Changeset 3996 for trunk/sources/HeuristicLab.Problems.DataAnalysis/3.3/Evaluators

Timestamp:

07/05/10 17:14:22 (14 years ago)

Author:

gkronber

Message:

Improved efficiency of analyzers and evaluators for regression problems. #1074

Location:

trunk/sources/HeuristicLab.Problems.DataAnalysis/3.3/Evaluators

Files:

• OnlineMeanAbsolutePercentageErrorEvaluator.cs (added)
• OnlineMeanSquaredErrorEvaluator.cs (added)
• OnlinePearsonsRSquaredEvaluator.cs (added)
• SimpleMSEEvaluator.cs (modified) (2 diffs)
• SimpleMeanAbsolutePercentageErrorEvaluator.cs (modified) (2 diffs)
• SimpleRSquaredEvaluator.cs (modified) (1 diff)

trunk/sources/HeuristicLab.Problems.DataAnalysis/3.3/Evaluators/SimpleMSEEvaluator.cs

@@ -45 +45 @@
 
     public static double Calculate(IEnumerable<double> original, IEnumerable<double> estimated) {
-      double sse = 0.0;
-      int cnt = 0;
+      var onlineMseEvaluator = new OnlineMeanSquaredErrorEvaluator();
       var originalEnumerator = original.GetEnumerator();
       var estimatedEnumerator = estimated.GetEnumerator();
@@ -52 +51 @@
         double e = estimatedEnumerator.Current;
         double o = originalEnumerator.Current;
-        if (!double.IsNaN(e) && !double.IsInfinity(e) &&
-            !double.IsNaN(o) && !double.IsInfinity(o)) {
-          double error = e - o;
-          sse += error * error;
-          cnt++;
-        }
+        onlineMseEvaluator.Add(o, e);
       }
       if (estimatedEnumerator.MoveNext() || originalEnumerator.MoveNext()) {
         throw new ArgumentException("Number of elements in original and estimated enumeration doesn't match.");
-      } else if (cnt == 0) {
-        throw new ArgumentException("Mean squared errors is not defined for input vectors of NaN or Inf");
       } else {
-        double mse = sse / cnt;
-        return mse;
+        return onlineMseEvaluator.MeanSquaredError;
       }
     }

trunk/sources/HeuristicLab.Problems.DataAnalysis/3.3/Evaluators/SimpleMeanAbsolutePercentageErrorEvaluator.cs

@@ -48 +48 @@
 
     public static double Calculate(IEnumerable<double> original, IEnumerable<double> estimated) {
-      double sre = 0;
-      int cnt = 0;
+      OnlineMeanAbsolutePercentageErrorEvaluator onlineEvaluator = new OnlineMeanAbsolutePercentageErrorEvaluator();
       var originalEnumerator = original.GetEnumerator();
       var estimatedEnumerator = estimated.GetEnumerator();
@@ -55 +54 @@
         double e = estimatedEnumerator.Current;
         double o = originalEnumerator.Current;
-        if (!double.IsNaN(e) && !double.IsInfinity(e) &&
-            !double.IsNaN(o) && !double.IsInfinity(o) && !o.IsAlmost(0.0)) {
-          sre += Math.Abs((e - o) / o);
-          cnt++;
-        }
+        onlineEvaluator.Add(o, e);
       }
       if (estimatedEnumerator.MoveNext() || originalEnumerator.MoveNext()) {
         throw new ArgumentException("Number of elements in original and estimated enumeration doesn't match.");
-      } else if (cnt == 0) {
-        throw new ArgumentException("Average relative error is not defined for input vectors of NaN or Inf");
       } else {
-        return sre / cnt;
+        return onlineEvaluator.MeanAbsolutePercentageError;
       }
     }

trunk/sources/HeuristicLab.Problems.DataAnalysis/3.3/Evaluators/SimpleRSquaredEvaluator.cs

@@ -49 +49 @@
 
     public static double Calculate(IEnumerable<double> original, IEnumerable<double> estimated) {
+      var onlinePearsonRSquaredEvaluator = new OnlinePearsonsRSquaredEvaluator();
       var originalEnumerator = original.GetEnumerator();
       var estimatedEnumerator = estimated.GetEnumerator();
-      originalEnumerator.MoveNext();
-      estimatedEnumerator.MoveNext();
-      double e = estimatedEnumerator.Current;
-      double o = originalEnumerator.Current;
 
-      // stable and iterative calculation of R² in one pass over original and estimated
-      double sum_sq_x = 0.0;
-      double sum_sq_y = 0.0;
-      double sum_coproduct = 0.0;
-      if (IsInvalidValue(o) || IsInvalidValue(e)) {
-        throw new ArgumentException("R² is not defined for variables with NaN or infinity values.");
-      }
-      double mean_x = o;
-      double mean_y = e;
-      int n = 1;
       while (originalEnumerator.MoveNext() & estimatedEnumerator.MoveNext()) {
-        e = estimatedEnumerator.Current;
-        o = originalEnumerator.Current;
-        double sweep = (n - 1.0) / n;
-        if (IsInvalidValue(o) || IsInvalidValue(e)) {
-          throw new ArgumentException("Correlation coefficient is not defined for variables with NaN or infinity values.");
-        }
-        double delta_x = o - mean_x;
-        double delta_y = e - mean_y;
-        sum_sq_x += delta_x * delta_x * sweep;
-        sum_sq_y += delta_y * delta_y * sweep;
-        sum_coproduct += delta_x * delta_y * sweep;
-        mean_x += delta_x / n;
-        mean_y += delta_y / n;
-        n++;
+        double e = estimatedEnumerator.Current;
+        double o = originalEnumerator.Current;
+        onlinePearsonRSquaredEvaluator.Add(o, e);
       }
       if (estimatedEnumerator.MoveNext() || originalEnumerator.MoveNext()) {
         throw new ArgumentException("Number of elements in original and estimated enumeration doesn't match.");
       } else {
-        double pop_sd_x = Math.Sqrt(sum_sq_x / n);
-        double pop_sd_y = Math.Sqrt(sum_sq_y / n);
-        double cov_x_y = sum_coproduct / n;
-
-        if (pop_sd_x.IsAlmost(0.0) || pop_sd_y.IsAlmost(0.0))
-          return 0.0;
-        else {
-          double r = cov_x_y / (pop_sd_x * pop_sd_y);
-          return r * r;
-        }
+        return onlinePearsonRSquaredEvaluator.RSquared;
       }
     }