Support probability estimation in SVC (#218)

* Add test for svm model with probability estimation * Extract buildPredictCommand method * Fix test to use PHP_EOL * Add predictProbability method (not completed) * Add test for DataTransformer::predictions * Fix SVM to use PHP_EOL * Support probability estimation in SVM * Add documentation * Add InvalidOperationException class * Throw InvalidOperationException before executing libsvm if probability estimation is not supported
DuaneMurray · Feb 6, 2018 · ec091b5 · ec091b5
1 parent ed775fb
commit ec091b5
Show file tree

Hide file tree

Showing 6 changed files with 268 additions and 11 deletions.
diff --git a/docs/machine-learning/classification/svc.md b/docs/machine-learning/classification/svc.md
@@ -47,3 +47,42 @@ $classifier->predict([3, 2]);
 $classifier->predict([[3, 2], [1, 5]]);
 // return ['b', 'a']
 ```
+
+### Probability estimation
+
+To predict probabilities you must build a classifier with `$probabilityEstimates` set to true. Example:
+
+```
+use Phpml\Classification\SVC;
+use Phpml\SupportVectorMachine\Kernel;
+
+$samples = [[1, 3], [1, 4], [2, 4], [3, 1], [4, 1], [4, 2]];
+$labels = ['a', 'a', 'a', 'b', 'b', 'b'];
+
+$classifier = new SVC(
+    Kernel::LINEAR, // $kernel
+    1.0,            // $cost
+    3,              // $degree
+    null,           // $gamma
+    0.0,            // $coef0
+    0.001,          // $tolerance
+    100,            // $cacheSize
+    true,           // $shrinking
+    true            // $probabilityEstimates, set to true
+);
+
+$classifier->train($samples, $labels);
+```
+
+Then use `predictProbability` method instead of `predict`:
+
+```
+$classifier->predictProbability([3, 2]);
+// return ['a' => 0.349833, 'b' => 0.650167]
+
+$classifier->predictProbability([[3, 2], [1, 5]]);
+// return [
+//   ['a' => 0.349833, 'b' => 0.650167],
+//   ['a' => 0.922664, 'b' => 0.0773364],
+// ]
+```
diff --git a/src/Phpml/Exception/InvalidOperationException.php b/src/Phpml/Exception/InvalidOperationException.php
@@ -0,0 +1,11 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Exception;
+
+use Exception;
+
+class InvalidOperationException extends Exception
+{
+}
diff --git a/src/Phpml/SupportVectorMachine/DataTransformer.php b/src/Phpml/SupportVectorMachine/DataTransformer.php
@@ -49,6 +49,37 @@ public static function predictions(string $rawPredictions, array $labels): array
         return $results;
     }
 
+    public static function probabilities(string $rawPredictions, array $labels): array
+    {
+        $numericLabels = self::numericLabels($labels);
+
+        $predictions = explode(PHP_EOL, trim($rawPredictions));
+
+        $header = array_shift($predictions);
+        $headerColumns = explode(' ', $header);
+        array_shift($headerColumns);
+
+        $columnLabels = [];
+        foreach ($headerColumns as $numericLabel) {
+            $columnLabels[] = array_search($numericLabel, $numericLabels);
+        }
+
+        $results = [];
+        foreach ($predictions as $rawResult) {
+            $probabilities = explode(' ', $rawResult);
+            array_shift($probabilities);
+
+            $result = [];
+            foreach ($probabilities as $i => $prob) {
+                $result[$columnLabels[$i]] = (float) $prob;
+            }
+
+            $results[] = $result;
+        }
+
+        return $results;
+    }
+
     public static function numericLabels(array $labels): array
     {
         $numericLabels = [];

diff --git a/src/Phpml/SupportVectorMachine/SupportVectorMachine.php b/src/Phpml/SupportVectorMachine/SupportVectorMachine.php
@@ -5,6 +5,7 @@
 namespace Phpml\SupportVectorMachine;
 
 use Phpml\Exception\InvalidArgumentException;
+use Phpml\Exception\InvalidOperationException;
 use Phpml\Exception\LibsvmCommandException;
 use Phpml\Helper\Trainable;
 
@@ -178,13 +179,61 @@ public function getModel(): string
      * @throws LibsvmCommandException
      */
     public function predict(array $samples)
+    {
+        $predictions = $this->runSvmPredict($samples, false);
+
+        if (in_array($this->type, [Type::C_SVC, Type::NU_SVC])) {
+            $predictions = DataTransformer::predictions($predictions, $this->targets);
+        } else {
+            $predictions = explode(PHP_EOL, trim($predictions));
+        }
+
+        if (!is_array($samples[0])) {
+            return $predictions[0];
+        }
+
+        return $predictions;
+    }
+
+    /**
+     * @return array|string
+     *
+     * @throws LibsvmCommandException
+     */
+    public function predictProbability(array $samples)
+    {
+        if (!$this->probabilityEstimates) {
+            throw new InvalidOperationException('Model does not support probabiliy estimates');
+        }
+
+        $predictions = $this->runSvmPredict($samples, true);
+
+        if (in_array($this->type, [Type::C_SVC, Type::NU_SVC])) {
+            $predictions = DataTransformer::probabilities($predictions, $this->targets);
+        } else {
+            $predictions = explode(PHP_EOL, trim($predictions));
+        }
+
+        if (!is_array($samples[0])) {
+            return $predictions[0];
+        }
+
+        return $predictions;
+    }
+
+    private function runSvmPredict(array $samples, bool $probabilityEstimates): string
     {
         $testSet = DataTransformer::testSet($samples);
         file_put_contents($testSetFileName = $this->varPath.uniqid('phpml', true), $testSet);
         file_put_contents($modelFileName = $testSetFileName.'-model', $this->model);
         $outputFileName = $testSetFileName.'-output';
 
-        $command = sprintf('%ssvm-predict%s %s %s %s', $this->binPath, $this->getOSExtension(), $testSetFileName, $modelFileName, $outputFileName);
+        $command = $this->buildPredictCommand(
+            $testSetFileName,
+            $modelFileName,
+            $outputFileName,
+            $probabilityEstimates
+        );
         $output = [];
         exec(escapeshellcmd($command).' 2>&1', $output, $return);
 
@@ -198,16 +247,6 @@ public function predict(array $samples)
             throw LibsvmCommandException::failedToRun($command, array_pop($output));
         }
 
-        if (in_array($this->type, [Type::C_SVC, Type::NU_SVC])) {
-            $predictions = DataTransformer::predictions($predictions, $this->targets);
-        } else {
-            $predictions = explode(PHP_EOL, trim($predictions));
-        }
-
-        if (!is_array($samples[0])) {
-            return $predictions[0];
-        }
-
         return $predictions;
     }
 
@@ -246,6 +285,23 @@ private function buildTrainCommand(string $trainingSetFileName, string $modelFil
         );
     }
 
+    private function buildPredictCommand(
+        string $testSetFileName,
+        string $modelFileName,
+        string $outputFileName,
+        bool $probabilityEstimates
+    ): string {
+        return sprintf(
+            '%ssvm-predict%s -b %d %s %s %s',
+            $this->binPath,
+            $this->getOSExtension(),
+            $probabilityEstimates ? 1 : 0,
+            escapeshellarg($testSetFileName),
+            escapeshellarg($modelFileName),
+            escapeshellarg($outputFileName)
+        );
+    }
+
     private function ensureDirectorySeparator(string &$path): void
     {
         if (substr($path, -1) !== DIRECTORY_SEPARATOR) {

diff --git a/tests/Phpml/SupportVectorMachine/DataTransformerTest.php b/tests/Phpml/SupportVectorMachine/DataTransformerTest.php
@@ -37,4 +37,45 @@ public function testTransformSamplesToTestSet(): void
 
         $this->assertEquals($testSet, DataTransformer::testSet($samples));
     }
+
+    public function testPredictions(): void
+    {
+        $labels = ['a', 'a', 'b', 'b'];
+        $rawPredictions = implode(PHP_EOL, [0, 1, 0, 0]);
+
+        $predictions = ['a', 'b', 'a', 'a'];
+
+        $this->assertEquals($predictions, DataTransformer::predictions($rawPredictions, $labels));
+    }
+
+    public function testProbabilities(): void
+    {
+        $labels = ['a', 'b', 'c'];
+        $rawPredictions = implode(PHP_EOL, [
+            'labels 0 1 2',
+            '1 0.1 0.7 0.2',
+            '2 0.2 0.3 0.5',
+            '0 0.6 0.1 0.3',
+        ]);
+
+        $probabilities = [
+            [
+                'a' => 0.1,
+                'b' => 0.7,
+                'c' => 0.2,
+            ],
+            [
+                'a' => 0.2,
+                'b' => 0.3,
+                'c' => 0.5,
+            ],
+            [
+                'a' => 0.6,
+                'b' => 0.1,
+                'c' => 0.3,
+            ],
+        ];
+
+        $this->assertEquals($probabilities, DataTransformer::probabilities($rawPredictions, $labels));
+    }
 }
diff --git a/tests/Phpml/SupportVectorMachine/SupportVectorMachineTest.php b/tests/Phpml/SupportVectorMachine/SupportVectorMachineTest.php
@@ -5,6 +5,7 @@
 namespace Phpml\Tests\SupportVectorMachine;
 
 use Phpml\Exception\InvalidArgumentException;
+use Phpml\Exception\InvalidOperationException;
 use Phpml\Exception\LibsvmCommandException;
 use Phpml\SupportVectorMachine\Kernel;
 use Phpml\SupportVectorMachine\SupportVectorMachine;
@@ -37,6 +38,31 @@ public function testTrainCSVCModelWithLinearKernel(): void
         $this->assertEquals($model, $svm->getModel());
     }
 
+    public function testTrainCSVCModelWithProbabilityEstimate(): void
+    {
+        $samples = [[1, 3], [1, 4], [2, 4], [3, 1], [4, 1], [4, 2]];
+        $labels = ['a', 'a', 'a', 'b', 'b', 'b'];
+
+        $svm = new SupportVectorMachine(
+            Type::C_SVC,
+            Kernel::LINEAR,
+            100.0,
+            0.5,
+            3,
+            null,
+            0.0,
+            0.1,
+            0.01,
+            100,
+            true,
+            true
+        );
+        $svm->train($samples, $labels);
+
+        $this->assertContains(PHP_EOL.'probA ', $svm->getModel());
+        $this->assertContains(PHP_EOL.'probB ', $svm->getModel());
+    }
+
     public function testPredictSampleWithLinearKernel(): void
     {
         $samples = [[1, 3], [1, 4], [2, 4], [3, 1], [4, 1], [4, 2]];
@@ -83,6 +109,41 @@ public function testPredictSampleFromMultipleClassWithRbfKernel(): void
         $this->assertEquals('c', $predictions[2]);
     }
 
+    public function testPredictProbability(): void
+    {
+        $samples = [[1, 3], [1, 4], [2, 4], [3, 1], [4, 1], [4, 2]];
+        $labels = ['a', 'a', 'a', 'b', 'b', 'b'];
+
+        $svm = new SupportVectorMachine(
+            Type::C_SVC,
+            Kernel::LINEAR,
+            100.0,
+            0.5,
+            3,
+            null,
+            0.0,
+            0.1,
+            0.01,
+            100,
+            true,
+            true
+        );
+        $svm->train($samples, $labels);
+
+        $predictions = $svm->predictProbability([
+            [3, 2],
+            [2, 3],
+            [4, -5],
+        ]);
+
+        $this->assertTrue($predictions[0]['a'] < $predictions[0]['b']);
+        $this->assertTrue($predictions[1]['a'] > $predictions[1]['b']);
+        $this->assertTrue($predictions[2]['a'] < $predictions[2]['b']);
+
+        // Should be true because the latter is farther from the decision boundary
+        $this->assertTrue($predictions[0]['b'] < $predictions[2]['b']);
+    }
+
     public function testThrowExceptionWhenVarPathIsNotWritable(): void
     {
         $this->expectException(InvalidArgumentException::class);
@@ -124,4 +185,22 @@ public function testThrowExceptionWhenLibsvmFailsDuringPredict(): void
         $svm = new SupportVectorMachine(Type::C_SVC, Kernel::RBF);
         $svm->predict([1]);
     }
+
+    public function testThrowExceptionWhenPredictProbabilityCalledWithoutProperModel(): void
+    {
+        $this->expectException(InvalidOperationException::class);
+        $this->expectExceptionMessage('Model does not support probabiliy estimates');
+
+        $samples = [[1, 3], [1, 4], [2, 4], [3, 1], [4, 1], [4, 2]];
+        $labels = ['a', 'a', 'a', 'b', 'b', 'b'];
+
+        $svm = new SupportVectorMachine(Type::C_SVC, Kernel::LINEAR, 100.0);
+        $svm->train($samples, $labels);
+
+        $predictions = $svm->predictProbability([
+            [3, 2],
+            [2, 3],
+            [4, -5],
+        ]);
+    }
 }