BAEL-3881

algorithms-miscellaneous-6/src/main/java/com/baeldung/algorithms/gradientdescent/GradientDescent.java

@@ -0,0 +1,33 @@
+package com.baeldung.algorithms.gradientdescent;
+
+import java.util.function.Function;
+
+public class GradientDescent {
+
+    private final double precision = 0.000001;
+
+    public double findLocalMinimum(Function<Double, Double> f, double initialX) {
+        double stepCoefficient = 0.1;
+        double previousStep = 1.0;
+        double currentX = initialX;
+        double previousX = initialX;
+        double previousY = f.apply(previousX);
+        int iter = 100;
+
+        currentX += stepCoefficient * previousY;
+
+        while (previousStep > precision && iter > 0) {
+            iter--;
+            double currentY = f.apply(currentX);
+            if (currentY > previousY) {
+                stepCoefficient = -stepCoefficient / 2;
+            }
+            previousX = currentX;
+            currentX += stepCoefficient * previousY;
+            previousY = currentY;
+            previousStep = StrictMath.abs(currentX - previousX);
+        }
+        return currentX;
+    }
+
+}

algorithms-miscellaneous-6/src/test/java/com/baeldung/algorithms/gradientdescent/GradientDescentUnitTest.java

@@ -0,0 +1,20 @@
+package com.baeldung.algorithms.gradientdescent;
+
+import static org.junit.Assert.assertTrue;
+
+import java.util.function.Function;
+
+import org.junit.Test;
+
+public class GradientDescentUnitTest {
+
+    @Test
+    public void givenFunction_whenStartingPointIsOne_thenLocalMinimumIsFound() {
+        Function<Double, Double> df = x ->
+            StrictMath.abs(StrictMath.pow(x, 3)) - (3 * StrictMath.pow(x, 2)) + x;
+        GradientDescent gd = new GradientDescent();
+        double res = gd.findLocalMinimum(df, 1);
+        assertTrue(res > 1.78);
+        assertTrue(res < 1.84);
+    }
+}