add learn with freezy weight

Author: asm-jaime

MLTests/GradientMulInOneOutTests.cs

@@ -15,7 +15,7 @@ private double WeightsSum(double[] vecA, double[] vecB)
             return result;
         }
 
-        private double[] WeightsSub(double[] vecA, double sub)
+        private double[] WeightSub(double[] vecA, double sub)
         {
             var result = new double[vecA.Length];
             for (int i = 0; i < vecA.Length; i++)
@@ -26,6 +26,17 @@ private double[] WeightsSub(double[] vecA, double sub)
 
         }
 
+        private double[] WeightsAlphaSub(double[] vecA, double[] vecB, double alpha)
+        {
+            var result = new double[vecA.Length];
+            for (int i = 0; i < vecA.Length; i++)
+            {
+                result[i] = vecA[i] - vecB[i] * alpha;
+            }
+            return result;
+
+        }
+
         private double[] WeightsMul(double[] vecA, double[] vecB)
         {
             var result = new double[vecA.Length];
@@ -45,7 +56,32 @@ private double[] GetLearnByGradientMul(double[] weights, double[] input, double
                 var error = Math.Pow(prediction - predictionGoal, 2);
                 var delta = prediction - predictionGoal;
                 var deltaWeight = delta * prediction;
-                result = WeightsSub(result, alpha * deltaWeight);
+                result = WeightSub(result, alpha * deltaWeight);
+            }
+            return result;
+        }
+        
+        private double[] ElementMul(double[] vecA, double element)
+        {
+            var result = new double[vecA.Length];
+            for (int i = 0; i < vecA.Length; i++)
+            {
+                result[i] = vecA[i] * element;
+            }
+            return result;
+        }
+
+        private double[] GetLearnByGradientFreezyOne(double[] weights, int freezyIndex, double[] input, double predictionGoal, double alpha, double iterations)
+        {
+            var result = weights.ToArray();
+            for(int iteration = 0; iteration < iterations; ++iteration)
+            {
+                var prediction = WeightsSum(input, result);
+                var error = Math.Pow(prediction - predictionGoal, 2);
+                var delta = prediction - predictionGoal;
+                var deltaWeights = ElementMul(input, delta);
+                deltaWeights[freezyIndex] = 0;
+                result = WeightsAlphaSub(result, deltaWeights, alpha);
             }
             return result;
         }
@@ -72,5 +108,28 @@ public void ShouldGradientLearnByMultipleInputOneOutput()
 
             //var learnWeights = 
         }
+
+        [Test]
+        public void ShouldGradientLearnByMultipleWhenFreezyOneWeight()
+        {
+            var weights = new double[] { 0.1, 0.2, -0.1 };
+            var alpha = 0.5;
+            var iterations = 20;
+
+            var toes = new double[] { 8.5, 9.5, 9.9, 9.0 };
+            var wlrec = new double[] { 0.65, 0.8, 0.8, 0.9 };
+            var nfans = new double[] { 1.2, 1.3, 0.5, 1.0 };
+
+            var winOrLoseData = new double[] {1, 1, 0, -1};
+
+            var predictionGoal = winOrLoseData[0];
+            var input = new double[] { toes[0], wlrec[0], nfans[0] };
+
+            var result = GetLearnByGradientFreezyOne(weights, 0, input, predictionGoal, alpha, iterations);
+
+            Assert.That(WeightsSum(result, input), Is.EqualTo(1.0));
+
+            //var learnWeights = 
+        }
     }
 }