1
|
|
|
<?php |
2
|
|
|
|
3
|
|
|
declare(strict_types=1); |
4
|
|
|
|
5
|
|
|
namespace Phpml\Metric; |
6
|
|
|
|
7
|
|
|
use Phpml\Exception\InvalidArgumentException; |
8
|
|
|
|
9
|
|
|
class ClassificationReport |
10
|
|
|
{ |
11
|
|
|
public const MICRO_AVERAGE = 1; |
12
|
|
|
|
13
|
|
|
public const MACRO_AVERAGE = 2; |
14
|
|
|
|
15
|
|
|
public const WEIGHTED_AVERAGE = 3; |
16
|
|
|
|
17
|
|
|
/** |
18
|
|
|
* @var array |
19
|
|
|
*/ |
20
|
|
|
private $truePositive = []; |
21
|
|
|
|
22
|
|
|
/** |
23
|
|
|
* @var array |
24
|
|
|
*/ |
25
|
|
|
private $falsePositive = []; |
26
|
|
|
|
27
|
|
|
/** |
28
|
|
|
* @var array |
29
|
|
|
*/ |
30
|
|
|
private $falseNegative = []; |
31
|
|
|
|
32
|
|
|
/** |
33
|
|
|
* @var array |
34
|
|
|
*/ |
35
|
|
|
private $support = []; |
36
|
|
|
|
37
|
|
|
/** |
38
|
|
|
* @var array |
39
|
|
|
*/ |
40
|
|
|
private $precision = []; |
41
|
|
|
|
42
|
|
|
/** |
43
|
|
|
* @var array |
44
|
|
|
*/ |
45
|
|
|
private $recall = []; |
46
|
|
|
|
47
|
|
|
/** |
48
|
|
|
* @var array |
49
|
|
|
*/ |
50
|
|
|
private $f1score = []; |
51
|
|
|
|
52
|
|
|
/** |
53
|
|
|
* @var array |
54
|
|
|
*/ |
55
|
|
|
private $average = []; |
56
|
|
|
|
57
|
|
|
public function __construct(array $actualLabels, array $predictedLabels, int $average = self::MACRO_AVERAGE) |
58
|
|
|
{ |
59
|
|
|
$averagingMethods = range(self::MICRO_AVERAGE, self::WEIGHTED_AVERAGE); |
60
|
|
|
if (!in_array($average, $averagingMethods, true)) { |
61
|
|
|
throw new InvalidArgumentException('Averaging method must be MICRO_AVERAGE, MACRO_AVERAGE or WEIGHTED_AVERAGE'); |
62
|
|
|
} |
63
|
|
|
|
64
|
|
|
$this->aggregateClassificationResults($actualLabels, $predictedLabels); |
65
|
|
|
$this->computeMetrics(); |
66
|
|
|
$this->computeAverage($average); |
67
|
|
|
} |
68
|
|
|
|
69
|
|
|
public function getPrecision(): array |
70
|
|
|
{ |
71
|
|
|
return $this->precision; |
72
|
|
|
} |
73
|
|
|
|
74
|
|
|
public function getRecall(): array |
75
|
|
|
{ |
76
|
|
|
return $this->recall; |
77
|
|
|
} |
78
|
|
|
|
79
|
|
|
public function getF1score(): array |
80
|
|
|
{ |
81
|
|
|
return $this->f1score; |
82
|
|
|
} |
83
|
|
|
|
84
|
|
|
public function getSupport(): array |
85
|
|
|
{ |
86
|
|
|
return $this->support; |
87
|
|
|
} |
88
|
|
|
|
89
|
|
|
public function getAverage(): array |
90
|
|
|
{ |
91
|
|
|
return $this->average; |
92
|
|
|
} |
93
|
|
|
|
94
|
|
|
private function aggregateClassificationResults(array $actualLabels, array $predictedLabels): void |
95
|
|
|
{ |
96
|
|
|
$truePositive = $falsePositive = $falseNegative = $support = self::getLabelIndexedArray($actualLabels, $predictedLabels); |
97
|
|
|
|
98
|
|
|
foreach ($actualLabels as $index => $actual) { |
99
|
|
|
$predicted = $predictedLabels[$index]; |
100
|
|
|
++$support[$actual]; |
101
|
|
|
|
102
|
|
|
if ($actual === $predicted) { |
103
|
|
|
++$truePositive[$actual]; |
104
|
|
|
} else { |
105
|
|
|
++$falsePositive[$predicted]; |
106
|
|
|
++$falseNegative[$actual]; |
107
|
|
|
} |
108
|
|
|
} |
109
|
|
|
|
110
|
|
|
$this->truePositive = $truePositive; |
111
|
|
|
$this->falsePositive = $falsePositive; |
112
|
|
|
$this->falseNegative = $falseNegative; |
113
|
|
|
$this->support = $support; |
114
|
|
|
} |
115
|
|
|
|
116
|
|
|
private function computeMetrics(): void |
117
|
|
|
{ |
118
|
|
|
foreach ($this->truePositive as $label => $tp) { |
119
|
|
|
$this->precision[$label] = $this->computePrecision($tp, $this->falsePositive[$label]); |
120
|
|
|
$this->recall[$label] = $this->computeRecall($tp, $this->falseNegative[$label]); |
121
|
|
|
$this->f1score[$label] = $this->computeF1Score((float) $this->precision[$label], (float) $this->recall[$label]); |
122
|
|
|
} |
123
|
|
|
} |
124
|
|
|
|
125
|
|
|
private function computeAverage(int $average): void |
126
|
|
|
{ |
127
|
|
|
switch ($average) { |
128
|
|
|
case self::MICRO_AVERAGE: |
129
|
|
|
$this->computeMicroAverage(); |
130
|
|
|
|
131
|
|
|
return; |
132
|
|
|
case self::MACRO_AVERAGE: |
133
|
|
|
$this->computeMacroAverage(); |
134
|
|
|
|
135
|
|
|
return; |
136
|
|
|
case self::WEIGHTED_AVERAGE: |
137
|
|
|
$this->computeWeightedAverage(); |
138
|
|
|
|
139
|
|
|
return; |
140
|
|
|
} |
141
|
|
|
} |
142
|
|
|
|
143
|
|
|
private function computeMicroAverage(): void |
144
|
|
|
{ |
145
|
|
|
$truePositive = (int) array_sum($this->truePositive); |
146
|
|
|
$falsePositive = (int) array_sum($this->falsePositive); |
147
|
|
|
$falseNegative = (int) array_sum($this->falseNegative); |
148
|
|
|
|
149
|
|
|
$precision = $this->computePrecision($truePositive, $falsePositive); |
150
|
|
|
$recall = $this->computeRecall($truePositive, $falseNegative); |
151
|
|
|
$f1score = $this->computeF1Score($precision, $recall); |
152
|
|
|
|
153
|
|
|
$this->average = compact('precision', 'recall', 'f1score'); |
154
|
|
|
} |
155
|
|
|
|
156
|
|
|
private function computeMacroAverage(): void |
157
|
|
|
{ |
158
|
|
|
foreach (['precision', 'recall', 'f1score'] as $metric) { |
159
|
|
|
$values = $this->{$metric}; |
160
|
|
|
if (count($values) == 0) { |
161
|
|
|
$this->average[$metric] = 0.0; |
162
|
|
|
|
163
|
|
|
continue; |
164
|
|
|
} |
165
|
|
|
|
166
|
|
|
$this->average[$metric] = array_sum($values) / count($values); |
167
|
|
|
} |
168
|
|
|
} |
169
|
|
|
|
170
|
|
|
private function computeWeightedAverage(): void |
171
|
|
|
{ |
172
|
|
|
foreach (['precision', 'recall', 'f1score'] as $metric) { |
173
|
|
|
$values = $this->{$metric}; |
174
|
|
|
if (count($values) == 0) { |
175
|
|
|
$this->average[$metric] = 0.0; |
176
|
|
|
|
177
|
|
|
continue; |
178
|
|
|
} |
179
|
|
|
|
180
|
|
|
$sum = 0; |
181
|
|
|
foreach ($values as $i => $value) { |
182
|
|
|
$sum += $value * $this->support[$i]; |
183
|
|
|
} |
184
|
|
|
|
185
|
|
|
$this->average[$metric] = $sum / array_sum($this->support); |
186
|
|
|
} |
187
|
|
|
} |
188
|
|
|
|
189
|
|
|
private function computePrecision(int $truePositive, int $falsePositive): float |
190
|
|
|
{ |
191
|
|
|
$divider = $truePositive + $falsePositive; |
192
|
|
|
if ($divider == 0) { |
193
|
|
|
return 0.0; |
194
|
|
|
} |
195
|
|
|
|
196
|
|
|
return $truePositive / $divider; |
197
|
|
|
} |
198
|
|
|
|
199
|
|
|
private function computeRecall(int $truePositive, int $falseNegative): float |
200
|
|
|
{ |
201
|
|
|
$divider = $truePositive + $falseNegative; |
202
|
|
|
if ($divider == 0) { |
203
|
|
|
return 0.0; |
204
|
|
|
} |
205
|
|
|
|
206
|
|
|
return $truePositive / $divider; |
207
|
|
|
} |
208
|
|
|
|
209
|
|
|
private function computeF1Score(float $precision, float $recall): float |
210
|
|
|
{ |
211
|
|
|
$divider = $precision + $recall; |
212
|
|
|
if ($divider == 0) { |
213
|
|
|
return 0.0; |
214
|
|
|
} |
215
|
|
|
|
216
|
|
|
return 2.0 * (($precision * $recall) / $divider); |
217
|
|
|
} |
218
|
|
|
|
219
|
|
|
private static function getLabelIndexedArray(array $actualLabels, array $predictedLabels): array |
220
|
|
|
{ |
221
|
|
|
$labels = array_values(array_unique(array_merge($actualLabels, $predictedLabels))); |
222
|
|
|
sort($labels); |
223
|
|
|
|
224
|
|
|
return (array) array_combine($labels, array_fill(0, count($labels), 0)); |
225
|
|
|
} |
226
|
|
|
} |
227
|
|
|
|