1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2011 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 |
|
---|
22 | using System.Collections.Generic;
|
---|
23 | using System.Linq;
|
---|
24 | using HeuristicLab.Common;
|
---|
25 | using HeuristicLab.Core;
|
---|
26 | using HeuristicLab.Data;
|
---|
27 | using HeuristicLab.Operators;
|
---|
28 | using HeuristicLab.Parameters;
|
---|
29 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
30 | using HeuristicLab.Optimization;
|
---|
31 | using System;
|
---|
32 |
|
---|
33 | namespace HeuristicLab.Problems.DataAnalysis {
|
---|
34 | /// <summary>
|
---|
35 | /// Represents a classification solution that uses a discriminant function and classification thresholds.
|
---|
36 | /// </summary>
|
---|
37 | [StorableClass]
|
---|
38 | [Item("DiscriminantFunctionClassificationSolution", "Represents a classification solution that uses a discriminant function and classification thresholds.")]
|
---|
39 | public class DiscriminantFunctionClassificationSolution : ClassificationSolution, IDiscriminantFunctionClassificationSolution {
|
---|
40 | [StorableConstructor]
|
---|
41 | protected DiscriminantFunctionClassificationSolution(bool deserializing) : base(deserializing) { }
|
---|
42 | protected DiscriminantFunctionClassificationSolution(DiscriminantFunctionClassificationSolution original, Cloner cloner)
|
---|
43 | : base(original, cloner) {
|
---|
44 | }
|
---|
45 | public DiscriminantFunctionClassificationSolution(IRegressionModel model, IClassificationProblemData problemData)
|
---|
46 | : this(new DiscriminantFunctionClassificationModel(model, problemData.ClassValues), problemData) {
|
---|
47 | }
|
---|
48 | public DiscriminantFunctionClassificationSolution(IDiscriminantFunctionClassificationModel model, IClassificationProblemData problemData)
|
---|
49 | : base(model, problemData) {
|
---|
50 | Model.ThresholdsChanged += new EventHandler(Model_ThresholdsChanged);
|
---|
51 | }
|
---|
52 |
|
---|
53 | #region IDiscriminantFunctionClassificationSolution Members
|
---|
54 |
|
---|
55 | public new IDiscriminantFunctionClassificationModel Model {
|
---|
56 | get { return (IDiscriminantFunctionClassificationModel)base.Model; }
|
---|
57 | }
|
---|
58 |
|
---|
59 | public IEnumerable<double> EstimatedValues {
|
---|
60 | get { return GetEstimatedValues(Enumerable.Range(0, ProblemData.Dataset.Rows)); }
|
---|
61 | }
|
---|
62 |
|
---|
63 | public IEnumerable<double> EstimatedTrainingValues {
|
---|
64 | get { return GetEstimatedValues(ProblemData.TrainingIndizes); }
|
---|
65 | }
|
---|
66 |
|
---|
67 | public IEnumerable<double> EstimatedTestValues {
|
---|
68 | get { return GetEstimatedValues(ProblemData.TestIndizes); }
|
---|
69 | }
|
---|
70 |
|
---|
71 | public IEnumerable<double> GetEstimatedValues(IEnumerable<int> rows) {
|
---|
72 | return Model.GetEstimatedValues(ProblemData.Dataset, rows);
|
---|
73 | }
|
---|
74 |
|
---|
75 | public IEnumerable<double> Thresholds {
|
---|
76 | get {
|
---|
77 | return Model.Thresholds;
|
---|
78 | }
|
---|
79 | set { Model.Thresholds = new List<double>(value); }
|
---|
80 | }
|
---|
81 |
|
---|
82 | public event EventHandler ThresholdsChanged;
|
---|
83 |
|
---|
84 | private void Model_ThresholdsChanged(object sender, EventArgs e) {
|
---|
85 | OnThresholdsChanged(e);
|
---|
86 | }
|
---|
87 |
|
---|
88 | protected virtual void OnThresholdsChanged(EventArgs e) {
|
---|
89 | var listener = ThresholdsChanged;
|
---|
90 | if (listener != null) listener(this, e);
|
---|
91 | }
|
---|
92 | #endregion
|
---|
93 |
|
---|
94 | public override IEnumerable<double> GetEstimatedClassValues(IEnumerable<int> rows) {
|
---|
95 | if (Model.Thresholds == null || Model.Thresholds.Count() == 0) RecalculateClassIntermediates();
|
---|
96 | return base.GetEstimatedClassValues(rows);
|
---|
97 | }
|
---|
98 |
|
---|
99 | private void RecalculateClassIntermediates() {
|
---|
100 | int slices = 100;
|
---|
101 | List<double> estimatedValues = EstimatedValues.ToList();
|
---|
102 | List<int> classInstances = (from classValue in ProblemData.Dataset.GetVariableValues(ProblemData.TargetVariable)
|
---|
103 | group classValue by classValue into grouping
|
---|
104 | select grouping.Count()).ToList();
|
---|
105 | double maxEstimatedValue = estimatedValues.Max();
|
---|
106 | double minEstimatedValue = estimatedValues.Min();
|
---|
107 | List<KeyValuePair<double, double>> estimatedTargetValues =
|
---|
108 | (from row in ProblemData.TrainingIndizes
|
---|
109 | select new KeyValuePair<double, double>(
|
---|
110 | estimatedValues[row],
|
---|
111 | ProblemData.Dataset[ProblemData.TargetVariable, row])).ToList();
|
---|
112 |
|
---|
113 | List<double> originalClasses = ProblemData.ClassValues.OrderBy(x => x).ToList();
|
---|
114 | int nClasses = originalClasses.Distinct().Count();
|
---|
115 | double[] thresholds = new double[nClasses + 1];
|
---|
116 | thresholds[0] = double.NegativeInfinity;
|
---|
117 | thresholds[thresholds.Length - 1] = double.PositiveInfinity;
|
---|
118 |
|
---|
119 | double thresholdIncrement = (maxEstimatedValue - minEstimatedValue) / slices;
|
---|
120 |
|
---|
121 | for (int i = 1; i < thresholds.Length - 1; i++) {
|
---|
122 | double lowerThreshold = thresholds[i - 1];
|
---|
123 | double actualThreshold = Math.Max(lowerThreshold, minEstimatedValue);
|
---|
124 | double lowestBestThreshold = double.NaN;
|
---|
125 | double highestBestThreshold = double.NaN;
|
---|
126 | double bestClassificationScore = double.PositiveInfinity;
|
---|
127 | bool seriesOfEqualClassificationScores = false;
|
---|
128 |
|
---|
129 | while (actualThreshold < maxEstimatedValue) {
|
---|
130 | double classificationScore = 0.0;
|
---|
131 |
|
---|
132 | foreach (KeyValuePair<double, double> estimatedTarget in estimatedTargetValues) {
|
---|
133 | //all positives
|
---|
134 | if (estimatedTarget.Value.IsAlmost(originalClasses[i - 1])) {
|
---|
135 | if (estimatedTarget.Key > lowerThreshold && estimatedTarget.Key < actualThreshold)
|
---|
136 | //true positive
|
---|
137 | classificationScore += ProblemData.GetClassificationPenalty(originalClasses[i - 1], originalClasses[i - 1]);
|
---|
138 | else
|
---|
139 | //false negative
|
---|
140 | classificationScore += ProblemData.GetClassificationPenalty(originalClasses[i], originalClasses[i - 1]);
|
---|
141 | }
|
---|
142 | //all negatives
|
---|
143 | else {
|
---|
144 | if (estimatedTarget.Key > lowerThreshold && estimatedTarget.Key < actualThreshold)
|
---|
145 | //false positive
|
---|
146 | classificationScore += ProblemData.GetClassificationPenalty(originalClasses[i - 1], originalClasses[i]);
|
---|
147 | else
|
---|
148 | //true negative, consider only upper class
|
---|
149 | classificationScore += ProblemData.GetClassificationPenalty(originalClasses[i], originalClasses[i]);
|
---|
150 | }
|
---|
151 | }
|
---|
152 |
|
---|
153 | //new best classification score found
|
---|
154 | if (classificationScore < bestClassificationScore) {
|
---|
155 | bestClassificationScore = classificationScore;
|
---|
156 | lowestBestThreshold = actualThreshold;
|
---|
157 | highestBestThreshold = actualThreshold;
|
---|
158 | seriesOfEqualClassificationScores = true;
|
---|
159 | }
|
---|
160 | //equal classification scores => if seriesOfEqualClassifcationScores == true update highest threshold
|
---|
161 | else if (Math.Abs(classificationScore - bestClassificationScore) < double.Epsilon && seriesOfEqualClassificationScores)
|
---|
162 | highestBestThreshold = actualThreshold;
|
---|
163 | //worse classificatoin score found reset seriesOfEqualClassifcationScores
|
---|
164 | else seriesOfEqualClassificationScores = false;
|
---|
165 |
|
---|
166 | actualThreshold += thresholdIncrement;
|
---|
167 | }
|
---|
168 | //scale lowest thresholds and highest found optimal threshold according to the misclassification matrix
|
---|
169 | double falseNegativePenalty = ProblemData.GetClassificationPenalty(originalClasses[i], originalClasses[i - 1]);
|
---|
170 | double falsePositivePenalty = ProblemData.GetClassificationPenalty(originalClasses[i - 1], originalClasses[i]);
|
---|
171 | thresholds[i] = (lowestBestThreshold * falsePositivePenalty + highestBestThreshold * falseNegativePenalty) / (falseNegativePenalty + falsePositivePenalty);
|
---|
172 | }
|
---|
173 | Thresholds = new List<double>(thresholds);
|
---|
174 | }
|
---|
175 | }
|
---|
176 | }
|
---|