
|
*/ |
27
|
|
|
public static function kmeans($data, $nclusters) |
28
|
|
|
{ |
29
|
|
|
$resp = []; |
30
|
|
|
$finished = false; |
31
|
|
|
$niter = 0; |
32
|
|
|
$maxiter = 100; |
33
|
|
|
$npoints = count($data); |
34
|
|
|
if ($npoints <= 0) throw new \Exception("Not enough data. "); |
35
|
|
|
$ndimensions = count($data[0]); |
36
|
|
|
$centroids = self::select_disjoint($data, $nclusters); |
37
|
|
|
|
38
|
|
|
while (!$finished && $niter < $maxiter) { |
39
|
|
|
// Assign each one of the points to one centroid |
40
|
|
|
$niter++; |
41
|
|
|
$nresp = []; |
42
|
|
|
for ($j = 0; $j < $npoints; $j++) { |
43
|
|
|
$best = -1; |
44
|
|
|
$bdist = INF; |
45
|
|
|
for ($i = 0; $i < $nclusters; $i++) { |
46
|
|
|
$ndist = self::eclideanDistance($data[$j], $centroids[$i]); |
47
|
|
|
if($bdist > $ndist) { |
48
|
|
|
$bdist = $ndist; |
49
|
|
|
$best = $i; |
50
|
|
|
} |
51
|
|
|
} |
52
|
|
|
$nresp[] = $best; |
53
|
|
|
|
54
|
|
|
} |
55
|
|
|
|
56
|
|
|
// Check change |
57
|
|
|
$finished = true; |
58
|
|
|
if (count($resp) > 0) { |
59
|
|
|
for ($j=0; $j < $npoints; $j++) { |
60
|
|
|
if ($resp[$j]!==$nresp[$j]) { |
61
|
|
|
$finished = false; |
62
|
|
|
break; |
63
|
|
|
} |
64
|
|
|
} |
65
|
|
|
} else { |
66
|
|
|
$finished = false; |
67
|
|
|
} |
68
|
|
|
$resp = $nresp; |
69
|
|
|
// Recalculate the centroids |
70
|
|
|
$centroids = self::initCentroids($nclusters, $ndimensions, function(){return 0;}); |
71
|
|
|
$counts = array_fill(0, $nclusters, 0); |
72
|
|
|
for ($j = 0; $j < $npoints; $j++) { |
73
|
|
|
$centroids[$resp[$j]] = Matrix::sumArray($centroids[$resp[$j]], $data[$j]); |
74
|
|
|
$counts[$resp[$j]]++; |
75
|
|
|
} |
76
|
|
|
$centroids = self::normalizeCentroids($centroids, $counts); |
77
|
|
|
} |
78
|
|
|
return ["clusters"=>$resp, "centroids"=>$centroids]; |
79
|
|
|
} |
80
|
|
|
|
81
|
|
|
|
82
|
|
|
/* |
83
|
|
|
* @param |
84
|
|
|
*/ |
85
|
|
|
private static function select_disjoint($data, $n) |
86
|
|
|
{ |
87
|
|
|
$resp = []; |
88
|
|
|
foreach ($data as $row) { |
89
|
|
|
if ( !self::contains_point($resp, $row) ) { |
90
|
|
|
$resp[] = $row; |
91
|
|
|
} |
92
|
|
|
if (count($resp) == $n) { |
93
|
|
|
return $resp; |
94
|
|
|
} |
95
|
|
|
} |
96
|
|
|
throw new \Exception("Not enough unique points."); |
97
|
|
|
} |
98
|
|
|
|
99
|
|
|
/* |
100
|
|
|
* @param |
101
|
|
|
*/ |
102
|
|
|
private static function contains_point($matrix, $array) |
103
|
|
|
{ |
104
|
|
|
foreach ($matrix as $row){ |
105
|
|
|
if (self::isEqual($row, $array)) return true; |
106
|
|
|
} |
107
|
|
|
return false; |
108
|
|
|
} |
109
|
|
|
|
110
|
|
|
/* |
111
|
|
|
* @param |
112
|
|
|
*/ |
113
|
|
|
private static function isEqual($array1, $array2) |
114
|
|
|
{ |
115
|
|
|
$len = count($array1); |
116
|
|
|
if($len != count($array2) ) return false; |
117
|
|
|
for ($i=0; $i<$len; $i++) { |
118
|
|
|
if ($array1[$i] != $array2[$i]) return false; |
119
|
|
|
} |
120
|
|
|
return true; |
121
|
|
|
} |
122
|
|
|
|
123
|
|
|
/* |
124
|
|
|
* @param |
125
|
|
|
*/ |
126
|
|
|
private static function normalizeCentroids($centroids, $counts) |
127
|
|
|
{ |
128
|
|
|
$resp = []; |
129
|
|
|
$n = count($centroids); |
130
|
|
|
$d = count($centroids[0]); |
131
|
|
|
for ($i=0;$i<$n;$i++) { |
132
|
|
|
$tmp = []; |
133
|
|
|
for ($j=0;$j<$d;$j++){ |
134
|
|
|
$tmp[] = $centroids[$i][$j]/$counts[$i]; |
135
|
|
|
} |
136
|
|
|
$resp[] = $tmp; |
137
|
|
|
} |
138
|
|
|
return $resp; |
139
|
|
|
} |
140
|
|
|
|
141
|
|
|
/* |
142
|
|
|
* @param |
143
|
|
|
*/ |
144
|
|
|
public static function initCentroids($nclusters, $ndimensions, $fvalue) |
145
|
|
|
{ |
146
|
|
|
$resp = []; |
147
|
|
|
for ($i = 0; $i < $nclusters; $i++) { |
148
|
|
|
$centroid = []; |
149
|
|
|
for ($d = 0; $d < $ndimensions; $d++) { |
150
|
|
|
$centroid[] = $fvalue(); |
151
|
|
|
} |
152
|
|
|
$resp[] = $centroid; |
153
|
|
|
} |
154
|
|
|
return $resp; |
155
|
|
|
} |
156
|
|
|
|
157
|
|
|
/* |
158
|
|
|
* @param |
159
|
|
|
*/ |
160
|
|
|
public static function eclideanDistance($p1, $p2) { |
161
|
|
|
$len = count($p1); |
162
|
|
|
$acum = 0; |
163
|
|
|
for($i=0; $i<$len; $i++) { |
164
|
|
|
$acum += ($p1[$i] - $p2[$i])**2; |
165
|
|
|
} |
166
|
|
|
return sqrt($acum); |
167
|
|
|
} |
168
|
|
|
|
169
|
|
|
} |
170
|
|
|
|