|
1
|
|
|
<?php |
|
2
|
|
|
|
|
3
|
|
|
declare(strict_types=1); |
|
4
|
|
|
|
|
5
|
|
|
namespace Phpml\Classification\Ensemble; |
|
6
|
|
|
|
|
7
|
|
|
use Phpml\Classification\Classifier; |
|
8
|
|
|
use Phpml\Classification\DecisionTree; |
|
9
|
|
|
use Phpml\Exception\InvalidArgumentException; |
|
10
|
|
|
use Phpml\Helper\Predictable; |
|
11
|
|
|
use Phpml\Helper\Trainable; |
|
12
|
|
|
use ReflectionClass; |
|
13
|
|
|
|
|
14
|
|
|
class Bagging implements Classifier |
|
15
|
|
|
{ |
|
16
|
|
|
use Trainable; |
|
17
|
|
|
use Predictable; |
|
18
|
|
|
|
|
19
|
|
|
/** |
|
20
|
|
|
* @var int |
|
21
|
|
|
*/ |
|
22
|
|
|
protected $numSamples; |
|
23
|
|
|
|
|
24
|
|
|
/** |
|
25
|
|
|
* @var int |
|
26
|
|
|
*/ |
|
27
|
|
|
protected $featureCount = 0; |
|
28
|
|
|
|
|
29
|
|
|
/** |
|
30
|
|
|
* @var int |
|
31
|
|
|
*/ |
|
32
|
|
|
protected $numClassifier; |
|
33
|
|
|
|
|
34
|
|
|
/** |
|
35
|
|
|
* @var string |
|
36
|
|
|
*/ |
|
37
|
|
|
protected $classifier = DecisionTree::class; |
|
38
|
|
|
|
|
39
|
|
|
/** |
|
40
|
|
|
* @var array |
|
41
|
|
|
*/ |
|
42
|
|
|
protected $classifierOptions = ['depth' => 20]; |
|
43
|
|
|
|
|
44
|
|
|
/** |
|
45
|
|
|
* @var array |
|
46
|
|
|
*/ |
|
47
|
|
|
protected $classifiers = []; |
|
48
|
|
|
|
|
49
|
|
|
/** |
|
50
|
|
|
* @var float |
|
51
|
|
|
*/ |
|
52
|
|
|
protected $subsetRatio = 0.7; |
|
53
|
|
|
|
|
54
|
|
|
/** |
|
55
|
|
|
* Creates an ensemble classifier with given number of base classifiers |
|
56
|
|
|
* Default number of base classifiers is 50. |
|
57
|
|
|
* The more number of base classifiers, the better performance but at the cost of procesing time |
|
58
|
|
|
*/ |
|
59
|
|
|
public function __construct(int $numClassifier = 50) |
|
60
|
|
|
{ |
|
61
|
|
|
$this->numClassifier = $numClassifier; |
|
62
|
|
|
} |
|
63
|
|
|
|
|
64
|
|
|
/** |
|
65
|
|
|
* This method determines the ratio of samples used to create the 'bootstrap' subset, |
|
66
|
|
|
* e.g., random samples drawn from the original dataset with replacement (allow repeats), |
|
67
|
|
|
* to train each base classifier. |
|
68
|
|
|
* |
|
69
|
|
|
* @return $this |
|
70
|
|
|
* |
|
71
|
|
|
* @throws InvalidArgumentException |
|
72
|
|
|
*/ |
|
73
|
|
|
public function setSubsetRatio(float $ratio) |
|
74
|
|
|
{ |
|
75
|
|
|
if ($ratio < 0.1 || $ratio > 1.0) { |
|
76
|
|
|
throw new InvalidArgumentException('Subset ratio should be between 0.1 and 1.0'); |
|
77
|
|
|
} |
|
78
|
|
|
|
|
79
|
|
|
$this->subsetRatio = $ratio; |
|
80
|
|
|
|
|
81
|
|
|
return $this; |
|
82
|
|
|
} |
|
83
|
|
|
|
|
84
|
|
|
/** |
|
85
|
|
|
* This method is used to set the base classifier. Default value is |
|
86
|
|
|
* DecisionTree::class, but any class that implements the <i>Classifier</i> |
|
87
|
|
|
* can be used. <br> |
|
88
|
|
|
* While giving the parameters of the classifier, the values should be |
|
89
|
|
|
* given in the order they are in the constructor of the classifier and parameter |
|
90
|
|
|
* names are neglected. |
|
91
|
|
|
* |
|
92
|
|
|
* @return $this |
|
93
|
|
|
*/ |
|
94
|
|
|
public function setClassifer(string $classifier, array $classifierOptions = []) |
|
95
|
|
|
{ |
|
96
|
|
|
$this->classifier = $classifier; |
|
97
|
|
|
$this->classifierOptions = $classifierOptions; |
|
98
|
|
|
|
|
99
|
|
|
return $this; |
|
100
|
|
|
} |
|
101
|
|
|
|
|
102
|
|
|
public function train(array $samples, array $targets): void |
|
103
|
|
|
{ |
|
104
|
|
|
$this->samples = array_merge($this->samples, $samples); |
|
105
|
|
|
$this->targets = array_merge($this->targets, $targets); |
|
106
|
|
|
$this->featureCount = count($samples[0]); |
|
107
|
|
|
$this->numSamples = count($this->samples); |
|
108
|
|
|
|
|
109
|
|
|
// Init classifiers and train them with bootstrap samples |
|
110
|
|
|
$this->classifiers = $this->initClassifiers(); |
|
111
|
|
|
$index = 0; |
|
112
|
|
|
foreach ($this->classifiers as $classifier) { |
|
113
|
|
|
[$samples, $targets] = $this->getRandomSubset($index); |
|
114
|
|
|
$classifier->train($samples, $targets); |
|
115
|
|
|
++$index; |
|
116
|
|
|
} |
|
117
|
|
|
} |
|
118
|
|
|
|
|
119
|
|
|
protected function getRandomSubset(int $index): array |
|
120
|
|
|
{ |
|
121
|
|
|
$samples = []; |
|
122
|
|
|
$targets = []; |
|
123
|
|
|
srand($index); |
|
124
|
|
|
$bootstrapSize = $this->subsetRatio * $this->numSamples; |
|
125
|
|
|
for ($i = 0; $i < $bootstrapSize; ++$i) { |
|
126
|
|
|
$rand = random_int(0, $this->numSamples - 1); |
|
127
|
|
|
$samples[] = $this->samples[$rand]; |
|
128
|
|
|
$targets[] = $this->targets[$rand]; |
|
129
|
|
|
} |
|
130
|
|
|
|
|
131
|
|
|
return [$samples, $targets]; |
|
132
|
|
|
} |
|
133
|
|
|
|
|
134
|
|
|
protected function initClassifiers(): array |
|
135
|
|
|
{ |
|
136
|
|
|
$classifiers = []; |
|
137
|
|
|
for ($i = 0; $i < $this->numClassifier; ++$i) { |
|
138
|
|
|
$ref = new ReflectionClass($this->classifier); |
|
139
|
|
|
/** @var Classifier $obj */ |
|
140
|
|
|
$obj = count($this->classifierOptions) === 0 ? $ref->newInstance() : $ref->newInstanceArgs($this->classifierOptions); |
|
141
|
|
|
|
|
142
|
|
|
$classifiers[] = $this->initSingleClassifier($obj); |
|
143
|
|
|
} |
|
144
|
|
|
|
|
145
|
|
|
return $classifiers; |
|
146
|
|
|
} |
|
147
|
|
|
|
|
148
|
|
|
protected function initSingleClassifier(Classifier $classifier): Classifier |
|
149
|
|
|
{ |
|
150
|
|
|
return $classifier; |
|
151
|
|
|
} |
|
152
|
|
|
|
|
153
|
|
|
/** |
|
154
|
|
|
* @return mixed |
|
155
|
|
|
*/ |
|
156
|
|
|
protected function predictSample(array $sample) |
|
157
|
|
|
{ |
|
158
|
|
|
$predictions = []; |
|
159
|
|
|
foreach ($this->classifiers as $classifier) { |
|
160
|
|
|
/** @var Classifier $classifier */ |
|
161
|
|
|
$predictions[] = $classifier->predict($sample); |
|
162
|
|
|
} |
|
163
|
|
|
|
|
164
|
|
|
$counts = array_count_values($predictions); |
|
165
|
|
|
arsort($counts); |
|
166
|
|
|
reset($counts); |
|
167
|
|
|
|
|
168
|
|
|
return key($counts); |
|
169
|
|
|
} |
|
170
|
|
|
} |
|
171
|
|
|
|
php-ai /
php-ml
