Implement SelectKBest algo for feature selection

src/FeatureExtraction/TfIdfTransformer.php

@@ -20,7 +20,7 @@ class TfIdfTransformer implements Transformer
         }
     }
 
-    public function fit(array $samples): void
+    public function fit(array $samples, ?array $targets = null): void
     {
         $this->countTokensFrequency($samples);
 

src/FeatureExtraction/TokenCountVectorizer.php

@@ -41,7 +41,7 @@ class TokenCountVectorizer implements Transformer
         $this->minDF = $minDF;
     }
 
-    public function fit(array $samples): void
+    public function fit(array $samples, ?array $targets = null): void
     {
         $this->buildVocabulary($samples);
     }

src/FeatureSelection/ScoringFunction.php

@@ -0,0 +1,10 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\FeatureSelection;
+
+interface ScoringFunction
+{
+    public function score(array $samples, array $targets): array;
+}

src/FeatureSelection/ScoringFunction/ANOVAFValue.php

@@ -0,0 +1,21 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\FeatureSelection\ScoringFunction;
+
+use Phpml\FeatureSelection\ScoringFunction;
+use Phpml\Math\Statistic\ANOVA;
+
+final class ANOVAFValue implements ScoringFunction
+{
+    public function score(array $samples, array $targets): array
+    {
+        $grouped = [];
+        foreach ($samples as $index => $sample) {
+            $grouped[$targets[$index]][] = $sample;
+        }
+
+        return ANOVA::oneWayF(array_values($grouped));
+    }
+}

src/FeatureSelection/SelectKBest.php

@@ -0,0 +1,78 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\FeatureSelection;
+
+use Phpml\Exception\InvalidArgumentException;
+use Phpml\Exception\InvalidOperationException;
+use Phpml\FeatureSelection\ScoringFunction\ANOVAFValue;
+use Phpml\Transformer;
+
+final class SelectKBest implements Transformer
+{
+    /**
+     * @var ScoringFunction
+     */
+    private $scoringFunction;
+
+    /**
+     * @var int
+     */
+    private $k;
+
+    /**
+     * @var array|null
+     */
+    private $scores = null;
+
+    /**
+     * @var array|null
+     */
+    private $keepColumns = null;
+
+    public function __construct(?ScoringFunction $scoringFunction = null, int $k = 10)
+    {
+        if ($scoringFunction === null) {
+            $scoringFunction = new ANOVAFValue();
+        }
+
+        $this->scoringFunction = $scoringFunction;
+        $this->k = $k;
+    }
+
+    public function fit(array $samples, ?array $targets = null): void
+    {
+        if ($targets === null || empty($targets)) {
+            throw InvalidArgumentException::arrayCantBeEmpty();
+        }
+
+        $this->scores = $sorted = $this->scoringFunction->score($samples, $targets);
+        if ($this->k >= count($sorted)) {
+            return;
+        }
+
+        arsort($sorted);
+        $this->keepColumns = array_slice($sorted, 0, $this->k, true);
+    }
+
+    public function transform(array &$samples): void
+    {
+        if ($this->keepColumns === null) {
+            return;
+        }
+
+        foreach ($samples as &$sample) {
+            $sample = array_values(array_intersect_key($sample, $this->keepColumns));
+        }
+    }
+
+    public function scores(): array
+    {
+        if ($this->scores === null) {
+            throw new InvalidOperationException('SelectKBest require to fit first to get scores');
+        }
+
+        return $this->scores;
+    }
+}

src/FeatureSelection/VarianceThreshold.php

@@ -33,11 +33,9 @@ final class VarianceThreshold implements Transformer
         }
 
         $this->threshold = $threshold;
-        $this->variances = [];
-        $this->keepColumns = [];
     }
 
-    public function fit(array $samples): void
+    public function fit(array $samples, ?array $targets = null): void
     {
         $this->variances = array_map(function (array $column) {
             return Variance::population($column);

src/Math/Statistic/ANOVA.php

@@ -0,0 +1,137 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Math\Statistic;
+
+use Phpml\Exception\InvalidArgumentException;
+
+/**
+ * Analysis of variance
+ * https://en.wikipedia.org/wiki/Analysis_of_variance
+ */
+final class ANOVA
+{
+    /**
+     * The one-way ANOVA tests the null hypothesis that 2 or more groups have
+     * the same population mean. The test is applied to samples from two or
+     * more groups, possibly with differing sizes.
+     *
+     * @param array|array[] $samples - each row is class samples
+     *
+     * @return array|float[]
+     */
+    public static function oneWayF(array $samples): array
+    {
+        $classes = count($samples);
+        if ($classes < 2) {
+            throw InvalidArgumentException::arraySizeToSmall(2);
+        }
+
+        $samplesPerClass = array_map(function (array $class): int {
+            return count($class);
+        }, $samples);
+        $allSamples = array_sum($samplesPerClass);
+        $ssAllSamples = self::sumOfSquaresPerFeature($samples);
+        $sumSamples = self::sumOfFeaturesPerClass($samples);
+        $squareSumSamples = self::sumOfSquares($sumSamples);
+        $sumSamplesSquare = self::squaresSum($sumSamples);
+        $ssbn = self::calculateSsbn($samples, $sumSamplesSquare, $samplesPerClass, $squareSumSamples, $allSamples);
+        $sswn = self::calculateSswn($ssbn, $ssAllSamples, $squareSumSamples, $allSamples);
+        $dfbn = $classes - 1;
+        $dfwn = $allSamples - $classes;
+
+        $msb = array_map(function ($s) use ($dfbn) {
+            return $s / $dfbn;
+        }, $ssbn);
+        $msw = array_map(function ($s) use ($dfwn) {
+            return $s / $dfwn;
+        }, $sswn);
+
+        $f = [];
+        foreach ($msb as $index => $msbValue) {
+            $f[$index] = $msbValue / $msw[$index];
+        }
+
+        return $f;
+    }
+
+    private static function sumOfSquaresPerFeature(array $samples): array
+    {
+        $sum = array_fill(0, count($samples[0][0]), 0);
+        foreach ($samples as $class) {
+            foreach ($class as $sample) {
+                foreach ($sample as $index => $feature) {
+                    $sum[$index] += $feature ** 2;
+                }
+            }
+        }
+
+        return $sum;
+    }
+
+    private static function sumOfFeaturesPerClass(array $samples): array
+    {
+        return array_map(function (array $class) {
+            $sum = array_fill(0, count($class[0]), 0);
+            foreach ($class as $sample) {
+                foreach ($sample as $index => $feature) {
+                    $sum[$index] += $feature;
+                }
+            }
+
+            return $sum;
+        }, $samples);
+    }
+
+    private static function sumOfSquares(array $sums): array
+    {
+        $squares = array_fill(0, count($sums[0]), 0);
+        foreach ($sums as $row) {
+            foreach ($row as $index => $sum) {
+                $squares[$index] += $sum;
+            }
+        }
+
+        return array_map(function ($sum) {
+            return $sum ** 2;
+        }, $squares);
+    }
+
+    private static function squaresSum(array $sums): array
+    {
+        foreach ($sums as &$row) {
+            foreach ($row as &$sum) {
+                $sum = $sum ** 2;
+            }
+        }
+
+        return $sums;
+    }
+
+    private static function calculateSsbn(array $samples, array $sumSamplesSquare, array $samplesPerClass, array $squareSumSamples, int $allSamples): array
+    {
+        $ssbn = array_fill(0, count($samples[0][0]), 0);
+        foreach ($sumSamplesSquare as $classIndex => $class) {
+            foreach ($class as $index => $feature) {
+                $ssbn[$index] += $feature / $samplesPerClass[$classIndex];
+            }
+        }
+
+        foreach ($squareSumSamples as $index => $sum) {
+            $ssbn[$index] -= $sum / $allSamples;
+        }
+
+        return $ssbn;
+    }
+
+    private static function calculateSswn(array $ssbn, array $ssAllSamples, array $squareSumSamples, int $allSamples): array
+    {
+        $sswn = [];
+        foreach ($ssAllSamples as $index => $ss) {
+            $sswn[$index] = ($ss - $squareSumSamples[$index] / $allSamples) - $ssbn[$index];
+        }
+
+        return $sswn;
+    }
+}

src/Preprocessing/Imputer.php

@@ -43,7 +43,7 @@ class Imputer implements Preprocessor
         $this->samples = $samples;
     }
 
-    public function fit(array $samples): void
+    public function fit(array $samples, ?array $targets = null): void
     {
         $this->samples = $samples;
     }

src/Preprocessing/Normalizer.php

@@ -48,7 +48,7 @@ class Normalizer implements Preprocessor
         $this->norm = $norm;
     }
 
-    public function fit(array $samples): void
+    public function fit(array $samples, ?array $targets = null): void
     {
         if ($this->fitted) {
             return;

src/Transformer.php

@@ -7,12 +7,9 @@ namespace Phpml;
 interface Transformer
 {
     /**
-     * @param array $samples
+     * most transformers don't require targets to train so null allow to use fit method without setting targets
      */
-    public function fit(array $samples);
+    public function fit(array $samples, ?array $targets = null): void;
 
-    /**
-     * @param array $samples
-     */
-    public function transform(array &$samples);
+    public function transform(array &$samples): void;
 }

tests/Classification/MLPClassifierTest.php

@@ -59,7 +59,7 @@ class MLPClassifierTest extends TestCase
     public function testBackpropagationLearning(): void
     {
         // Single layer 2 classes.
-        $network = new MLPClassifier(2, [2], ['a', 'b']);
+        $network = new MLPClassifier(2, [2], ['a', 'b'], 1000);
         $network->train(
             [[1, 0], [0, 1], [1, 1], [0, 0]],
             ['a', 'b', 'a', 'b']
@@ -118,7 +118,7 @@ class MLPClassifierTest extends TestCase
     public function testBackpropagationLearningMultilayer(): void
     {
         // Multi-layer 2 classes.
-        $network = new MLPClassifier(5, [3, 2], ['a', 'b', 'c']);
+        $network = new MLPClassifier(5, [3, 2], ['a', 'b', 'c'], 2000);
         $network->train(
             [[1, 0, 0, 0, 0], [0, 1, 1, 0, 0], [1, 1, 1, 1, 1], [0, 0, 0, 0, 0]],
             ['a', 'b', 'a', 'c']
@@ -133,7 +133,7 @@ class MLPClassifierTest extends TestCase
     public function testBackpropagationLearningMulticlass(): void
     {
         // Multi-layer more than 2 classes.
-        $network = new MLPClassifier(5, [3, 2], ['a', 'b', 4]);
+        $network = new MLPClassifier(5, [3, 2], ['a', 'b', 4], 1000);
         $network->train(
             [[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 1, 0], [1, 1, 1, 1, 1], [0, 0, 0, 0, 0]],
             ['a', 'b', 'a', 'a', 4]
@@ -151,7 +151,7 @@ class MLPClassifierTest extends TestCase
      */
     public function testBackpropagationActivationFunctions(ActivationFunction $activationFunction): void
     {
-        $network = new MLPClassifier(5, [3], ['a', 'b'], 10000, $activationFunction);
+        $network = new MLPClassifier(5, [3], ['a', 'b'], 1000, $activationFunction);
         $network->train(
             [[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 1, 0], [1, 1, 1, 1, 1]],
             ['a', 'b', 'a', 'a']
@@ -178,7 +178,7 @@ class MLPClassifierTest extends TestCase
         // Instantinate new Percetron trained for OR problem
         $samples = [[0, 0], [1, 0], [0, 1], [1, 1]];
         $targets = [0, 1, 1, 1];
-        $classifier = new MLPClassifier(2, [2], [0, 1]);
+        $classifier = new MLPClassifier(2, [2], [0, 1], 1000);
         $classifier->train($samples, $targets);
         $testSamples = [[0, 0], [1, 0], [0, 1], [1, 1]];
         $predicted = $classifier->predict($testSamples);

tests/FeatureSelection/ScoringFunction/ANOVAFValueTest.php

@@ -0,0 +1,25 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Tests\FeatureSelection\ScoringFunction;
+
+use Phpml\Dataset\Demo\IrisDataset;
+use Phpml\FeatureSelection\ScoringFunction\ANOVAFValue;
+use PHPUnit\Framework\TestCase;
+
+final class ANOVAFValueTest extends TestCase
+{
+    public function testScoreForANOVAFValue(): void
+    {
+        $dataset = new IrisDataset();
+        $function = new ANOVAFValue();
+
+        self::assertEquals(
+            [119.2645, 47.3644, 1179.0343, 959.3244],
+            $function->score($dataset->getSamples(), $dataset->getTargets()),
+            '',
+            0.0001
+        );
+    }
+}

tests/FeatureSelection/SelectKBestTest.php

@@ -0,0 +1,61 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Tests\FeatureSelection;
+
+use Phpml\Dataset\Demo\IrisDataset;
+use Phpml\Exception\InvalidArgumentException;
+use Phpml\Exception\InvalidOperationException;
+use Phpml\FeatureSelection\ScoringFunction\ANOVAFValue;
+use Phpml\FeatureSelection\SelectKBest;
+use PHPUnit\Framework\TestCase;
+
+final class SelectKBestTest extends TestCase
+{
+    public function testSelectKBestWithDefaultScoringFunction(): void
+    {
+        $samples = [[1, 2, 1], [1, 3, 4], [5, 2, 1], [1, 3, 3], [1, 3, 4], [0, 3, 5]];
+        $targets = ['a', 'a', 'a', 'b', 'b', 'b'];
+        $selector = new SelectKBest(null, 2);
+        $selector->fit($samples, $targets);
+        $selector->transform($samples);
+
+        self::assertEquals([[2, 1], [3, 4], [2, 1], [3, 3], [3, 4], [3, 5]], $samples);
+    }
+
+    public function testSelectKBestWithKBiggerThanFeatures(): void
+    {
+        $samples = [[1, 2, 1], [1, 3, 4], [5, 2, 1], [1, 3, 3], [1, 3, 4], [0, 3, 5]];
+        $targets = ['a', 'a', 'a', 'b', 'b', 'b'];
+        $selector = new SelectKBest(null, 4);
+        $selector->fit($samples, $targets);
+        $selector->transform($samples);
+
+        self::assertEquals([[1, 2, 1], [1, 3, 4], [5, 2, 1], [1, 3, 3], [1, 3, 4], [0, 3, 5]], $samples);
+    }
+
+    public function testSelectKBestWithIrisDataset(): void
+    {
+        $dataset = new IrisDataset();
+        $selector = new SelectKBest(new ANOVAFValue(), 2);
+        $selector->fit($samples = $dataset->getSamples(), $dataset->getTargets());
+        $selector->transform($samples);
+
+        self::assertEquals(2, count($samples[0]));
+    }
+
+    public function testThrowExceptionOnEmptyTargets(): void
+    {
+        $this->expectException(InvalidArgumentException::class);
+        $selector = new SelectKBest(new ANOVAFValue(), 2);
+        $selector->fit([[1, 2, 3], [4, 5, 6]], []);
+    }
+
+    public function testThrowExceptionWhenNotTrained(): void
+    {
+        $this->expectException(InvalidOperationException::class);
+        $selector = new SelectKBest(new ANOVAFValue(), 2);
+        $selector->scores();
+    }
+}

tests/Math/Statistic/ANOVATest.php

@@ -0,0 +1,44 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Tests\Math\Statistic;
+
+use Phpml\Exception\InvalidArgumentException;
+use Phpml\Math\Statistic\ANOVA;
+use PHPUnit\Framework\TestCase;
+
+final class ANOVATest extends TestCase
+{
+    public function testOneWayF(): void
+    {
+        $samples = [
+            [[1, 2, 1], [1, 3, 4], [5, 2, 1]],
+            [[1, 3, 3], [1, 3, 4], [0, 3, 5]],
+        ];
+
+        $f = [1.47058824, 4.0, 3.0];
+
+        self::assertEquals($f, ANOVA::oneWayF($samples), '', 0.00000001);
+    }
+
+    public function testOneWayFWithDifferingSizes(): void
+    {
+        $samples = [
+            [[1, 2, 1], [1, 3, 4], [5, 2, 1]],
+            [[1, 3, 3], [1, 3, 4]],
+        ];
+
+        self::assertEquals([0.6, 2.4, 1.24615385], ANOVA::oneWayF($samples), '', 0.00000001);
+    }
+
+    public function testThrowExceptionOnToSmallSamples(): void
+    {
+        $this->expectException(InvalidArgumentException::class);
+        $samples = [
+            [[1, 2, 1], [1, 3, 4], [5, 2, 1]],
+        ];
+
+        ANOVA::oneWayF($samples);
+    }
+}