1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\Regression; |
6
|
|
|
|
7
|
|
|
use Phpml\Exception\InvalidOperationException; |
8
|
|
|
use Phpml\Math\Statistic\Mean; |
9
|
|
|
use Phpml\Math\Statistic\Variance; |
10
|
|
|
use Phpml\Tree\CART; |
11
|
|
|
use Phpml\Tree\Node\AverageNode; |
12
|
|
|
use Phpml\Tree\Node\BinaryNode; |
13
|
|
|
use Phpml\Tree\Node\DecisionNode; |
14
|
|
|
|
15
|
|
|
final class DecisionTreeRegressor extends CART implements Regression |
16
|
|
|
{ |
17
|
|
|
/** |
18
|
|
|
* @var int|null |
19
|
|
|
*/ |
20
|
|
|
protected $maxFeatures; |
21
|
|
|
|
22
|
|
|
/** |
23
|
|
|
* @var float |
24
|
|
|
*/ |
25
|
|
|
protected $tolerance; |
26
|
|
|
|
27
|
|
|
/** |
28
|
|
|
* @var array |
29
|
|
|
*/ |
30
|
|
|
protected $columns = []; |
31
|
|
|
|
32
|
|
|
public function train(array $samples, array $targets): void |
33
|
|
|
{ |
34
|
|
|
$features = count($samples[0]); |
35
|
|
|
|
36
|
|
|
$this->columns = range(0, $features - 1); |
37
|
|
|
$this->maxFeatures = $this->maxFeatures ?? (int) round(sqrt($features)); |
38
|
|
|
|
39
|
|
|
$this->grow($samples, $targets); |
40
|
|
|
|
41
|
|
|
$this->columns = []; |
42
|
|
|
} |
43
|
|
|
|
44
|
|
|
public function predict(array $samples) |
45
|
|
|
{ |
46
|
|
|
if ($this->bare()) { |
47
|
|
|
throw new InvalidOperationException('Regressor must be trained first'); |
48
|
|
|
} |
49
|
|
|
|
50
|
|
|
$predictions = []; |
51
|
|
|
|
52
|
|
|
foreach ($samples as $sample) { |
53
|
|
|
$node = $this->search($sample); |
|
|
|
|
54
|
|
|
|
55
|
|
|
$predictions[] = $node instanceof AverageNode |
56
|
|
|
? $node->outcome() |
57
|
|
|
: null; |
58
|
|
|
} |
59
|
|
|
|
60
|
|
|
return $predictions; |
61
|
|
|
} |
62
|
|
|
|
63
|
|
|
protected function split(array $samples, array $targets): DecisionNode |
64
|
|
|
{ |
65
|
|
|
$bestVariance = INF; |
66
|
|
|
$bestColumn = $bestValue = null; |
67
|
|
|
$bestGroups = []; |
68
|
|
|
|
69
|
|
|
shuffle($this->columns); |
70
|
|
|
|
71
|
|
|
foreach (array_slice($this->columns, 0, $this->maxFeatures) as $column) { |
72
|
|
|
$values = array_unique(array_column($samples, $column)); |
73
|
|
|
|
74
|
|
|
foreach ($values as $value) { |
75
|
|
|
$groups = $this->partition($column, $value, $samples, $targets); |
76
|
|
|
|
77
|
|
|
$variance = $this->splitImpurity($groups); |
78
|
|
|
|
79
|
|
|
if ($variance < $bestVariance) { |
80
|
|
|
$bestColumn = $column; |
81
|
|
|
$bestValue = $value; |
82
|
|
|
$bestGroups = $groups; |
83
|
|
|
$bestVariance = $variance; |
84
|
|
|
} |
85
|
|
|
|
86
|
|
|
if ($variance <= $this->tolerance) { |
87
|
|
|
break 2; |
88
|
|
|
} |
89
|
|
|
} |
90
|
|
|
} |
91
|
|
|
|
92
|
|
|
return new DecisionNode($bestColumn, $bestValue, $bestGroups, $bestVariance); |
93
|
|
|
} |
94
|
|
|
|
95
|
|
|
protected function terminate(array $targets): BinaryNode |
96
|
|
|
{ |
97
|
|
|
return new AverageNode(Mean::arithmetic($targets), Variance::population($targets), count($targets)); |
98
|
|
|
} |
99
|
|
|
|
100
|
|
|
protected function splitImpurity(array $groups): float |
101
|
|
|
{ |
102
|
|
|
$samplesCount = (int) array_sum(array_map(static function (array $group) { |
103
|
|
|
return count($group[0]); |
104
|
|
|
}, $groups)); |
105
|
|
|
|
106
|
|
|
$impurity = 0.; |
107
|
|
|
|
108
|
|
|
foreach ($groups as $group) { |
109
|
|
|
$k = count($group[1]); |
110
|
|
|
|
111
|
|
|
if ($k < 2) { |
112
|
|
|
continue 1; |
113
|
|
|
} |
114
|
|
|
|
115
|
|
|
$variance = Variance::population($group[1]); |
116
|
|
|
|
117
|
|
|
$impurity += ($k / $samplesCount) * $variance; |
118
|
|
|
} |
119
|
|
|
|
120
|
|
|
return $impurity; |
121
|
|
|
} |
122
|
|
|
|
123
|
|
|
/** |
124
|
|
|
* @param int|float $value |
125
|
|
|
*/ |
126
|
|
|
private function partition(int $column, $value, array $samples, array $targets): array |
127
|
|
|
{ |
128
|
|
|
$leftSamples = $leftTargets = $rightSamples = $rightTargets = []; |
129
|
|
|
foreach ($samples as $index => $sample) { |
130
|
|
|
if ($sample[$column] < $value) { |
131
|
|
|
$leftSamples[] = $sample; |
132
|
|
|
$leftTargets[] = $targets[$index]; |
133
|
|
|
} else { |
134
|
|
|
$rightSamples[] = $sample; |
135
|
|
|
$rightTargets[] = $targets[$index]; |
136
|
|
|
} |
137
|
|
|
} |
138
|
|
|
|
139
|
|
|
return [ |
140
|
|
|
[$leftSamples, $leftTargets], |
141
|
|
|
[$rightSamples, $rightTargets], |
142
|
|
|
]; |
143
|
|
|
} |
144
|
|
|
} |
145
|
|
|
|
This check looks for function or method calls that always return null and whose return value is assigned to a variable.
The method
getObject()
can return nothing but null, so it makes no sense to assign that value to a variable.The reason is most likely that a function or method is imcomplete or has been reduced for debug purposes.