Changeset 5417

Timestamp:

02/03/11 23:02:05 (14 years ago)

Author:

mkommend

Message:

Adapted RocCurvesView to treat binary classification problems differently (ticket #1289).

File:

• trunk/sources/HeuristicLab.Problems.DataAnalysis.Classification.Views/3.3/RocCurvesView.cs (modified) (2 diffs)

trunk/sources/HeuristicLab.Problems.DataAnalysis.Classification.Views/3.3/RocCurvesView.cs

@@ -123 +123 @@
           for (double lowerThreshold = minThreshold; lowerThreshold < maxThreshold; lowerThreshold += thresholdIncrement) {
             for (double upperThreshold = lowerThreshold + thresholdIncrement; upperThreshold < maxThreshold; upperThreshold += thresholdIncrement) {
+              //only adapt lower threshold for binary classification problems and upper class prediction              
+              if (classValues.Count == 2 && classValue == classValues[1]) upperThreshold = double.PositiveInfinity;
+
               int truePositives = 0;
               int falsePositives = 0;
@@ -142 +145 @@
               }
             }
+            //only adapt upper threshold for binary classification problems and upper class prediction              
+            if (classValues.Count == 2 && classValue == classValues[0]) lowerThreshold = double.PositiveInfinity;
           }