1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\DimensionReduction; |
6
|
|
|
|
7
|
|
|
use Exception; |
8
|
|
|
use Phpml\Math\Matrix; |
9
|
|
|
|
10
|
|
|
class LDA extends EigenTransformerBase |
11
|
|
|
{ |
12
|
|
|
/** |
13
|
|
|
* @var bool |
14
|
|
|
*/ |
15
|
|
|
public $fit = false; |
16
|
|
|
|
17
|
|
|
/** |
18
|
|
|
* @var array |
19
|
|
|
*/ |
20
|
|
|
public $labels = []; |
21
|
|
|
|
22
|
|
|
/** |
23
|
|
|
* @var array |
24
|
|
|
*/ |
25
|
|
|
public $means = []; |
26
|
|
|
|
27
|
|
|
/** |
28
|
|
|
* @var array |
29
|
|
|
*/ |
30
|
|
|
public $counts = []; |
31
|
|
|
|
32
|
|
|
/** |
33
|
|
|
* @var float[] |
34
|
|
|
*/ |
35
|
|
|
public $overallMean = []; |
36
|
|
|
|
37
|
|
|
/** |
38
|
|
|
* Linear Discriminant Analysis (LDA) is used to reduce the dimensionality |
39
|
|
|
* of the data. Unlike Principal Component Analysis (PCA), it is a supervised |
40
|
|
|
* technique that requires the class labels in order to fit the data to a |
41
|
|
|
* lower dimensional space. <br><br> |
42
|
|
|
* The algorithm can be initialized by speciyfing |
43
|
|
|
* either with the totalVariance(a value between 0.1 and 0.99) |
44
|
|
|
* or numFeatures (number of features in the dataset) to be preserved. |
45
|
|
|
* |
46
|
|
|
* @param float|null $totalVariance Total explained variance to be preserved |
47
|
|
|
* @param int|null $numFeatures Number of features to be preserved |
48
|
|
|
* |
49
|
|
|
* @throws \Exception |
50
|
|
|
*/ |
51
|
|
|
public function __construct(?float $totalVariance = null, ?int $numFeatures = null) |
52
|
|
|
{ |
53
|
|
|
if ($totalVariance !== null && ($totalVariance < 0.1 || $totalVariance > 0.99)) { |
54
|
|
|
throw new Exception('Total variance can be a value between 0.1 and 0.99'); |
55
|
|
|
} |
56
|
|
|
|
57
|
|
|
if ($numFeatures !== null && $numFeatures <= 0) { |
58
|
|
|
throw new Exception('Number of features to be preserved should be greater than 0'); |
59
|
|
|
} |
60
|
|
|
|
61
|
|
|
if ($totalVariance !== null && $numFeatures !== null) { |
62
|
|
|
throw new Exception('Either totalVariance or numFeatures should be specified in order to run the algorithm'); |
63
|
|
|
} |
64
|
|
|
|
65
|
|
|
if ($numFeatures !== null) { |
66
|
|
|
$this->numFeatures = $numFeatures; |
67
|
|
|
} |
68
|
|
|
|
69
|
|
|
if ($totalVariance !== null) { |
70
|
|
|
$this->totalVariance = $totalVariance; |
71
|
|
|
} |
72
|
|
|
} |
73
|
|
|
|
74
|
|
|
/** |
75
|
|
|
* Trains the algorithm to transform the given data to a lower dimensional space. |
76
|
|
|
*/ |
77
|
|
|
public function fit(array $data, array $classes): array |
78
|
|
|
{ |
79
|
|
|
$this->labels = $this->getLabels($classes); |
80
|
|
|
$this->means = $this->calculateMeans($data, $classes); |
81
|
|
|
|
82
|
|
|
$sW = $this->calculateClassVar($data, $classes); |
83
|
|
|
$sB = $this->calculateClassCov(); |
84
|
|
|
|
85
|
|
|
$S = $sW->inverse()->multiply($sB); |
86
|
|
|
$this->eigenDecomposition($S->toArray()); |
87
|
|
|
|
88
|
|
|
$this->fit = true; |
89
|
|
|
|
90
|
|
|
return $this->reduce($data); |
91
|
|
|
} |
92
|
|
|
|
93
|
|
|
/** |
94
|
|
|
* Transforms the given sample to a lower dimensional vector by using |
95
|
|
|
* the eigenVectors obtained in the last run of <code>fit</code>. |
96
|
|
|
* |
97
|
|
|
* @throws \Exception |
98
|
|
|
*/ |
99
|
|
|
public function transform(array $sample): array |
100
|
|
|
{ |
101
|
|
|
if (!$this->fit) { |
102
|
|
|
throw new Exception('LDA has not been fitted with respect to original dataset, please run LDA::fit() first'); |
103
|
|
|
} |
104
|
|
|
|
105
|
|
|
if (!is_array($sample[0])) { |
106
|
|
|
$sample = [$sample]; |
107
|
|
|
} |
108
|
|
|
|
109
|
|
|
return $this->reduce($sample); |
110
|
|
|
} |
111
|
|
|
|
112
|
|
|
/** |
113
|
|
|
* Returns unique labels in the dataset |
114
|
|
|
*/ |
115
|
|
|
protected function getLabels(array $classes): array |
116
|
|
|
{ |
117
|
|
|
$counts = array_count_values($classes); |
118
|
|
|
|
119
|
|
|
return array_keys($counts); |
120
|
|
|
} |
121
|
|
|
|
122
|
|
|
/** |
123
|
|
|
* Calculates mean of each column for each class and returns |
124
|
|
|
* n by m matrix where n is number of labels and m is number of columns |
125
|
|
|
*/ |
126
|
|
|
protected function calculateMeans(array $data, array $classes): array |
127
|
|
|
{ |
128
|
|
|
$means = []; |
129
|
|
|
$counts = []; |
130
|
|
|
$overallMean = array_fill(0, count($data[0]), 0.0); |
131
|
|
|
|
132
|
|
|
foreach ($data as $index => $row) { |
133
|
|
|
$label = array_search($classes[$index], $this->labels); |
134
|
|
|
|
135
|
|
|
foreach ($row as $col => $val) { |
136
|
|
|
if (!isset($means[$label][$col])) { |
137
|
|
|
$means[$label][$col] = 0.0; |
138
|
|
|
} |
139
|
|
|
|
140
|
|
|
$means[$label][$col] += $val; |
141
|
|
|
$overallMean[$col] += $val; |
142
|
|
|
} |
143
|
|
|
|
144
|
|
|
if (!isset($counts[$label])) { |
145
|
|
|
$counts[$label] = 0; |
146
|
|
|
} |
147
|
|
|
|
148
|
|
|
++$counts[$label]; |
149
|
|
|
} |
150
|
|
|
|
151
|
|
|
foreach ($means as $index => $row) { |
152
|
|
|
foreach ($row as $col => $sum) { |
153
|
|
|
$means[$index][$col] = $sum / $counts[$index]; |
154
|
|
|
} |
155
|
|
|
} |
156
|
|
|
|
157
|
|
|
// Calculate overall mean of the dataset for each column |
158
|
|
|
$numElements = array_sum($counts); |
159
|
|
|
$map = function ($el) use ($numElements) { |
160
|
|
|
return $el / $numElements; |
161
|
|
|
}; |
162
|
|
|
$this->overallMean = array_map($map, $overallMean); |
163
|
|
|
$this->counts = $counts; |
164
|
|
|
|
165
|
|
|
return $means; |
166
|
|
|
} |
167
|
|
|
|
168
|
|
|
/** |
169
|
|
|
* Returns in-class scatter matrix for each class, which |
170
|
|
|
* is a n by m matrix where n is number of classes and |
171
|
|
|
* m is number of columns |
172
|
|
|
*/ |
173
|
|
|
protected function calculateClassVar(array $data, array $classes): Matrix |
174
|
|
|
{ |
175
|
|
|
// s is an n (number of classes) by m (number of column) matrix |
176
|
|
|
$s = array_fill(0, count($data[0]), array_fill(0, count($data[0]), 0)); |
177
|
|
|
$sW = new Matrix($s, false); |
178
|
|
|
|
179
|
|
|
foreach ($data as $index => $row) { |
180
|
|
|
$label = array_search($classes[$index], $this->labels); |
181
|
|
|
$means = $this->means[$label]; |
182
|
|
|
|
183
|
|
|
$row = $this->calculateVar($row, $means); |
184
|
|
|
|
185
|
|
|
$sW = $sW->add($row); |
186
|
|
|
} |
187
|
|
|
|
188
|
|
|
return $sW; |
189
|
|
|
} |
190
|
|
|
|
191
|
|
|
/** |
192
|
|
|
* Returns between-class scatter matrix for each class, which |
193
|
|
|
* is an n by m matrix where n is number of classes and |
194
|
|
|
* m is number of columns |
195
|
|
|
*/ |
196
|
|
|
protected function calculateClassCov(): Matrix |
197
|
|
|
{ |
198
|
|
|
// s is an n (number of classes) by m (number of column) matrix |
199
|
|
|
$s = array_fill(0, count($this->overallMean), array_fill(0, count($this->overallMean), 0)); |
200
|
|
|
$sB = new Matrix($s, false); |
201
|
|
|
|
202
|
|
|
foreach ($this->means as $index => $classMeans) { |
203
|
|
|
$row = $this->calculateVar($classMeans, $this->overallMean); |
204
|
|
|
$N = $this->counts[$index]; |
205
|
|
|
$sB = $sB->add($row->multiplyByScalar($N)); |
206
|
|
|
} |
207
|
|
|
|
208
|
|
|
return $sB; |
209
|
|
|
} |
210
|
|
|
|
211
|
|
|
/** |
212
|
|
|
* Returns the result of the calculation (x - m)T.(x - m) |
213
|
|
|
*/ |
214
|
|
|
protected function calculateVar(array $row, array $means): Matrix |
215
|
|
|
{ |
216
|
|
|
$x = new Matrix($row, false); |
217
|
|
|
$m = new Matrix($means, false); |
218
|
|
|
$diff = $x->subtract($m); |
|
|
|
|
219
|
|
|
|
220
|
|
|
return $diff->transpose()->multiply($diff); |
221
|
|
|
} |
222
|
|
|
} |
223
|
|
|
|
It seems like the type of the argument is not accepted by the function/method which you are calling.
In some cases, in particular if PHP’s automatic type-juggling kicks in this might be fine. In other cases, however this might be a bug.
We suggest to add an explicit type cast like in the following example: