add hot-cold learn method

Author: asm-jaime

MLTests/PredictionCompareLearnTest.cs

@@ -5,6 +5,23 @@ namespace MLTests
 {
     public class PredictionCompareLearnTest
     {
+        public double GetPredictFromSimpleNetwork(double input, double weight)
+        {
+            var prediction = input * weight;
+            return prediction;
+        }
+
+        public double GetLearnNetwork(double weight, double error, double errorUp, double errorDn, double step)
+        {
+            var result = weight;
+            if(error > errorUp || error > errorDn)
+            {
+                if (errorDn < errorUp) result = result - step;
+                if (errorUp < errorDn) result = result + step;
+            }
+            return result;
+        }
+
         [Test]
         public void ShouldGetSimpleLearnError()
         {
@@ -18,5 +35,36 @@ public void ShouldGetSimpleLearnError()
             var error = Math.Round(Math.Pow(goal - prediction, 2), 2);
             error.Should().Be(0.3);
         }
+
+        [Test]
+        public void ShouldLearnByHotColdMethod()
+        {
+            var weight = 0.1;
+
+            var toes = 8.5;
+            var winOrLose = 1;
+
+            var input = toes;
+            var goal = winOrLose;
+
+            var prediction = GetPredictFromSimpleNetwork(input, weight);
+            var error = Math.Round(Math.Pow(prediction - goal, 2), 3);
+            error.Should().Be(0.022);
+
+            var step = 0.01;
+            var predictUp = GetPredictFromSimpleNetwork(input, weight + step);
+            var errorUp = Math.Round(Math.Pow(predictUp - goal, 2), 3);
+            errorUp.Should().Be(0.004);
+
+            step = 0.01;
+            var predictDn = GetPredictFromSimpleNetwork(input, weight - step);
+            var errorDn = Math.Round(Math.Pow(predictDn - goal, 2), 3);
+            errorDn.Should().Be(0.055);
+
+            var betterWeight = GetLearnNetwork(weight, error, errorUp, predictDn, step);
+            var betterPrediction = GetPredictFromSimpleNetwork(input, betterWeight);
+            var betterError = Math.Round(Math.Pow(betterPrediction - goal, 2), 3);
+            betterError.Should().Be(0.004);
+        }
     }
 }