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
|
|
|
|
11
|
|
|
class Perceptron implements Classifier |
12
|
|
|
{ |
13
|
|
|
use Predictable; |
14
|
|
|
|
15
|
|
|
/** |
16
|
|
|
* The function whose result will be used to calculate the network error |
17
|
|
|
* for each instance |
18
|
|
|
* |
19
|
|
|
* @var string |
20
|
|
|
*/ |
21
|
|
|
protected static $errorFunction = 'outputClass'; |
22
|
|
|
|
23
|
|
|
/** |
24
|
|
|
* @var array |
25
|
|
|
*/ |
26
|
|
|
protected $samples = []; |
27
|
|
|
|
28
|
|
|
/** |
29
|
|
|
* @var array |
30
|
|
|
*/ |
31
|
|
|
protected $targets = []; |
32
|
|
|
|
33
|
|
|
/** |
34
|
|
|
* @var array |
35
|
|
|
*/ |
36
|
|
|
protected $labels = []; |
37
|
|
|
|
38
|
|
|
/** |
39
|
|
|
* @var int |
40
|
|
|
*/ |
41
|
|
|
protected $featureCount = 0; |
42
|
|
|
|
43
|
|
|
/** |
44
|
|
|
* @var array |
45
|
|
|
*/ |
46
|
|
|
protected $weights; |
47
|
|
|
|
48
|
|
|
/** |
49
|
|
|
* @var float |
50
|
|
|
*/ |
51
|
|
|
protected $learningRate; |
52
|
|
|
|
53
|
|
|
/** |
54
|
|
|
* @var int |
55
|
|
|
*/ |
56
|
|
|
protected $maxIterations; |
57
|
|
|
|
58
|
|
|
/** |
59
|
|
|
* Initalize a perceptron classifier with given learning rate and maximum |
60
|
|
|
* number of iterations used while training the perceptron <br> |
61
|
|
|
* |
62
|
|
|
* Learning rate should be a float value between 0.0(exclusive) and 1.0(inclusive) <br> |
63
|
|
|
* Maximum number of iterations can be an integer value greater than 0 |
64
|
|
|
* @param int $learningRate |
65
|
|
|
* @param int $maxIterations |
66
|
|
|
*/ |
67
|
|
|
public function __construct(float $learningRate = 0.001, int $maxIterations = 1000) |
68
|
|
|
{ |
69
|
|
|
if ($learningRate <= 0.0 || $learningRate > 1.0) { |
70
|
|
|
throw new \Exception("Learning rate should be a float value between 0.0(exclusive) and 1.0(inclusive)"); |
71
|
|
|
} |
72
|
|
|
|
73
|
|
|
if ($maxIterations <= 0) { |
74
|
|
|
throw new \Exception("Maximum number of iterations should be an integer greater than 0"); |
75
|
|
|
} |
76
|
|
|
|
77
|
|
|
$this->learningRate = $learningRate; |
78
|
|
|
$this->maxIterations = $maxIterations; |
79
|
|
|
} |
80
|
|
|
|
81
|
|
|
/** |
82
|
|
|
* @param array $samples |
83
|
|
|
* @param array $targets |
84
|
|
|
*/ |
85
|
|
|
public function train(array $samples, array $targets) |
86
|
|
|
{ |
87
|
|
|
$this->labels = array_keys(array_count_values($targets)); |
88
|
|
|
if (count($this->labels) > 2) { |
89
|
|
|
throw new \Exception("Perceptron is for only binary (two-class) classification"); |
90
|
|
|
} |
91
|
|
|
|
92
|
|
|
// Set all target values to either -1 or 1 |
93
|
|
|
$this->labels = [1 => $this->labels[0], -1 => $this->labels[1]]; |
94
|
|
|
foreach ($targets as $target) { |
95
|
|
|
$this->targets[] = $target == $this->labels[1] ? 1 : -1; |
96
|
|
|
} |
97
|
|
|
|
98
|
|
|
// Set samples and feature count vars |
99
|
|
|
$this->samples = array_merge($this->samples, $samples); |
100
|
|
|
$this->featureCount = count($this->samples[0]); |
101
|
|
|
|
102
|
|
|
// Init weights with random values |
103
|
|
|
$this->weights = array_fill(0, $this->featureCount + 1, 0); |
104
|
|
|
foreach ($this->weights as &$weight) { |
105
|
|
|
$weight = rand() / (float) getrandmax(); |
106
|
|
|
} |
107
|
|
|
// Do training |
108
|
|
|
$this->runTraining(); |
109
|
|
|
} |
110
|
|
|
|
111
|
|
|
/** |
112
|
|
|
* Adapts the weights with respect to given samples and targets |
113
|
|
|
* by use of perceptron learning rule |
114
|
|
|
*/ |
115
|
|
|
protected function runTraining() |
116
|
|
|
{ |
117
|
|
|
$currIter = 0; |
118
|
|
|
while ($this->maxIterations > $currIter++) { |
119
|
|
|
foreach ($this->samples as $index => $sample) { |
120
|
|
|
$target = $this->targets[$index]; |
121
|
|
|
$prediction = $this->{static::$errorFunction}($sample); |
122
|
|
|
$update = $target - $prediction; |
123
|
|
|
// Update bias |
124
|
|
|
$this->weights[0] += $update * $this->learningRate; // Bias |
125
|
|
|
// Update other weights |
126
|
|
|
for ($i=1; $i <= $this->featureCount; $i++) { |
127
|
|
|
$this->weights[$i] += $update * $sample[$i - 1] * $this->learningRate; |
128
|
|
|
} |
129
|
|
|
} |
130
|
|
|
} |
131
|
|
|
} |
132
|
|
|
|
133
|
|
|
/** |
134
|
|
|
* Calculates net output of the network as a float value for the given input |
135
|
|
|
* |
136
|
|
|
* @param array $sample |
137
|
|
|
* @return int |
138
|
|
|
*/ |
139
|
|
|
protected function output(array $sample) |
140
|
|
|
{ |
141
|
|
|
$sum = 0; |
142
|
|
|
foreach ($this->weights as $index => $w) { |
143
|
|
|
if ($index == 0) { |
144
|
|
|
$sum += $w; |
145
|
|
|
} else { |
146
|
|
|
$sum += $w * $sample[$index - 1]; |
147
|
|
|
} |
148
|
|
|
} |
149
|
|
|
|
150
|
|
|
return $sum; |
151
|
|
|
} |
152
|
|
|
|
153
|
|
|
/** |
154
|
|
|
* Returns the class value (either -1 or 1) for the given input |
155
|
|
|
* |
156
|
|
|
* @param array $sample |
157
|
|
|
* @return int |
158
|
|
|
*/ |
159
|
|
|
protected function outputClass(array $sample) |
160
|
|
|
{ |
161
|
|
|
return $this->output($sample) > 0 ? 1 : -1; |
162
|
|
|
} |
163
|
|
|
|
164
|
|
|
/** |
165
|
|
|
* @param array $sample |
166
|
|
|
* @return mixed |
167
|
|
|
*/ |
168
|
|
|
protected function predictSample(array $sample) |
169
|
|
|
{ |
170
|
|
|
$predictedClass = $this->outputClass($sample); |
171
|
|
|
|
172
|
|
|
return $this->labels[ $predictedClass ]; |
173
|
|
|
} |
174
|
|
|
} |
175
|
|
|
|