Fix Optimizer initial theta randomization (#239)

* Fix Optimizer initial theta randomization

* Add more tests for LUDecomposition and FuzzyCMeans

src/Helper/Optimizer/Optimizer.php

@@ -5,10 +5,12 @@ declare(strict_types=1);
 namespace Phpml\Helper\Optimizer;
 
 use Closure;
-use Exception;
+use Phpml\Exception\InvalidArgumentException;
 
 abstract class Optimizer
 {
+    public $initialTheta;
+
     /**
      * Unknown variables to be found
      *
@@ -33,21 +35,16 @@ abstract class Optimizer
         // Inits the weights randomly
         $this->theta = [];
         for ($i = 0; $i < $this->dimensions; ++$i) {
-            $this->theta[] = random_int(0, getrandmax()) / (float) getrandmax();
+            $this->theta[] = (random_int(0, PHP_INT_MAX) / PHP_INT_MAX) + 0.1;
         }
+
+        $this->initialTheta = $this->theta;
     }
 
-    /**
-     * Sets the weights manually
-     *
-     * @return $this
-     *
-     * @throws \Exception
-     */
     public function setInitialTheta(array $theta)
     {
         if (count($theta) != $this->dimensions) {
-            throw new Exception("Number of values in the weights array should be ${this}->dimensions");
+            throw new InvalidArgumentException(sprintf('Number of values in the weights array should be %s', $this->dimensions));
         }
 
         $this->theta = $theta;

src/Helper/Optimizer/StochasticGD.php

@@ -5,6 +5,7 @@ declare(strict_types=1);
 namespace Phpml\Helper\Optimizer;
 
 use Closure;
+use Phpml\Exception\InvalidArgumentException;
 
 /**
  * Stochastic Gradient Descent optimization method
@@ -88,6 +89,17 @@ class StochasticGD extends Optimizer
         $this->dimensions = $dimensions;
     }
 
+    public function setInitialTheta(array $theta)
+    {
+        if (count($theta) != $this->dimensions + 1) {
+            throw new InvalidArgumentException(sprintf('Number of values in the weights array should be %s', $this->dimensions + 1));
+        }
+
+        $this->theta = $theta;
+
+        return $this;
+    }
+
     /**
      * Sets minimum value for the change in the theta values
      * between iterations to continue the iterations.<br>

src/Math/LinearAlgebra/LUDecomposition.php

@@ -225,25 +225,14 @@ class LUDecomposition
         return true;
     }
 
-    /**
-     * Count determinants
-     *
-     * @return float|int d matrix determinant
-     *
-     * @throws MatrixException
-     */
-    public function det()
+    public function det(): float
     {
-        if ($this->m !== $this->n) {
-            throw MatrixException::notSquareMatrix();
-        }
-
         $d = $this->pivsign;
         for ($j = 0; $j < $this->n; ++$j) {
             $d *= $this->LU[$j][$j];
         }
 
-        return $d;
+        return (float) $d;
     }
 
     /**

tests/Clustering/FuzzyCMeansTest.php

@@ -5,6 +5,7 @@ declare(strict_types=1);
 namespace Phpml\Tests\Clustering;
 
 use Phpml\Clustering\FuzzyCMeans;
+use Phpml\Exception\InvalidArgumentException;
 use PHPUnit\Framework\TestCase;
 
 class FuzzyCMeansTest extends TestCase
@@ -45,4 +46,19 @@ class FuzzyCMeansTest extends TestCase
             $this->assertEquals(1, array_sum($col));
         }
     }
+
+    /**
+     * @dataProvider invalidClusterNumberProvider
+     */
+    public function testInvalidClusterNumber(int $clusters): void
+    {
+        $this->expectException(InvalidArgumentException::class);
+
+        new FuzzyCMeans($clusters);
+    }
+
+    public function invalidClusterNumberProvider(): array
+    {
+        return [[0], [-1]];
+    }
 }

tests/Helper/Optimizer/ConjugateGradientTest.php

@@ -4,6 +4,7 @@ declare(strict_types=1);
 
 namespace Phpml\Tests\Helper\Optimizer;
 
+use Phpml\Exception\InvalidArgumentException;
 use Phpml\Helper\Optimizer\ConjugateGradient;
 use PHPUnit\Framework\TestCase;
 
@@ -35,6 +36,34 @@ class ConjugateGradientTest extends TestCase
         $this->assertEquals([-1, 2], $theta, '', 0.1);
     }
 
+    public function testRunOptimizationWithCustomInitialTheta(): void
+    {
+        // 200 samples from y = -1 + 2x (i.e. theta = [-1, 2])
+        $samples = [];
+        $targets = [];
+        for ($i = -100; $i <= 100; ++$i) {
+            $x = $i / 100;
+            $samples[] = [$x];
+            $targets[] = -1 + 2 * $x;
+        }
+
+        $callback = function ($theta, $sample, $target) {
+            $y = $theta[0] + $theta[1] * $sample[0];
+            $cost = ($y - $target) ** 2 / 2;
+            $grad = $y - $target;
+
+            return [$cost, $grad];
+        };
+
+        $optimizer = new ConjugateGradient(1);
+        // set very weak theta to trigger very bad result
+        $optimizer->setInitialTheta([0.0000001, 0.0000001]);
+
+        $theta = $optimizer->runOptimization($samples, $targets, $callback);
+
+        $this->assertEquals([-1.087708, 2.212034], $theta, '', 0.000001);
+    }
+
     public function testRunOptimization2Dim(): void
     {
         // 100 samples from y = -1 + 2x0 - 3x1 (i.e. theta = [-1, 2, -3])
@@ -62,4 +91,12 @@ class ConjugateGradientTest extends TestCase
 
         $this->assertEquals([-1, 2, -3], $theta, '', 0.1);
     }
+
+    public function testThrowExceptionOnInvalidTheta(): void
+    {
+        $opimizer = new ConjugateGradient(2);
+
+        $this->expectException(InvalidArgumentException::class);
+        $opimizer->setInitialTheta([0.15]);
+    }
 }

tests/Math/LinearAlgebra/LUDecompositionTest.php

@@ -0,0 +1,31 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Phpml\Tests\Math\LinearAlgebra;
+
+use Phpml\Exception\MatrixException;
+use Phpml\Math\LinearAlgebra\LUDecomposition;
+use Phpml\Math\Matrix;
+use PHPUnit\Framework\TestCase;
+
+/**
+ * LUDecomposition is used and tested in Matrix::inverse method so not all tests are required
+ */
+final class LUDecompositionTest extends TestCase
+{
+    public function testNotSquareMatrix(): void
+    {
+        $this->expectException(MatrixException::class);
+
+        new LUDecomposition(new Matrix([1, 2, 3, 4, 5]));
+    }
+
+    public function testSolveWithInvalidMatrix(): void
+    {
+        $this->expectException(MatrixException::class);
+
+        $lu = new LUDecomposition(new Matrix([[1, 2], [3, 4]]));
+        $lu->solve(new Matrix([1, 2, 3]));
+    }
+}