|
1
|
|
|
<?php |
|
2
|
|
|
|
|
3
|
|
|
declare(strict_types=1); |
|
4
|
|
|
|
|
5
|
|
|
namespace Phpml\NeuralNetwork\Network; |
|
6
|
|
|
|
|
7
|
|
|
use Phpml\Estimator; |
|
8
|
|
|
use Phpml\Exception\InvalidArgumentException; |
|
9
|
|
|
use Phpml\NeuralNetwork\Training\Backpropagation; |
|
10
|
|
|
use Phpml\NeuralNetwork\ActivationFunction; |
|
11
|
|
|
use Phpml\NeuralNetwork\Layer; |
|
12
|
|
|
use Phpml\NeuralNetwork\Node\Bias; |
|
13
|
|
|
use Phpml\NeuralNetwork\Node\Input; |
|
14
|
|
|
use Phpml\NeuralNetwork\Node\Neuron; |
|
15
|
|
|
use Phpml\NeuralNetwork\Node\Neuron\Synapse; |
|
16
|
|
|
use Phpml\Helper\Predictable; |
|
17
|
|
|
|
|
18
|
|
|
abstract class MultilayerPerceptron extends LayeredNetwork implements Estimator |
|
19
|
|
|
{ |
|
20
|
|
|
use Predictable; |
|
21
|
|
|
|
|
22
|
|
|
/** |
|
23
|
|
|
* @var array |
|
24
|
|
|
*/ |
|
25
|
|
|
protected $classes = []; |
|
26
|
|
|
|
|
27
|
|
|
/** |
|
28
|
|
|
* @var int |
|
29
|
|
|
*/ |
|
30
|
|
|
private $iterations; |
|
31
|
|
|
|
|
32
|
|
|
/** |
|
33
|
|
|
* @var Backpropagation |
|
34
|
|
|
*/ |
|
35
|
|
|
protected $backpropagation = null; |
|
36
|
|
|
|
|
37
|
|
|
/** |
|
38
|
|
|
* @param int $inputLayerFeatures |
|
39
|
|
|
* @param array $hiddenLayers |
|
40
|
|
|
* @param array $classes |
|
41
|
|
|
* @param int $iterations |
|
42
|
|
|
* @param ActivationFunction|null $activationFunction |
|
43
|
|
|
* @param int $theta |
|
44
|
|
|
* |
|
45
|
|
|
* @throws InvalidArgumentException |
|
46
|
|
|
*/ |
|
47
|
|
|
public function __construct(int $inputLayerFeatures, array $hiddenLayers, array $classes, int $iterations = 10000, ActivationFunction $activationFunction = null, int $theta = 1) |
|
48
|
|
|
{ |
|
49
|
|
|
if (empty($hiddenLayers)) { |
|
50
|
|
|
throw InvalidArgumentException::invalidLayersNumber(); |
|
51
|
|
|
} |
|
52
|
|
|
|
|
53
|
|
|
$nClasses = count($classes); |
|
54
|
|
|
if ($nClasses < 2) { |
|
55
|
|
|
throw InvalidArgumentException::invalidClassesNumber(); |
|
56
|
|
|
} |
|
57
|
|
|
$this->classes = array_values($classes); |
|
58
|
|
|
|
|
59
|
|
|
$this->iterations = $iterations; |
|
60
|
|
|
|
|
61
|
|
|
$this->addInputLayer($inputLayerFeatures); |
|
62
|
|
|
$this->addNeuronLayers($hiddenLayers, $activationFunction); |
|
63
|
|
|
$this->addNeuronLayers([$nClasses], $activationFunction); |
|
64
|
|
|
|
|
65
|
|
|
$this->addBiasNodes(); |
|
66
|
|
|
$this->generateSynapses(); |
|
67
|
|
|
|
|
68
|
|
|
$this->backpropagation = new Backpropagation($theta); |
|
69
|
|
|
} |
|
70
|
|
|
|
|
71
|
|
|
/** |
|
72
|
|
|
* @param array $samples |
|
73
|
|
|
* @param array $targets |
|
74
|
|
|
*/ |
|
75
|
|
|
public function train(array $samples, array $targets) |
|
76
|
|
|
{ |
|
77
|
|
|
for ($i = 0; $i < $this->iterations; ++$i) { |
|
78
|
|
|
$this->trainSamples($samples, $targets); |
|
79
|
|
|
} |
|
80
|
|
|
} |
|
81
|
|
|
|
|
82
|
|
|
/** |
|
83
|
|
|
* @param array $sample |
|
84
|
|
|
* @param mixed $target |
|
85
|
|
|
*/ |
|
86
|
|
|
protected abstract function trainSample(array $sample, $target); |
|
|
|
|
|
|
87
|
|
|
|
|
88
|
|
|
/** |
|
89
|
|
|
* @param array $sample |
|
90
|
|
|
* @return mixed |
|
91
|
|
|
*/ |
|
92
|
|
|
protected abstract function predictSample(array $sample); |
|
|
|
|
|
|
93
|
|
|
|
|
94
|
|
|
/** |
|
95
|
|
|
* @param int $nodes |
|
96
|
|
|
*/ |
|
97
|
|
|
private function addInputLayer(int $nodes) |
|
98
|
|
|
{ |
|
99
|
|
|
$this->addLayer(new Layer($nodes, Input::class)); |
|
100
|
|
|
} |
|
101
|
|
|
|
|
102
|
|
|
/** |
|
103
|
|
|
* @param array $layers |
|
104
|
|
|
* @param ActivationFunction|null $activationFunction |
|
105
|
|
|
*/ |
|
106
|
|
|
private function addNeuronLayers(array $layers, ActivationFunction $activationFunction = null) |
|
107
|
|
|
{ |
|
108
|
|
|
foreach ($layers as $neurons) { |
|
109
|
|
|
$this->addLayer(new Layer($neurons, Neuron::class, $activationFunction)); |
|
110
|
|
|
} |
|
111
|
|
|
} |
|
112
|
|
|
|
|
113
|
|
|
private function generateSynapses() |
|
114
|
|
|
{ |
|
115
|
|
|
$layersNumber = count($this->layers) - 1; |
|
116
|
|
|
for ($i = 0; $i < $layersNumber; ++$i) { |
|
117
|
|
|
$currentLayer = $this->layers[$i]; |
|
118
|
|
|
$nextLayer = $this->layers[$i + 1]; |
|
119
|
|
|
$this->generateLayerSynapses($nextLayer, $currentLayer); |
|
120
|
|
|
} |
|
121
|
|
|
} |
|
122
|
|
|
|
|
123
|
|
|
private function addBiasNodes() |
|
124
|
|
|
{ |
|
125
|
|
|
$biasLayers = count($this->layers) - 1; |
|
126
|
|
|
for ($i = 0; $i < $biasLayers; ++$i) { |
|
127
|
|
|
$this->layers[$i]->addNode(new Bias()); |
|
128
|
|
|
} |
|
129
|
|
|
} |
|
130
|
|
|
|
|
131
|
|
|
/** |
|
132
|
|
|
* @param Layer $nextLayer |
|
133
|
|
|
* @param Layer $currentLayer |
|
134
|
|
|
*/ |
|
135
|
|
|
private function generateLayerSynapses(Layer $nextLayer, Layer $currentLayer) |
|
136
|
|
|
{ |
|
137
|
|
|
foreach ($nextLayer->getNodes() as $nextNeuron) { |
|
138
|
|
|
if ($nextNeuron instanceof Neuron) { |
|
139
|
|
|
$this->generateNeuronSynapses($currentLayer, $nextNeuron); |
|
140
|
|
|
} |
|
141
|
|
|
} |
|
142
|
|
|
} |
|
143
|
|
|
|
|
144
|
|
|
/** |
|
145
|
|
|
* @param Layer $currentLayer |
|
146
|
|
|
* @param Neuron $nextNeuron |
|
147
|
|
|
*/ |
|
148
|
|
|
private function generateNeuronSynapses(Layer $currentLayer, Neuron $nextNeuron) |
|
149
|
|
|
{ |
|
150
|
|
|
foreach ($currentLayer->getNodes() as $currentNeuron) { |
|
151
|
|
|
$nextNeuron->addSynapse(new Synapse($currentNeuron)); |
|
152
|
|
|
} |
|
153
|
|
|
} |
|
154
|
|
|
|
|
155
|
|
|
/** |
|
156
|
|
|
* @param array $samples |
|
157
|
|
|
* @param array $targets |
|
158
|
|
|
*/ |
|
159
|
|
|
private function trainSamples(array $samples, array $targets) |
|
160
|
|
|
{ |
|
161
|
|
|
foreach ($targets as $key => $target) { |
|
162
|
|
|
$this->trainSample($samples[$key], $target); |
|
163
|
|
|
} |
|
164
|
|
|
} |
|
165
|
|
|
} |
|
166
|
|
|
|
php-ai /
php-ml
Loading history...