Changeset 2226

Timestamp:

08/03/09 15:18:36 (15 years ago)

Author:

gkronber

Message:

• Added extension method IsAlmost to type double
• changed evaluation semantics of boolean operations in HL3TreeEvaluator

(#713)

Location:

trunk/sources

Files:

• HeuristicLab.GP.StructureIdentification/3.3/HL3TreeEvaluator.cs (modified) (7 diffs)
• HeuristicLab.LinearRegression/3.2/LinearRegressionOperator.cs (modified) (3 diffs)
• HeuristicLab.Modeling/3.2/DoubleExtensions.cs (added)
• HeuristicLab.Modeling/3.2/HeuristicLab.Modeling-3.2.csproj (modified) (1 diff)
• HeuristicLab.Modeling/3.2/SimpleEvaluatorBase.cs (modified) (1 diff)
• HeuristicLab.Modeling/3.2/SimpleMeanAbsolutePercentageOfRangeErrorEvaluator.cs (modified) (1 diff)
• HeuristicLab.Modeling/3.2/SimpleVarianceAccountedForEvaluator.cs (modified) (1 diff)
• HeuristicLab.Modeling/3.2/VariableEvaluationImpactCalculator.cs (modified) (1 diff)

trunk/sources/HeuristicLab.GP.StructureIdentification/3.3/HL3TreeEvaluator.cs

@@ -22 +22 @@
 using System;
 using System.Diagnostics;
+using HeuristicLab.Modeling; // double.IsAlmost extension
 
 namespace HeuristicLab.GP.StructureIdentification {
@@ -29 +30 @@
   /// </summary>
   public class HL3TreeEvaluator : TreeEvaluatorBase {
-
-
     protected override double EvaluateBakedCode() {
       Instr currInstr = codeArr[PC++];
@@ -36 +35 @@
         case EvaluatorSymbolTable.VARIABLE: {
             int row = sampleIndex + currInstr.i_arg1;
-            if (row < 0 || row >= dataset.Rows) return double.NaN;
+            if (row < 0 || row >= dataset.Rows) throw new InvalidOperationException("Out of range access to dataset row: " + row);
             else return currInstr.d_arg0 * dataset.GetValue(row, currInstr.i_arg0);
           }
@@ -44 +43 @@
         case EvaluatorSymbolTable.DIFFERENTIAL: {
             int row = sampleIndex + currInstr.i_arg1;
-            if (row < 0 || row >= dataset.Rows) return double.NaN;
+            if (row < 0 || row >= dataset.Rows) throw new InvalidOperationException("Out of range access to dataset row: " + row);
             else if (row < 1) return 0.0;
             else {
@@ -117 +116 @@
             return Math.Tan(EvaluateBakedCode());
           }
-        case EvaluatorSymbolTable.AND: { // only defined for inputs 1 and 0
+        case EvaluatorSymbolTable.AND: {
             double result = EvaluateBakedCode();
             for (int i = 1; i < currInstr.arity; i++) {
-              if (result == 0.0) SkipBakedCode();
+              if (result < 0.0) SkipBakedCode();
               else {
                 result = EvaluateBakedCode();
               }
-              Debug.Assert(result == 0.0 || result == 1.0);
             }
-            return result;
+            return Math.Sign(result);
           }
         case EvaluatorSymbolTable.EQU: {
             double x = EvaluateBakedCode();
             double y = EvaluateBakedCode();
-            if (Math.Abs(x - y) < EPSILON) return 1.0; else return 0.0;
+            if (x.IsAlmost(y)) return 1.0; else return -1.0;
           }
         case EvaluatorSymbolTable.GT: {
@@ -137 +135 @@
             double y = EvaluateBakedCode();
             if (x > y) return 1.0;
-            else return 0.0;
+            else return -1.0;
           }
-        case EvaluatorSymbolTable.IFTE: { // only defined for condition 0 or 1
+        case EvaluatorSymbolTable.IFTE: {
             double condition = EvaluateBakedCode();
-            Debug.Assert(condition == 0.0 || condition == 1.0);
             double result;
-            if (condition == 0.0) {
+            if (condition < 0.0) {
               result = EvaluateBakedCode(); SkipBakedCode();
             } else {
@@ -154 +151 @@
             double y = EvaluateBakedCode();
             if (x < y) return 1.0;
-            else return 0.0;
+            else return -1.0;
           }
-        case EvaluatorSymbolTable.NOT: { // only defined for inputs 0 or 1
-            double result = EvaluateBakedCode();
-            Debug.Assert(result == 0.0 || result == 1.0);
-            return Math.Abs(result - 1.0);
+        case EvaluatorSymbolTable.NOT: {
+            return -EvaluateBakedCode();
           }
-        case EvaluatorSymbolTable.OR: { // only defined for inputs 0 or 1
+        case EvaluatorSymbolTable.OR: {
             double result = EvaluateBakedCode();
             for (int i = 1; i < currInstr.arity; i++) {
-              if (result > 0.0) SkipBakedCode();
+              if (result >= 0.0) SkipBakedCode();
               else {
                 result = EvaluateBakedCode();
-                Debug.Assert(result == 0.0 || result == 1.0);
               }
             }
-            return result;
+            return Math.Sign(result);
           }
-        case EvaluatorSymbolTable.XOR: { // only defined for inputs 0 or 1
+        case EvaluatorSymbolTable.XOR: {
             double x = EvaluateBakedCode();
             double y = EvaluateBakedCode();
-            return Math.Abs(x - y);
+            if (x > y) {
+              double tmp = x;
+              x = y;
+              y = tmp;
+            }
+            // invariant y >= x 
+            if (y < 0.0 || x > 0.0) return -1.0;
+            else return 1.0;
           }
         default: {

trunk/sources/HeuristicLab.LinearRegression/3.2/LinearRegressionOperator.cs

@@ -26 +26 @@
 using HeuristicLab.Data;
 using HeuristicLab.DataAnalysis;
+using HeuristicLab.Modeling;
 using HeuristicLab.GP;
 using HeuristicLab.GP.StructureIdentification;
@@ -54 +55 @@
       double[] coefficients = CalculateCoefficients(inputMatrix, targetVector);
       IFunctionTree tree = CreateModel(coefficients, allowedColumns.Select(i => dataset.GetVariableName(i)).ToList());
-      
+
       scope.AddVariable(new HeuristicLab.Core.Variable(scope.TranslateName("LinearRegressionModel"), new GeneticProgrammingModel(tree)));
       return null;
-    }
-
-    private bool IsAlmost(double x, double y) {
-      return Math.Abs(x - y) < 1.0E-12;
     }
 
@@ -125 +122 @@
       for (int i = 0; i < dataset.Columns; i++) {
         if (i == targetVariable) continue;
-        if (!IsAlmost(dataset.GetMinimum(i, start, end), 0.0) ||
-            !IsAlmost(dataset.GetMaximum(i, start, end), 0.0))
+        if (!dataset.GetMinimum(i, start, end).IsAlmost(0.0) ||
+            !dataset.GetMaximum(i, start, end).IsAlmost(0.0))
           allowedColumns.Add(i);
       }

trunk/sources/HeuristicLab.Modeling/3.2/HeuristicLab.Modeling-3.2.csproj

@@ -83 +83 @@
   <ItemGroup>
     <Compile Include="ClassificationProblemInjector.cs" />
+    <Compile Include="DoubleExtensions.cs" />
     <Compile Include="VariableImpactCalculatorBase.cs" />
     <Compile Include="VariableEvaluationImpactCalculator.cs" />

trunk/sources/HeuristicLab.Modeling/3.2/SimpleEvaluatorBase.cs

@@ -31 +31 @@
 
     public abstract double Evaluate(double[,] values);
-
-    protected static bool IsAlmost(double x, double y) {
-      return Math.Abs(x - y) < 1.0E-12;
-    }
   }
 }

trunk/sources/HeuristicLab.Modeling/3.2/SimpleMeanAbsolutePercentageOfRangeErrorEvaluator.cs

@@ -53 +53 @@
       double range = Statistics.Range(originalValues);
       if (double.IsInfinity(range)) throw new ArgumentException("Range of elements in values is infinity");
-      if (IsAlmost(range, 0.0)) throw new ArgumentException("Range of elements in values is zero");
+      if (range.IsAlmost(0.0)) throw new ArgumentException("Range of elements in values is zero");
 
       for (int i = 0; i < values.GetLength(0); i++) {

trunk/sources/HeuristicLab.Modeling/3.2/SimpleVarianceAccountedForEvaluator.cs

@@ -69 +69 @@
       double errorsVariance = Statistics.Variance(errors);
       double originalsVariance = Statistics.Variance(originalTargetVariableValues);
-      if (IsAlmost(originalsVariance, 0.0))
-        if (IsAlmost(errorsVariance, 0.0)) {
+      if (originalsVariance.IsAlmost(0.0))
+        if (errorsVariance.IsAlmost(0.0)) {
           return 1.0;
         } else {

trunk/sources/HeuristicLab.Modeling/3.2/VariableEvaluationImpactCalculator.cs

@@ -64 +64 @@
     protected override double[] PostProcessImpacts(double[] impacts) {
       double mseSum = impacts.Sum();
-      if (IsAlmost(mseSum, 0.0)) mseSum = 1.0;
+      if (mseSum.IsAlmost(0.0)) mseSum = 1.0;
       for (int i = 0; i < impacts.Length; i++) {
         impacts[i] = impacts[i] / mseSum;