Free cookie consent management tool by TermsFeed Policy Generator

source: trunk/sources/HeuristicLab.GP.StructureIdentification.TimeSeries/TheilInequalityCoefficientEvaluator.cs @ 695

Last change on this file since 695 was 695, checked in by gkronber, 15 years ago

implemented #324 (Bias, variance and covariance decomposition of theil's inequality)

File size: 5.9 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2008 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;
23using System.Collections.Generic;
24using System.Linq;
25using System.Text;
26using HeuristicLab.Core;
27using HeuristicLab.Data;
28using HeuristicLab.GP.StructureIdentification;
29using HeuristicLab.DataAnalysis;
30
31namespace HeuristicLab.GP.StructureIdentification.TimeSeries {
32  public class TheilInequalityCoefficientEvaluator : GPEvaluatorBase {
33    private DoubleData theilInequaliy;
34    private DoubleData uBias;
35    private DoubleData uVariance;
36    private DoubleData uCovariance;
37
38    public override string Description {
39      get {
40        return @"Evaluates 'FunctionTree' for all samples of 'Dataset' and calculates
41the 'Theil inequality coefficient (Theil's U2 not U1!)' of estimated values vs. real values of 'TargetVariable'.
42
43U2 = Sqrt(1/N * Sum(P_t - A_t)^2 ) / Sqrt(1/N * Sum(A_t)^2 )
44
45where P_t is the predicted change of the target variable and A_t is the measured (original) change.
46(P_t = y'_t - y_(t-1), A_t = y_t - y_(t-1)).
47
48U2 is 0 for a perfect prediction and 1 for the naive model y'_t = y_(t-1). An U2 > 1 means the
49model is worse than the naive model (=> model is useless).";
50      }
51    }
52
53    public TheilInequalityCoefficientEvaluator()
54      : base() {
55      AddVariableInfo(new VariableInfo("TheilInequalityCoefficient", "Theil's inequality coefficient (U2) of the model", typeof(DoubleData), VariableKind.New));
56      AddVariableInfo(new VariableInfo("TheilInequalityCoefficientBias", "Bias proportion of Theil's inequality coefficient", typeof(DoubleData), VariableKind.New));
57      AddVariableInfo(new VariableInfo("TheilInequalityCoefficientVariance", "Variance proportion of Theil's inequality coefficient", typeof(DoubleData), VariableKind.New));
58      AddVariableInfo(new VariableInfo("TheilInequalityCoefficientCovariance", "Covariance proportion of Theil's inequality coefficient", typeof(DoubleData), VariableKind.New));
59    }
60
61    public override IOperation Apply(IScope scope) {
62      theilInequaliy = GetVariableValue<DoubleData>("TheilInequalityCoefficient", scope, false, false);
63      if(theilInequaliy == null) {
64        theilInequaliy = new DoubleData();
65        scope.AddVariable(new HeuristicLab.Core.Variable(scope.TranslateName("TheilInequalityCoefficient"), theilInequaliy));
66      }
67      uBias = GetVariableValue<DoubleData>("TheilInequalityCoefficientBias", scope, false, false);
68      if(uBias == null) {
69        uBias = new DoubleData();
70        scope.AddVariable(new HeuristicLab.Core.Variable(scope.TranslateName("TheilInequalityCoefficientBias"), uBias));
71      }
72      uVariance = GetVariableValue<DoubleData>("TheilInequalityCoefficientVariance", scope, false, false);
73      if(uVariance == null) {
74        uVariance = new DoubleData();
75        scope.AddVariable(new HeuristicLab.Core.Variable(scope.TranslateName("TheilInequalityCoefficientVariance"), uVariance));
76      }
77      uCovariance = GetVariableValue<DoubleData>("TheilInequalityCoefficientCovariance", scope, false, false);
78      if(uCovariance == null) {
79        uCovariance = new DoubleData();
80        scope.AddVariable(new HeuristicLab.Core.Variable(scope.TranslateName("TheilInequalityCoefficientCovariance"), uCovariance));
81      }
82      return base.Apply(scope);
83    }
84
85    public override void Evaluate(int start, int end) {
86      double errorsSquaredSum = 0.0;
87      double originalSquaredSum = 0.0;
88      double[] estimatedChanges = new double[end-start];
89      double[] originalChanges = new double[end-start];
90      int nSamples = 0;
91      for(int sample = start; sample < end; sample++) {
92        double prevValue = GetOriginalValue(sample - 1);
93        double estimatedChange = GetEstimatedValue(sample) - prevValue;
94        double originalChange = GetOriginalValue(sample) - prevValue;
95        SetOriginalValue(sample, estimatedChange + prevValue);
96        if(!double.IsNaN(originalChange) && !double.IsInfinity(originalChange)) {
97          double error = estimatedChange - originalChange;
98          errorsSquaredSum += error * error;
99          originalSquaredSum += originalChange * originalChange;
100          estimatedChanges[sample - start] = estimatedChange;
101          originalChanges[sample - start] = originalChange;
102          nSamples++;
103        }
104      }
105      double quality = Math.Sqrt(errorsSquaredSum / nSamples) / Math.Sqrt(originalSquaredSum / nSamples);
106      if(double.IsNaN(quality) || double.IsInfinity(quality))
107        quality = double.MaxValue;
108      theilInequaliy.Data = quality; // U2
109
110      // decomposition into U_bias + U_variance + U_covariance parts
111      double bias = Statistics.Mean(estimatedChanges) - Statistics.Mean(originalChanges);
112      bias *= bias; // squared
113      uBias.Data = bias / (errorsSquaredSum / nSamples);
114
115      double variance = Statistics.StandardDeviation(estimatedChanges) - Statistics.StandardDeviation(originalChanges);
116      variance *= variance; // squared
117      uVariance.Data = variance / (errorsSquaredSum / nSamples);
118
119      // all parts add up to one so I don't have to calculate the correlation coefficient for the covariance propotion
120      uCovariance.Data = 1.0 - uBias.Data - uVariance.Data;
121    }
122  }
123}
Note: See TracBrowser for help on using the repository browser.