1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\Preprocessing; |
6
|
|
|
|
7
|
|
|
use Phpml\Exception\NormalizerException; |
8
|
|
|
use Phpml\Math\Statistic\Mean; |
9
|
|
|
use Phpml\Math\Statistic\StandardDeviation; |
10
|
|
|
|
11
|
|
|
class Normalizer implements Preprocessor |
12
|
|
|
{ |
13
|
|
|
public const NORM_L1 = 1; |
14
|
|
|
|
15
|
|
|
public const NORM_L2 = 2; |
16
|
|
|
|
17
|
|
|
public const NORM_STD = 3; |
18
|
|
|
|
19
|
|
|
/** |
20
|
|
|
* @var int |
21
|
|
|
*/ |
22
|
|
|
private $norm; |
23
|
|
|
|
24
|
|
|
/** |
25
|
|
|
* @var bool |
26
|
|
|
*/ |
27
|
|
|
private $fitted = false; |
28
|
|
|
|
29
|
|
|
/** |
30
|
|
|
* @var array |
31
|
|
|
*/ |
32
|
|
|
private $std = []; |
33
|
|
|
|
34
|
|
|
/** |
35
|
|
|
* @var array |
36
|
|
|
*/ |
37
|
|
|
private $mean = []; |
38
|
|
|
|
39
|
|
|
/** |
40
|
|
|
* @throws NormalizerException |
41
|
|
|
*/ |
42
|
|
|
public function __construct(int $norm = self::NORM_L2) |
43
|
|
|
{ |
44
|
|
|
if (!in_array($norm, [self::NORM_L1, self::NORM_L2, self::NORM_STD], true)) { |
45
|
|
|
throw new NormalizerException('Unknown norm supplied.'); |
46
|
|
|
} |
47
|
|
|
|
48
|
|
|
$this->norm = $norm; |
49
|
|
|
} |
50
|
|
|
|
51
|
|
|
public function fit(array $samples, ?array $targets = null): void |
52
|
|
|
{ |
53
|
|
|
if ($this->fitted) { |
54
|
|
|
return; |
55
|
|
|
} |
56
|
|
|
|
57
|
|
|
if ($this->norm === self::NORM_STD) { |
58
|
|
|
$features = range(0, count($samples[0]) - 1); |
59
|
|
|
foreach ($features as $i) { |
60
|
|
|
$values = array_column($samples, $i); |
61
|
|
|
$this->std[$i] = StandardDeviation::population($values); |
62
|
|
|
$this->mean[$i] = Mean::arithmetic($values); |
63
|
|
|
} |
64
|
|
|
} |
65
|
|
|
|
66
|
|
|
$this->fitted = true; |
67
|
|
|
} |
68
|
|
|
|
69
|
|
|
public function transform(array &$samples, ?array &$targets = null): void |
70
|
|
|
{ |
71
|
|
|
$methods = [ |
72
|
|
|
self::NORM_L1 => 'normalizeL1', |
73
|
|
|
self::NORM_L2 => 'normalizeL2', |
74
|
|
|
self::NORM_STD => 'normalizeSTD', |
75
|
|
|
]; |
76
|
|
|
$method = $methods[$this->norm]; |
77
|
|
|
|
78
|
|
|
$this->fit($samples); |
79
|
|
|
|
80
|
|
|
foreach ($samples as &$sample) { |
81
|
|
|
$this->{$method}($sample); |
82
|
|
|
} |
83
|
|
|
} |
84
|
|
|
|
85
|
|
|
private function normalizeL1(array &$sample): void |
86
|
|
|
{ |
87
|
|
|
$norm1 = 0; |
88
|
|
|
foreach ($sample as $feature) { |
89
|
|
|
$norm1 += abs($feature); |
90
|
|
|
} |
91
|
|
|
|
92
|
|
|
if ($norm1 == 0) { |
|
|
|
|
93
|
|
|
$count = count($sample); |
94
|
|
|
$sample = array_fill(0, $count, 1.0 / $count); |
95
|
|
|
} else { |
96
|
|
|
array_walk($sample, function (&$feature) use ($norm1): void { |
97
|
|
|
$feature /= $norm1; |
98
|
|
|
}); |
99
|
|
|
} |
100
|
|
|
} |
101
|
|
|
|
102
|
|
|
private function normalizeL2(array &$sample): void |
103
|
|
|
{ |
104
|
|
|
$norm2 = 0; |
105
|
|
|
foreach ($sample as $feature) { |
106
|
|
|
$norm2 += $feature * $feature; |
107
|
|
|
} |
108
|
|
|
|
109
|
|
|
$norm2 **= .5; |
110
|
|
|
|
111
|
|
|
if ($norm2 == 0) { |
|
|
|
|
112
|
|
|
$sample = array_fill(0, count($sample), 1); |
113
|
|
|
} else { |
114
|
|
|
array_walk($sample, function (&$feature) use ($norm2): void { |
115
|
|
|
$feature /= $norm2; |
116
|
|
|
}); |
117
|
|
|
} |
118
|
|
|
} |
119
|
|
|
|
120
|
|
|
private function normalizeSTD(array &$sample): void |
121
|
|
|
{ |
122
|
|
|
foreach (array_keys($sample) as $i) { |
123
|
|
|
if ($this->std[$i] != 0) { |
124
|
|
|
$sample[$i] = ($sample[$i] - $this->mean[$i]) / $this->std[$i]; |
125
|
|
|
} else { |
126
|
|
|
// Same value for all samples. |
127
|
|
|
$sample[$i] = 0; |
128
|
|
|
} |
129
|
|
|
} |
130
|
|
|
} |
131
|
|
|
} |
132
|
|
|
|