|
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])) { |
|
45
|
|
|
throw NormalizerException::unknownNorm(); |
|
46
|
|
|
} |
|
47
|
|
|
|
|
48
|
|
|
$this->norm = $norm; |
|
49
|
|
|
} |
|
50
|
|
|
|
|
51
|
|
|
/** |
|
52
|
|
|
* @param array $samples |
|
53
|
|
|
*/ |
|
54
|
|
|
public function preprocess(array $samples) |
|
55
|
|
|
{ |
|
56
|
|
|
$this->fit($samples); |
|
57
|
|
|
} |
|
58
|
|
|
|
|
59
|
|
|
public function fit(array $samples): void |
|
60
|
|
|
{ |
|
61
|
|
|
if ($this->fitted) { |
|
62
|
|
|
return; |
|
63
|
|
|
} |
|
64
|
|
|
|
|
65
|
|
|
if ($this->norm == self::NORM_STD) { |
|
66
|
|
|
$features = range(0, count($samples[0]) - 1); |
|
67
|
|
|
foreach ($features as $i) { |
|
68
|
|
|
$values = array_column($samples, $i); |
|
69
|
|
|
$this->std[$i] = StandardDeviation::population($values); |
|
70
|
|
|
$this->mean[$i] = Mean::arithmetic($values); |
|
71
|
|
|
} |
|
72
|
|
|
} |
|
73
|
|
|
|
|
74
|
|
|
$this->fitted = true; |
|
75
|
|
|
} |
|
76
|
|
|
|
|
77
|
|
|
public function transform(array &$samples): void |
|
78
|
|
|
{ |
|
79
|
|
|
$methods = [ |
|
80
|
|
|
self::NORM_L1 => 'normalizeL1', |
|
81
|
|
|
self::NORM_L2 => 'normalizeL2', |
|
82
|
|
|
self::NORM_STD => 'normalizeSTD', |
|
83
|
|
|
]; |
|
84
|
|
|
$method = $methods[$this->norm]; |
|
85
|
|
|
|
|
86
|
|
|
$this->fit($samples); |
|
87
|
|
|
|
|
88
|
|
|
foreach ($samples as &$sample) { |
|
89
|
|
|
$this->{$method}($sample); |
|
90
|
|
|
} |
|
91
|
|
|
} |
|
92
|
|
|
|
|
93
|
|
|
private function normalizeL1(array &$sample): void |
|
94
|
|
|
{ |
|
95
|
|
|
$norm1 = 0; |
|
96
|
|
|
foreach ($sample as $feature) { |
|
97
|
|
|
$norm1 += abs($feature); |
|
98
|
|
|
} |
|
99
|
|
|
|
|
100
|
|
|
if ($norm1 == 0) { |
|
101
|
|
|
$count = count($sample); |
|
102
|
|
|
$sample = array_fill(0, $count, 1.0 / $count); |
|
103
|
|
|
} else { |
|
104
|
|
|
foreach ($sample as &$feature) { |
|
105
|
|
|
$feature /= $norm1; |
|
106
|
|
|
} |
|
107
|
|
|
} |
|
108
|
|
|
} |
|
109
|
|
|
|
|
110
|
|
|
private function normalizeL2(array &$sample): void |
|
111
|
|
|
{ |
|
112
|
|
|
$norm2 = 0; |
|
113
|
|
|
foreach ($sample as $feature) { |
|
114
|
|
|
$norm2 += $feature * $feature; |
|
115
|
|
|
} |
|
116
|
|
|
|
|
117
|
|
|
$norm2 = sqrt((float) $norm2); |
|
118
|
|
|
|
|
119
|
|
|
if ($norm2 == 0) { |
|
120
|
|
|
$sample = array_fill(0, count($sample), 1); |
|
121
|
|
|
} else { |
|
122
|
|
|
foreach ($sample as &$feature) { |
|
123
|
|
|
$feature /= $norm2; |
|
124
|
|
|
} |
|
125
|
|
|
} |
|
126
|
|
|
} |
|
127
|
|
|
|
|
128
|
|
|
private function normalizeSTD(array &$sample): void |
|
129
|
|
|
{ |
|
130
|
|
|
foreach ($sample as $i => $val) { |
|
131
|
|
|
if ($this->std[$i] != 0) { |
|
132
|
|
|
$sample[$i] = ($sample[$i] - $this->mean[$i]) / $this->std[$i]; |
|
133
|
|
|
} else { |
|
134
|
|
|
// Same value for all samples. |
|
135
|
|
|
$sample[$i] = 0; |
|
136
|
|
|
} |
|
137
|
|
|
} |
|
138
|
|
|
} |
|
139
|
|
|
} |
|
140
|
|
|
|
php-ai /
php-ml
