1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\Classification\Linear; |
6
|
|
|
|
7
|
|
|
use Phpml\Helper\Predictable; |
8
|
|
|
use Phpml\Helper\Trainable; |
9
|
|
|
use Phpml\Classification\Classifier; |
10
|
|
|
use Phpml\Classification\Linear\Perceptron; |
11
|
|
|
use Phpml\Preprocessing\Normalizer; |
12
|
|
|
|
13
|
|
|
class Adaline extends Perceptron |
14
|
|
|
{ |
15
|
|
|
|
16
|
|
|
/** |
17
|
|
|
* Batch training is the default Adaline training algorithm |
18
|
|
|
*/ |
19
|
|
|
const BATCH_TRAINING = 1; |
20
|
|
|
|
21
|
|
|
/** |
22
|
|
|
* Online training: Stochastic gradient descent learning |
23
|
|
|
*/ |
24
|
|
|
const ONLINE_TRAINING = 2; |
25
|
|
|
|
26
|
|
|
/** |
27
|
|
|
* The function whose result will be used to calculate the network error |
28
|
|
|
* for each instance |
29
|
|
|
* |
30
|
|
|
* @var string |
31
|
|
|
*/ |
32
|
|
|
protected static $errorFunction = 'output'; |
33
|
|
|
|
34
|
|
|
/** |
35
|
|
|
* Training type may be either 'Batch' or 'Online' learning |
36
|
|
|
* |
37
|
|
|
* @var string |
38
|
|
|
*/ |
39
|
|
|
protected $trainingType; |
40
|
|
|
|
41
|
|
|
/** |
42
|
|
|
* @var Normalizer |
43
|
|
|
*/ |
44
|
|
|
private $normalizer; |
45
|
|
|
|
46
|
|
|
/** |
47
|
|
|
* Initalize an Adaline (ADAptive LInear NEuron) classifier with given learning rate and maximum |
48
|
|
|
* number of iterations used while training the classifier <br> |
49
|
|
|
* |
50
|
|
|
* Learning rate should be a float value between 0.0(exclusive) and 1.0 (inclusive) <br> |
51
|
|
|
* Maximum number of iterations can be an integer value greater than 0 <br> |
52
|
|
|
* If normalizeInputs is set to true, then every input given to the algorithm will be standardized |
53
|
|
|
* by use of standard deviation and mean calculation |
54
|
|
|
* |
55
|
|
|
* @param int $learningRate |
56
|
|
|
* @param int $maxIterations |
57
|
|
|
*/ |
58
|
|
|
public function __construct(float $learningRate = 0.001, int $maxIterations = 1000, |
59
|
|
|
bool $normalizeInputs = true, int $trainingType = self::BATCH_TRAINING) |
60
|
|
|
{ |
61
|
|
|
if ($normalizeInputs) { |
62
|
|
|
$this->normalizer = new Normalizer(Normalizer::NORM_STD); |
63
|
|
|
} |
64
|
|
|
|
65
|
|
|
if (! in_array($trainingType, [self::BATCH_TRAINING, self::ONLINE_TRAINING])) { |
66
|
|
|
throw new \Exception("Adaline can only be trained with batch and online/stochastic gradient descent algorithm"); |
67
|
|
|
} |
68
|
|
|
$this->trainingType = $trainingType; |
|
|
|
|
69
|
|
|
|
70
|
|
|
parent::__construct($learningRate, $maxIterations); |
71
|
|
|
} |
72
|
|
|
|
73
|
|
|
/** |
74
|
|
|
* @param array $samples |
75
|
|
|
* @param array $targets |
76
|
|
|
*/ |
77
|
|
|
public function train(array $samples, array $targets) |
78
|
|
|
{ |
79
|
|
|
if ($this->normalizer) { |
80
|
|
|
$this->normalizer->transform($samples); |
81
|
|
|
} |
82
|
|
|
|
83
|
|
|
parent::train($samples, $targets); |
84
|
|
|
} |
85
|
|
|
|
86
|
|
|
/** |
87
|
|
|
* Adapts the weights with respect to given samples and targets |
88
|
|
|
* by use of gradient descent learning rule |
89
|
|
|
*/ |
90
|
|
|
protected function runTraining() |
91
|
|
|
{ |
92
|
|
|
// If online training is chosen, then the parent runTraining method |
93
|
|
|
// will be executed with the 'output' method as the error function |
94
|
|
|
if ($this->trainingType == self::ONLINE_TRAINING) { |
95
|
|
|
return parent::runTraining(); |
96
|
|
|
} |
97
|
|
|
|
98
|
|
|
// Batch learning is executed: |
99
|
|
|
$currIter = 0; |
100
|
|
|
while ($this->maxIterations > $currIter++) { |
101
|
|
|
$outputs = array_map([$this, 'output'], $this->samples); |
102
|
|
|
$updates = array_map([$this, 'gradient'], $this->targets, $outputs); |
103
|
|
|
$sum = array_sum($updates); |
104
|
|
|
|
105
|
|
|
// Updates all weights at once |
106
|
|
|
for ($i=0; $i <= $this->featureCount; $i++) { |
107
|
|
|
if ($i == 0) { |
108
|
|
|
$this->weights[0] += $this->learningRate * $sum; |
109
|
|
|
} else { |
110
|
|
|
$col = array_column($this->samples, $i - 1); |
111
|
|
|
$error = 0; |
112
|
|
|
foreach ($col as $index => $val) { |
113
|
|
|
$error += $val * $updates[$index]; |
114
|
|
|
} |
115
|
|
|
|
116
|
|
|
$this->weights[$i] += $this->learningRate * $error; |
117
|
|
|
} |
118
|
|
|
} |
119
|
|
|
} |
120
|
|
|
} |
121
|
|
|
|
122
|
|
|
/** |
123
|
|
|
* Returns the direction of gradient given the desired and actual outputs |
124
|
|
|
* |
125
|
|
|
* @param int $desired |
126
|
|
|
* @param int $output |
127
|
|
|
* @return int |
128
|
|
|
*/ |
129
|
|
|
protected function gradient($desired, $output) |
130
|
|
|
{ |
131
|
|
|
return $desired - $output; |
132
|
|
|
} |
133
|
|
|
|
134
|
|
|
/** |
135
|
|
|
* @param array $sample |
136
|
|
|
* @return mixed |
137
|
|
|
*/ |
138
|
|
|
public function predictSample(array $sample) |
139
|
|
|
{ |
140
|
|
|
if ($this->normalizer) { |
141
|
|
|
$samples = [$sample]; |
142
|
|
|
$this->normalizer->transform($samples); |
143
|
|
|
$sample = $samples[0]; |
144
|
|
|
} |
145
|
|
|
|
146
|
|
|
return parent::predictSample($sample); |
147
|
|
|
} |
148
|
|
|
} |
149
|
|
|
|
This check looks for assignments to scalar types that may be of the wrong type.
To ensure the code behaves as expected, it may be a good idea to add an explicit type cast.