source: branches/ClassificationEnsembleVoting/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/WeightCalculators/MedianThresholdCalculator.cs @ 8814

Last change on this file since 8814 was 8814, checked in by sforsten, 7 years ago

#1776:

  • improved performance of confidence calculation
  • fixed bug in median confidence calculation
  • fixed bug in average confidence calculation
  • confidence calculation is now easier for training and test
  • removed obsolete view ClassificationEnsembleSolutionConfidenceAccuracyDependence
File size: 5.6 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2012 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
4 *
5 * This file is part of HeuristicLab.
6 *
7 * HeuristicLab is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 3 of the License, or
10 * (at your option) any later version.
11 *
12 * HeuristicLab is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
19 */
20#endregion
21
22using System.Collections;
23using System.Collections.Generic;
24using System.Linq;
25using HeuristicLab.Common;
26using HeuristicLab.Core;
27using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
28
29namespace HeuristicLab.Problems.DataAnalysis {
30  [StorableClass]
31  [Item("MedianThresholdCalculator", "")]
32  public class MedianThresholdCalculator : DiscriminantClassificationWeightCalculator {
33
34    public MedianThresholdCalculator()
35      : base() {
36    }
37    [StorableConstructor]
38    protected MedianThresholdCalculator(bool deserializing) : base(deserializing) { }
39    protected MedianThresholdCalculator(MedianThresholdCalculator original, Cloner cloner)
40      : base(original, cloner) {
41    }
42    public override IDeepCloneable Clone(Cloner cloner) {
43      return new MedianThresholdCalculator(this, cloner);
44    }
45
46    protected double[] threshold;
47    protected double[] classValues;
48
49    protected override IEnumerable<double> DiscriminantCalculateWeights(IEnumerable<IDiscriminantFunctionClassificationSolution> discriminantSolutions) {
50      classValues = discriminantSolutions.First().Model.ClassValues.ToArray();
51      var modelThresholds = discriminantSolutions.Select(x => x.Model.Thresholds.ToArray());
52      threshold = new double[modelThresholds.First().GetLength(0)];
53      for (int i = 0; i < modelThresholds.First().GetLength(0); i++) {
54        threshold[i] = GetMedian(modelThresholds.Select(x => x[i]).ToList());
55      }
56      return Enumerable.Repeat<double>(1, discriminantSolutions.Count());
57    }
58
59    protected override double GetDiscriminantConfidence(IEnumerable<IDiscriminantFunctionClassificationSolution> solutions, int index, double estimatedClassValue, CheckPoint handler) {
60      Dataset dataset = solutions.First().ProblemData.Dataset;
61      IList<double> values = solutions.Where(s => handler(s.ProblemData, index)).Select(s => s.Model.GetEstimatedValues(dataset, Enumerable.Repeat(index, 1)).First()).ToList();
62      if (values.Count <= 0)
63        return double.NaN;
64      double median = GetMedian(values);
65      return GetMedianConfidence(median, estimatedClassValue);
66    }
67
68    public override IEnumerable<double> GetDiscriminantConfidence(IEnumerable<IDiscriminantFunctionClassificationSolution> solutions, IEnumerable<int> indices, IEnumerable<double> estimatedClassValue, CheckPoint handler) {
69
70      Dataset dataset = solutions.First().ProblemData.Dataset;
71      List<int> indicesList = indices.ToList();
72      var solValues = solutions.ToDictionary(x => x, x => x.Model.GetEstimatedValues(dataset, indicesList).ToArray());
73      double[] confidences = new double[indices.Count()];
74      double[] estimatedClassValueArr = estimatedClassValue.ToArray();
75
76      for (int i = 0; i < indicesList.Count; i++) {
77        var values = solValues.Where(x => handler(x.Key.ProblemData, indicesList[i])).Select(x => x.Value[i]).ToList();
78        if (values.Count <= 0) {
79          confidences[i] = double.NaN;
80        } else {
81          double median = GetMedian(values);
82          confidences[i] = GetMedianConfidence(median, estimatedClassValueArr[i]);
83        }
84      }
85
86      return confidences;
87    }
88
89    protected double GetMedianConfidence(double median, double estimatedClassValue) {
90      for (int i = 0; i < classValues.Length; i++) {
91        if (estimatedClassValue.Equals(classValues[i])) {
92          //special case: median is higher than value of highest class
93          if (i == classValues.Length - 1 && median >= estimatedClassValue) {
94            return 1;
95          }
96          //special case: median is lower than value of lowest class
97          if (i == 0 && median < estimatedClassValue) {
98            return 1;
99          }
100          //special case: median is not between threshold of estimated class value
101          if ((i < classValues.Length - 1 && median >= threshold[i + 1]) || median <= threshold[i]) {
102            return 0;
103          }
104
105          double thresholdToClassDistance, thresholdToMedianValueDistance;
106          if (median >= classValues[i]) {
107            thresholdToClassDistance = threshold[i + 1] - classValues[i];
108            thresholdToMedianValueDistance = threshold[i + 1] - median;
109          } else {
110            thresholdToClassDistance = classValues[i] - threshold[i];
111            thresholdToMedianValueDistance = median - threshold[i];
112          }
113          return (1 / thresholdToClassDistance) * thresholdToMedianValueDistance;
114        }
115      }
116      return double.NaN;
117    }
118
119    protected double GetMedian(IList<double> estimatedValues) {
120      int count = estimatedValues.Count;
121      if (count % 2 == 0)
122        return 0.5 * (estimatedValues[count / 2 - 1] + estimatedValues[count / 2]);
123      else
124        return estimatedValues[count / 2];
125    }
126  }
127}
Note: See TracBrowser for help on using the repository browser.