1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\Helper\Optimizer; |
6
|
|
|
|
7
|
|
|
use Closure; |
8
|
|
|
|
9
|
|
|
/** |
10
|
|
|
* Batch version of Gradient Descent to optimize the weights |
11
|
|
|
* of a classifier given samples, targets and the objective function to minimize |
12
|
|
|
*/ |
13
|
|
|
class GD extends StochasticGD |
14
|
|
|
{ |
15
|
|
|
/** |
16
|
|
|
* Number of samples given |
17
|
|
|
* |
18
|
|
|
* @var int|null |
19
|
|
|
*/ |
20
|
|
|
protected $sampleCount; |
21
|
|
|
|
22
|
|
|
public function runOptimization(array $samples, array $targets, Closure $gradientCb): array |
23
|
|
|
{ |
24
|
|
|
$this->samples = $samples; |
25
|
|
|
$this->targets = $targets; |
26
|
|
|
$this->gradientCb = $gradientCb; |
27
|
|
|
$this->sampleCount = count($this->samples); |
28
|
|
|
|
29
|
|
|
// Batch learning is executed: |
30
|
|
|
$currIter = 0; |
31
|
|
|
$this->costValues = []; |
32
|
|
|
while ($this->maxIterations > $currIter++) { |
33
|
|
|
$theta = $this->theta; |
34
|
|
|
|
35
|
|
|
// Calculate update terms for each sample |
36
|
|
|
[$errors, $updates, $totalPenalty] = $this->gradient($theta); |
|
|
|
|
37
|
|
|
|
38
|
|
|
$this->updateWeightsWithUpdates($updates, $totalPenalty); |
39
|
|
|
|
40
|
|
|
$this->costValues[] = array_sum($errors) / $this->sampleCount; |
41
|
|
|
|
42
|
|
|
if ($this->earlyStop($theta)) { |
43
|
|
|
break; |
44
|
|
|
} |
45
|
|
|
} |
46
|
|
|
|
47
|
|
|
$this->clear(); |
48
|
|
|
|
49
|
|
|
return $this->theta; |
50
|
|
|
} |
51
|
|
|
|
52
|
|
|
/** |
53
|
|
|
* Calculates gradient, cost function and penalty term for each sample |
54
|
|
|
* then returns them as an array of values |
55
|
|
|
*/ |
56
|
|
|
protected function gradient(array $theta): array |
57
|
|
|
{ |
58
|
|
|
$costs = []; |
59
|
|
|
$gradient = []; |
60
|
|
|
$totalPenalty = 0; |
61
|
|
|
|
62
|
|
|
foreach ($this->samples as $index => $sample) { |
63
|
|
|
$target = $this->targets[$index]; |
64
|
|
|
|
65
|
|
|
$result = ($this->gradientCb)($theta, $sample, $target); |
66
|
|
|
[$cost, $grad, $penalty] = array_pad($result, 3, 0); |
|
|
|
|
67
|
|
|
|
68
|
|
|
$costs[] = $cost; |
69
|
|
|
$gradient[] = $grad; |
70
|
|
|
$totalPenalty += $penalty; |
|
|
|
|
71
|
|
|
} |
72
|
|
|
|
73
|
|
|
$totalPenalty /= $this->sampleCount; |
74
|
|
|
|
75
|
|
|
return [$costs, $gradient, $totalPenalty]; |
76
|
|
|
} |
77
|
|
|
|
78
|
|
|
protected function updateWeightsWithUpdates(array $updates, float $penalty): void |
79
|
|
|
{ |
80
|
|
|
// Updates all weights at once |
81
|
|
|
for ($i = 0; $i <= $this->dimensions; ++$i) { |
82
|
|
|
if ($i === 0) { |
83
|
|
|
$this->theta[0] -= $this->learningRate * array_sum($updates); |
84
|
|
|
} else { |
85
|
|
|
$col = array_column($this->samples, $i - 1); |
86
|
|
|
|
87
|
|
|
$error = 0; |
88
|
|
|
foreach ($col as $index => $val) { |
89
|
|
|
$error += $val * $updates[$index]; |
90
|
|
|
} |
91
|
|
|
|
92
|
|
|
$this->theta[$i] -= $this->learningRate * |
93
|
|
|
($error + $penalty * $this->theta[$i]); |
94
|
|
|
} |
95
|
|
|
} |
96
|
|
|
} |
97
|
|
|
|
98
|
|
|
/** |
99
|
|
|
* Clears the optimizer internal vars after the optimization process. |
100
|
|
|
*/ |
101
|
|
|
protected function clear(): void |
102
|
|
|
{ |
103
|
|
|
$this->sampleCount = null; |
104
|
|
|
parent::clear(); |
105
|
|
|
} |
106
|
|
|
} |
107
|
|
|
|
This check marks access to variables or properties that have not been declared yet. While PHP has no explicit notion of declaring a variable, accessing it before a value is assigned to it is most likely a bug.