|
1
|
|
|
<?php |
|
2
|
|
|
|
|
3
|
|
|
declare(strict_types=1); |
|
4
|
|
|
|
|
5
|
|
|
namespace Phpml\Helper; |
|
6
|
|
|
|
|
7
|
|
|
trait OneVsRest |
|
8
|
|
|
{ |
|
9
|
|
|
|
|
10
|
|
|
/** |
|
11
|
|
|
* @var array |
|
12
|
|
|
*/ |
|
13
|
|
|
protected $classifiers = []; |
|
14
|
|
|
|
|
15
|
|
|
/** |
|
16
|
|
|
* All provided training targets' labels. |
|
17
|
|
|
* |
|
18
|
|
|
* @var array |
|
19
|
|
|
*/ |
|
20
|
|
|
protected $allLabels = []; |
|
21
|
|
|
|
|
22
|
|
|
/** |
|
23
|
|
|
* @var array |
|
24
|
|
|
*/ |
|
25
|
|
|
protected $costValues = []; |
|
26
|
|
|
|
|
27
|
|
|
/** |
|
28
|
|
|
* Train a binary classifier in the OvR style |
|
29
|
|
|
* |
|
30
|
|
|
* @param array $samples |
|
31
|
|
|
* @param array $targets |
|
32
|
|
|
*/ |
|
33
|
|
|
public function train(array $samples, array $targets) |
|
34
|
|
|
{ |
|
35
|
|
|
return $this->trainBylabel($samples, $targets); |
|
36
|
|
|
} |
|
37
|
|
|
|
|
38
|
|
|
/** |
|
39
|
|
|
* @param array $samples |
|
40
|
|
|
* @param array $targets |
|
41
|
|
|
* @param array $allLabels All training set labels |
|
42
|
|
|
* @return void |
|
43
|
|
|
*/ |
|
44
|
|
|
protected function trainByLabel(array $samples, array $targets, array $allLabels = array()) |
|
45
|
|
|
{ |
|
46
|
|
|
|
|
47
|
|
|
// Overwrites the current value if it exist. $allLabels must be provided for each partialTrain run. |
|
48
|
|
|
if (!empty($allLabels)) { |
|
49
|
|
|
$this->allLabels = $allLabels; |
|
50
|
|
|
} else { |
|
51
|
|
|
$this->allLabels = array_keys(array_count_values($targets)); |
|
52
|
|
|
} |
|
53
|
|
|
sort($this->allLabels, SORT_STRING); |
|
54
|
|
|
|
|
55
|
|
|
// If there are only two targets, then there is no need to perform OvR |
|
56
|
|
|
if (count($this->allLabels) == 2) { |
|
57
|
|
|
|
|
58
|
|
|
// Init classifier if required. |
|
59
|
|
|
if (empty($this->classifiers)) { |
|
60
|
|
|
$this->classifiers[0] = $this->getClassifierCopy(); |
|
61
|
|
|
} |
|
62
|
|
|
|
|
63
|
|
|
$this->classifiers[0]->trainBinary($samples, $targets, $this->allLabels); |
|
64
|
|
|
} else { |
|
65
|
|
|
// Train a separate classifier for each label and memorize them |
|
66
|
|
|
|
|
67
|
|
|
foreach ($this->allLabels as $label) { |
|
68
|
|
|
|
|
69
|
|
|
// Init classifier if required. |
|
70
|
|
|
if (empty($this->classifiers[$label])) { |
|
71
|
|
|
$this->classifiers[$label] = $this->getClassifierCopy(); |
|
72
|
|
|
} |
|
73
|
|
|
|
|
74
|
|
|
list($binarizedTargets, $classifierLabels) = $this->binarizeTargets($targets, $label); |
|
75
|
|
|
$this->classifiers[$label]->trainBinary($samples, $binarizedTargets, $classifierLabels); |
|
76
|
|
|
} |
|
77
|
|
|
} |
|
78
|
|
|
|
|
79
|
|
|
// If the underlying classifier is capable of giving the cost values |
|
80
|
|
|
// during the training, then assign it to the relevant variable |
|
81
|
|
|
if (method_exists($this->classifiers[0], 'getCostValues')) { |
|
82
|
|
|
$this->costValues = $this->classifiers[0]->getCostValues(); |
|
83
|
|
|
} |
|
84
|
|
|
|
|
85
|
|
|
} |
|
86
|
|
|
|
|
87
|
|
|
/** |
|
88
|
|
|
* Resets the internal vars used by OneVsRest instances. |
|
89
|
|
|
*/ |
|
90
|
|
|
public function reset() { |
|
91
|
|
|
$this->classifiers = []; |
|
92
|
|
|
$this->allLabels = []; |
|
93
|
|
|
$this->costValues = []; |
|
94
|
|
|
} |
|
95
|
|
|
|
|
96
|
|
|
/** |
|
97
|
|
|
* Returns an instance of the current class after cleaning up OneVsRest stuff. |
|
98
|
|
|
* |
|
99
|
|
|
* @return \Phpml\Estimator |
|
100
|
|
|
*/ |
|
101
|
|
|
protected function getClassifierCopy() { |
|
102
|
|
|
|
|
103
|
|
|
// Clone the current classifier, so that |
|
104
|
|
|
// we don't mess up its variables while training |
|
105
|
|
|
// multiple instances of this classifier |
|
106
|
|
|
$classifier = clone $this; |
|
107
|
|
|
$classifier->reset(); |
|
108
|
|
|
return $classifier; |
|
109
|
|
|
} |
|
110
|
|
|
|
|
111
|
|
|
/** |
|
112
|
|
|
* Groups all targets into two groups: Targets equal to |
|
113
|
|
|
* the given label and the others |
|
114
|
|
|
* |
|
115
|
|
|
* $targets is not passed by reference nor contains objects so this method |
|
116
|
|
|
* changes will not affect the caller $targets array. |
|
117
|
|
|
* |
|
118
|
|
|
* @param array $targets |
|
119
|
|
|
* @param mixed $label |
|
120
|
|
|
* @return array Binarized targets and target's labels |
|
121
|
|
|
*/ |
|
122
|
|
|
private function binarizeTargets($targets, $label) |
|
123
|
|
|
{ |
|
124
|
|
|
|
|
125
|
|
|
$notLabel = "not_$label"; |
|
126
|
|
|
foreach ($targets as $key => $target) { |
|
127
|
|
|
$targets[$key] = $target == $label ? $label : $notLabel; |
|
128
|
|
|
} |
|
129
|
|
|
|
|
130
|
|
|
$labels = array($label, $notLabel); |
|
131
|
|
|
return array($targets, $labels); |
|
132
|
|
|
} |
|
133
|
|
|
|
|
134
|
|
|
|
|
135
|
|
|
/** |
|
136
|
|
|
* @param array $sample |
|
137
|
|
|
* |
|
138
|
|
|
* @return mixed |
|
139
|
|
|
*/ |
|
140
|
|
|
protected function predictSample(array $sample) |
|
141
|
|
|
{ |
|
142
|
|
|
if (count($this->allLabels) == 2) { |
|
143
|
|
|
return $this->classifiers[0]->predictSampleBinary($sample); |
|
144
|
|
|
} |
|
145
|
|
|
|
|
146
|
|
|
$probs = []; |
|
147
|
|
|
|
|
148
|
|
|
foreach ($this->classifiers as $label => $predictor) { |
|
149
|
|
|
$probs[$label] = $predictor->predictProbability($sample, $label); |
|
150
|
|
|
} |
|
151
|
|
|
|
|
152
|
|
|
arsort($probs, SORT_NUMERIC); |
|
153
|
|
|
return key($probs); |
|
154
|
|
|
} |
|
155
|
|
|
|
|
156
|
|
|
/** |
|
157
|
|
|
* Each classifier should implement this method instead of train(samples, targets) |
|
158
|
|
|
* |
|
159
|
|
|
* @param array $samples |
|
160
|
|
|
* @param array $targets |
|
161
|
|
|
* @param array $labels |
|
162
|
|
|
*/ |
|
163
|
|
|
abstract protected function trainBinary(array $samples, array $targets, array $labels); |
|
164
|
|
|
|
|
165
|
|
|
/** |
|
166
|
|
|
* Each classifier that make use of OvR approach should be able to |
|
167
|
|
|
* return a probability for a sample to belong to the given label. |
|
168
|
|
|
* |
|
169
|
|
|
* @param array $sample |
|
170
|
|
|
* |
|
171
|
|
|
* @return mixed |
|
172
|
|
|
*/ |
|
173
|
|
|
abstract protected function predictProbability(array $sample, string $label); |
|
174
|
|
|
|
|
175
|
|
|
/** |
|
176
|
|
|
* Each classifier should implement this method instead of predictSample() |
|
177
|
|
|
* |
|
178
|
|
|
* @param array $sample |
|
179
|
|
|
* |
|
180
|
|
|
* @return mixed |
|
181
|
|
|
*/ |
|
182
|
|
|
abstract protected function predictSampleBinary(array $sample); |
|
183
|
|
|
} |
|
184
|
|
|
|
php-ai /
php-ml
