/
DecisionTree.php
484 lines (406 loc) · 14.6 KB
/
DecisionTree.php
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
<?php
declare(strict_types=1);
namespace Phpml\Classification;
use Phpml\Classification\DecisionTree\DecisionTreeLeaf;
use Phpml\Exception\InvalidArgumentException;
use Phpml\Helper\Predictable;
use Phpml\Helper\Trainable;
use Phpml\Math\Statistic\Mean;
class DecisionTree implements Classifier
{
use Trainable;
use Predictable;
public const CONTINUOUS = 1;
public const NOMINAL = 2;
/**
* @var int
*/
public $actualDepth = 0;
/**
* @var array
*/
protected $columnTypes = [];
/**
* @var DecisionTreeLeaf
*/
protected $tree;
/**
* @var int
*/
protected $maxDepth;
/**
* @var array
*/
private $labels = [];
/**
* @var int
*/
private $featureCount = 0;
/**
* @var int
*/
private $numUsableFeatures = 0;
/**
* @var array
*/
private $selectedFeatures = [];
/**
* @var array|null
*/
private $featureImportances;
/**
* @var array
*/
private $columnNames = [];
public function __construct(int $maxDepth = 10)
{
$this->maxDepth = $maxDepth;
}
public function train(array $samples, array $targets): void
{
$this->samples = array_merge($this->samples, $samples);
$this->targets = array_merge($this->targets, $targets);
$this->featureCount = count($this->samples[0]);
$this->columnTypes = self::getColumnTypes($this->samples);
$this->labels = array_keys(array_count_values($this->targets));
$this->tree = $this->getSplitLeaf(range(0, count($this->samples) - 1));
// Each time the tree is trained, feature importances are reset so that
// we will have to compute it again depending on the new data
$this->featureImportances = null;
// If column names are given or computed before, then there is no
// need to init it and accidentally remove the previous given names
if ($this->columnNames === []) {
$this->columnNames = range(0, $this->featureCount - 1);
} elseif (count($this->columnNames) > $this->featureCount) {
$this->columnNames = array_slice($this->columnNames, 0, $this->featureCount);
} elseif (count($this->columnNames) < $this->featureCount) {
$this->columnNames = array_merge(
$this->columnNames,
range(count($this->columnNames), $this->featureCount - 1)
);
}
}
public static function getColumnTypes(array $samples): array
{
$types = [];
$featureCount = count($samples[0]);
for ($i = 0; $i < $featureCount; ++$i) {
$values = array_column($samples, $i);
$isCategorical = self::isCategoricalColumn($values);
$types[] = $isCategorical ? self::NOMINAL : self::CONTINUOUS;
}
return $types;
}
/**
* @param mixed $baseValue
*/
public function getGiniIndex($baseValue, array $colValues, array $targets): float
{
$countMatrix = [];
foreach ($this->labels as $label) {
$countMatrix[$label] = [0, 0];
}
foreach ($colValues as $index => $value) {
$label = $targets[$index];
$rowIndex = $value === $baseValue ? 0 : 1;
++$countMatrix[$label][$rowIndex];
}
$giniParts = [0, 0];
for ($i = 0; $i <= 1; ++$i) {
$part = 0;
$sum = array_sum(array_column($countMatrix, $i));
if ($sum > 0) {
foreach ($this->labels as $label) {
$part += ($countMatrix[$label][$i] / (float) $sum) ** 2;
}
}
$giniParts[$i] = (1 - $part) * $sum;
}
return array_sum($giniParts) / count($colValues);
}
/**
* This method is used to set number of columns to be used
* when deciding a split at an internal node of the tree. <br>
* If the value is given 0, then all features are used (default behaviour),
* otherwise the given value will be used as a maximum for number of columns
* randomly selected for each split operation.
*
* @return $this
*
* @throws InvalidArgumentException
*/
public function setNumFeatures(int $numFeatures)
{
if ($numFeatures < 0) {
throw new InvalidArgumentException('Selected column count should be greater or equal to zero');
}
$this->numUsableFeatures = $numFeatures;
return $this;
}
/**
* A string array to represent columns. Useful when HTML output or
* column importances are desired to be inspected.
*
* @return $this
*
* @throws InvalidArgumentException
*/
public function setColumnNames(array $names)
{
if ($this->featureCount !== 0 && count($names) !== $this->featureCount) {
throw new InvalidArgumentException(sprintf('Length of the given array should be equal to feature count %s', $this->featureCount));
}
$this->columnNames = $names;
return $this;
}
public function getHtml(): string
{
return $this->tree->getHTML($this->columnNames);
}
/**
* This will return an array including an importance value for
* each column in the given dataset. The importance values are
* normalized and their total makes 1.<br/>
*/
public function getFeatureImportances(): array
{
if ($this->featureImportances !== null) {
return $this->featureImportances;
}
$sampleCount = count($this->samples);
$this->featureImportances = [];
foreach ($this->columnNames as $column => $columnName) {
$nodes = $this->getSplitNodesByColumn($column, $this->tree);
$importance = 0;
foreach ($nodes as $node) {
$importance += $node->getNodeImpurityDecrease($sampleCount);
}
$this->featureImportances[$columnName] = $importance;
}
// Normalize & sort the importances
$total = array_sum($this->featureImportances);
if ($total > 0) {
array_walk($this->featureImportances, function (&$importance) use ($total): void {
$importance /= $total;
});
arsort($this->featureImportances);
}
return $this->featureImportances;
}
protected function getSplitLeaf(array $records, int $depth = 0): DecisionTreeLeaf
{
$split = $this->getBestSplit($records);
$split->level = $depth;
if ($this->actualDepth < $depth) {
$this->actualDepth = $depth;
}
// Traverse all records to see if all records belong to the same class,
// otherwise group the records so that we can classify the leaf
// in case maximum depth is reached
$leftRecords = [];
$rightRecords = [];
$remainingTargets = [];
$prevRecord = null;
$allSame = true;
foreach ($records as $recordNo) {
// Check if the previous record is the same with the current one
$record = $this->samples[$recordNo];
if ($prevRecord !== null && $prevRecord != $record) {
$allSame = false;
}
$prevRecord = $record;
// According to the split criteron, this record will
// belong to either left or the right side in the next split
if ($split->evaluate($record)) {
$leftRecords[] = $recordNo;
} else {
$rightRecords[] = $recordNo;
}
// Group remaining targets
$target = $this->targets[$recordNo];
if (!array_key_exists($target, $remainingTargets)) {
$remainingTargets[$target] = 1;
} else {
++$remainingTargets[$target];
}
}
if ($allSame || $depth >= $this->maxDepth || count($remainingTargets) === 1) {
$split->isTerminal = true;
arsort($remainingTargets);
$split->classValue = (string) key($remainingTargets);
} else {
if (isset($leftRecords[0])) {
$split->leftLeaf = $this->getSplitLeaf($leftRecords, $depth + 1);
}
if (isset($rightRecords[0])) {
$split->rightLeaf = $this->getSplitLeaf($rightRecords, $depth + 1);
}
}
return $split;
}
protected function getBestSplit(array $records): DecisionTreeLeaf
{
$targets = array_intersect_key($this->targets, array_flip($records));
$samples = (array) array_combine(
$records,
$this->preprocess(array_intersect_key($this->samples, array_flip($records)))
);
$bestGiniVal = 1;
$bestSplit = null;
$features = $this->getSelectedFeatures();
foreach ($features as $i) {
$colValues = [];
foreach ($samples as $index => $row) {
$colValues[$index] = $row[$i];
}
$counts = array_count_values($colValues);
arsort($counts);
$baseValue = key($counts);
if ($baseValue === null) {
continue;
}
$gini = $this->getGiniIndex($baseValue, $colValues, $targets);
if ($bestSplit === null || $bestGiniVal > $gini) {
$split = new DecisionTreeLeaf();
$split->value = $baseValue;
$split->giniIndex = $gini;
$split->columnIndex = $i;
$split->isContinuous = $this->columnTypes[$i] === self::CONTINUOUS;
$split->records = $records;
// If a numeric column is to be selected, then
// the original numeric value and the selected operator
// will also be saved into the leaf for future access
if ($this->columnTypes[$i] === self::CONTINUOUS) {
$matches = [];
preg_match("/^([<>=]{1,2})\s*(.*)/", (string) $split->value, $matches);
$split->operator = $matches[1];
$split->numericValue = (float) $matches[2];
}
$bestSplit = $split;
$bestGiniVal = $gini;
}
}
return $bestSplit;
}
/**
* Returns available features/columns to the tree for the decision making
* process. <br>
*
* If a number is given with setNumFeatures() method, then a random selection
* of features up to this number is returned. <br>
*
* If some features are manually selected by use of setSelectedFeatures(),
* then only these features are returned <br>
*
* If any of above methods were not called beforehand, then all features
* are returned by default.
*/
protected function getSelectedFeatures(): array
{
$allFeatures = range(0, $this->featureCount - 1);
if ($this->numUsableFeatures === 0 && count($this->selectedFeatures) === 0) {
return $allFeatures;
}
if (count($this->selectedFeatures) > 0) {
return $this->selectedFeatures;
}
$numFeatures = $this->numUsableFeatures;
if ($numFeatures > $this->featureCount) {
$numFeatures = $this->featureCount;
}
shuffle($allFeatures);
$selectedFeatures = array_slice($allFeatures, 0, $numFeatures);
sort($selectedFeatures);
return $selectedFeatures;
}
protected function preprocess(array $samples): array
{
// Detect and convert continuous data column values into
// discrete values by using the median as a threshold value
$columns = [];
for ($i = 0; $i < $this->featureCount; ++$i) {
$values = array_column($samples, $i);
if ($this->columnTypes[$i] == self::CONTINUOUS) {
$median = Mean::median($values);
foreach ($values as &$value) {
if ($value <= $median) {
$value = "<= ${median}";
} else {
$value = "> ${median}";
}
}
}
$columns[] = $values;
}
// Below method is a strange yet very simple & efficient method
// to get the transpose of a 2D array
return array_map(null, ...$columns);
}
protected static function isCategoricalColumn(array $columnValues): bool
{
$count = count($columnValues);
// There are two main indicators that *may* show whether a
// column is composed of discrete set of values:
// 1- Column may contain string values and non-float values
// 2- Number of unique values in the column is only a small fraction of
// all values in that column (Lower than or equal to %20 of all values)
$numericValues = array_filter($columnValues, 'is_numeric');
$floatValues = array_filter($columnValues, 'is_float');
if (count($floatValues) > 0) {
return false;
}
if (count($numericValues) !== $count) {
return true;
}
$distinctValues = array_count_values($columnValues);
return count($distinctValues) <= $count / 5;
}
/**
* Used to set predefined features to consider while deciding which column to use for a split
*/
protected function setSelectedFeatures(array $selectedFeatures): void
{
$this->selectedFeatures = $selectedFeatures;
}
/**
* Collects and returns an array of internal nodes that use the given
* column as a split criterion
*/
protected function getSplitNodesByColumn(int $column, DecisionTreeLeaf $node): array
{
if ($node->isTerminal) {
return [];
}
$nodes = [];
if ($node->columnIndex === $column) {
$nodes[] = $node;
}
$lNodes = [];
$rNodes = [];
if ($node->leftLeaf !== null) {
$lNodes = $this->getSplitNodesByColumn($column, $node->leftLeaf);
}
if ($node->rightLeaf !== null) {
$rNodes = $this->getSplitNodesByColumn($column, $node->rightLeaf);
}
return array_merge($nodes, $lNodes, $rNodes);
}
/**
* @return mixed
*/
protected function predictSample(array $sample)
{
$node = $this->tree;
do {
if ($node->isTerminal) {
return $node->classValue;
}
if ($node->evaluate($sample)) {
$node = $node->leftLeaf;
} else {
$node = $node->rightLeaf;
}
} while ($node);
return $this->labels[0];
}
}