Changeset 11914 for trunk/sources/HeuristicLab.Analysis/3.3/Statistics

Timestamp:

02/05/15 16:09:10 (10 years ago)

Author:

ascheibe

Message:

#2031 implemented review comments

Location:

trunk/sources/HeuristicLab.Analysis/3.3/Statistics

Files:

• BonferroniHolm.cs (modified) (1 diff)
• EnumerableStatisticsExtension.cs (modified) (1 diff)
• Fitting (added)
• Fitting/ExpFitting.cs (moved) (moved from trunk/sources/HeuristicLab.Analysis/3.3/Statistics/ExpFitting.cs) (4 diffs)
• Fitting/IFitting.cs (moved) (moved from trunk/sources/HeuristicLab.Analysis/3.3/Statistics/IFitting.cs) (1 diff)
• Fitting/LinearLeastSquaresFitting.cs (moved) (moved from trunk/sources/HeuristicLab.Analysis/3.3/Statistics/LinearLeastSquaresFitting.cs) (4 diffs)
• Fitting/LogFitting.cs (moved) (moved from trunk/sources/HeuristicLab.Analysis/3.3/Statistics/LogFitting.cs) (2 diffs)
• SampleSizeDetermination.cs (modified) (2 diffs)

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/BonferroniHolm.cs

@@ -41 +41 @@
         pValuesIndizes.Add(i, pValues[i]);
       }
-      var sortedPValues = pValuesIndizes.OrderBy(x => x.Value);
+      var sortedPValues = pValuesIndizes.OrderBy(x => x.Value).ToArray();
 
       for (int i = 1; i < k + 1; i++) {
         alphaNiveau[i - 1] = globalAlpha / (k - i + 1);
-        int idx = sortedPValues.ElementAt(i - 1).Key;
+        int idx = sortedPValues[i - 1].Key;
 
         if (i == 1) {
           //true means reject
-          decision[idx] = sortedPValues.ElementAt(i - 1).Value < alphaNiveau[i - 1];
-          adjustedPValues[idx] = sortedPValues.ElementAt(i - 1).Value * (k - i + 1);
+          decision[idx] = sortedPValues[i - 1].Value < alphaNiveau[i - 1];
+          adjustedPValues[idx] = sortedPValues[i - 1].Value * (k - i + 1);
         } else {
-          decision[idx] = decision[sortedPValues.ElementAt(i - 2).Key] ? (sortedPValues.ElementAt(i - 1).Value < alphaNiveau[i - 1]) : false;
-          adjustedPValues[idx] = Math.Max(adjustedPValues[sortedPValues.ElementAt(i - 2).Key], sortedPValues.ElementAt(i - 1).Value * (k - i + 1));
+          decision[idx] = decision[sortedPValues[i - 2].Key] && (sortedPValues[i - 1].Value < alphaNiveau[i - 1]);
+          adjustedPValues[idx] = Math.Max(adjustedPValues[sortedPValues[i - 2].Key], sortedPValues[i - 1].Value * (k - i + 1));
         }
         if (adjustedPValues[idx] > 1.0) {

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/EnumerableStatisticsExtension.cs

@@ -28 +28 @@
   public static class EnumerableStatisticExtensions {
     public static Tuple<double, double> ConfidenceIntervals(this IEnumerable<double> values, double alpha) {
-      if (values.Count() <= 1) return new Tuple<double, double>(double.NaN, double.NaN);
+      return ConfidenceIntervals(values.ToArray(), alpha);
+    }
 
+    public static Tuple<double, double> ConfidenceIntervals(this double[] values, double alpha) {
       double lower, upper;
+      int n = values.Length;
+      if (n <= 1) return new Tuple<double, double>(double.NaN, double.NaN);
+
       double s = values.StandardDeviation();
       double x = values.Average();
-      int n = values.Count();
       double t = alglib.invstudenttdistribution(n - 1, (1.0 - alpha) / 2.0);
 

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/Fitting/ExpFitting.cs

@@ -25 +25 @@
 namespace HeuristicLab.Analysis.Statistics {
   public class ExpFitting : IFitting {
-    private void LogFunc(double[] c, double[] x, ref double func, object obj) {
+    private void ExpFunc(double[] c, double[] x, ref double func, object obj) {
       func = Math.Exp(-c[0] * Math.Pow(x[0], 2));
     }
@@ -34 +34 @@
     }
 
-    public void Calculate(double[] dataPoints, out double p0, out double p1) {
+    public void Calculate(double[] dataPoints, out double p0) {
       var stdX = GetDefaultXValues(dataPoints.Count());
-      Calculate(dataPoints, stdX, out p0, out p1);
+      Calculate(dataPoints, stdX, out p0);
     }
 
-    public void Calculate(double[] y, double[] x, out double p0, out double p1) {
+    public void Calculate(double[] y, double[] x, out double p0) {
       if (y.Count() != x.Count()) {
         throw new ArgumentException("The lenght of x and y needs do be equal. ");
@@ -60 +60 @@
       alglib.lsfitcreatef(xx, y, c, diffstep, out state);
       alglib.lsfitsetcond(state, epsf, epsx, maxits);
-      alglib.lsfitfit(state, LogFunc, null, null);
+      alglib.lsfitfit(state, ExpFunc, null, null);
       alglib.lsfitresults(state, out info, out c, out rep);
 
       p0 = c[0];
-      p1 = c[0];
     }
 
-    public DataRow CalculateFittedLine(double[] dataPoints, string rowName) {
-      DataRow newRow = new DataRow(rowName);
-      double c0, c1;
-      Calculate(dataPoints, out c0, out c1);
+    public DataRow CalculateFittedLine(double[] dataPoints) {
+      DataRow newRow = new DataRow();
+      double c0;
+      Calculate(dataPoints, out c0);
       var stdX = GetDefaultXValues(dataPoints.Count());
 
@@ -80 +79 @@
     }
 
-    public DataRow CalculateFittedLine(double[] y, double[] x, string rowName) {
-      DataRow newRow = new DataRow(rowName);
-      double c0, c1;
-      Calculate(y, x, out c0, out c1);
+    public DataRow CalculateFittedLine(double[] y, double[] x) {
+      DataRow newRow = new DataRow();
+      double c0;
+      Calculate(y, x, out c0);
 
       for (int i = 0; i < x.Count(); i++) {

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/Fitting/IFitting.cs

@@ -22 +22 @@
 namespace HeuristicLab.Analysis.Statistics {
   public interface IFitting {
-    void Calculate(double[] dataPoints, out double p0, out double p1);
-    void Calculate(double[] y, double[] x, out double p0, out double p1);
-
-    DataRow CalculateFittedLine(double[] dataPoints, string rowName);
-    DataRow CalculateFittedLine(double[] y, double[] x, string rowName);
+    DataRow CalculateFittedLine(double[] dataPoints);
+    DataRow CalculateFittedLine(double[] y, double[] x);
   }
 }

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/Fitting/LinearLeastSquaresFitting.cs

@@ -25 +25 @@
 namespace HeuristicLab.Analysis.Statistics {
   public class LinearLeastSquaresFitting : IFitting {
-    public void Calculate(double[] dataPoints, out double p0, out double p1) {
+    public void Calculate(double[] dataPoints, out double slope, out double intercept) {
       var stdX = Enumerable.Range(0, dataPoints.Count()).Select(x => (double)x).ToArray();
-      Calculate(dataPoints, stdX, out p0, out p1);
+      Calculate(dataPoints, stdX, out slope, out intercept);
     }
 
-    public void Calculate(double[] y, double[] x, out double p0, out double p1) {
+    public void Calculate(double[] y, double[] x, out double slope, out double intercept) {
       if (y.Count() != x.Count()) {
         throw new ArgumentException("The lenght of x and y needs do be equal. ");
@@ -48 +48 @@
       }
 
-      p0 = (sxy - (n * avgx * avgy)) / (sxx - (n * avgx * avgx));
-      p1 = avgy - p0 * avgx;
+      slope = (sxy - (n * avgx * avgy)) / (sxx - (n * avgx * avgx));
+      intercept = avgy - slope * avgx;
     }
 
-    public double CalculateError(double[] dataPoints, double p0, double p1) {
+    public double CalculateError(double[] dataPoints, double slope, double intercept) {
       double r;
       double avgy = dataPoints.Average();
@@ -59 +59 @@
 
       for (int i = 0; i < dataPoints.Count(); i++) {
-        double y = p0 * i + p1;
+        double y = slope * i + intercept;
         sstot += Math.Pow(dataPoints[i] - avgy, 2);
         sserr += Math.Pow(dataPoints[i] - y, 2);
@@ -68 +68 @@
     }
 
-    public DataRow CalculateFittedLine(double[] y, double[] x, string rowName) {
-      double k, d;
-      Calculate(y, x, out k, out d);
+    public DataRow CalculateFittedLine(double[] y, double[] x) {
+      double slope, intercept;
+      Calculate(y, x, out slope, out intercept);
 
-      DataRow newRow = new DataRow(rowName);
+      DataRow newRow = new DataRow();
       for (int i = 0; i < x.Count(); i++) {
-        newRow.Values.Add(k * x[i] + d);
+        newRow.Values.Add(slope * x[i] + intercept);
       }
       return newRow;
     }
 
-    public DataRow CalculateFittedLine(double[] dataPoints, string rowName) {
-      DataRow newRow = new DataRow(rowName);
-      double c0, c1;
-      Calculate(dataPoints, out c0, out c1);
+    public DataRow CalculateFittedLine(double[] dataPoints) {
+      DataRow newRow = new DataRow();
+      double slope, intercept;
+      Calculate(dataPoints, out slope, out intercept);
       var stdX = Enumerable.Range(0, dataPoints.Count()).Select(x => (double)x).ToArray();
 
       for (int i = 0; i < stdX.Count(); i++) {
-        newRow.Values.Add(c0 * stdX[i] + c1);
+        newRow.Values.Add(slope * stdX[i] + intercept);
       }
 

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/Fitting/LogFitting.cs

@@ -67 +67 @@
     }
 
-    public DataRow CalculateFittedLine(double[] dataPoints, string rowName) {
-      DataRow newRow = new DataRow(rowName);
+    public DataRow CalculateFittedLine(double[] dataPoints) {
+      DataRow newRow = new DataRow();
       double c0, c1;
       Calculate(dataPoints, out c0, out c1);
@@ -80 +80 @@
     }
 
-    public DataRow CalculateFittedLine(double[] y, double[] x, string rowName) {
-      DataRow newRow = new DataRow(rowName);
+    public DataRow CalculateFittedLine(double[] y, double[] x) {
+      DataRow newRow = new DataRow();
       double c0, c1;
       Calculate(y, x, out c0, out c1);

trunk/sources/HeuristicLab.Analysis/3.3/Statistics/SampleSizeDetermination.cs

@@ -36 +36 @@
       if (conf < 0.0 || conf > 1.0) throw new ArgumentException("The confidence interval must be between zero and one.");
 
-      var confInterval = samples.ConfidenceIntervals(0.95);
+      var confInterval = samples.ConfidenceIntervals(conf);
       double e = (confInterval.Item2 - confInterval.Item1) / 2;
       double s = samples.StandardDeviation();
@@ -43 +43 @@
 
       double result = Math.Pow(s, 2) / ((Math.Pow(e, 2) / Math.Pow(z, 2)) + (Math.Pow(s, 2) / n));
-
-      result = Math.Ceiling(result);
-      if (result > int.MaxValue)
-        return int.MaxValue;
-      else
-        return (int)result;
-    }
-
-    public static int DetermineSampleSizeByEstimatingMeanForLargeSampleSizes(double[] samples, double conf = 0.95) {
-      if (conf < 0.0 || conf > 1.0) throw new ArgumentException("The confidence interval must be between zero and one.");
-
-      var confInterval = samples.ConfidenceIntervals(0.95);
-      double e = (confInterval.Item2 - confInterval.Item1) / 2;
-      double s = samples.StandardDeviation();
-      double z = alglib.invnormaldistribution((conf + 1) / 2);
-
-      double result = Math.Pow(z, 2) * (Math.Pow(s, 2) / Math.Pow(e, 2));
 
       result = Math.Ceiling(result);