Free cookie consent management tool by TermsFeed Policy Generator

source: trunk/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionSolutionBase.cs @ 17982

Last change on this file since 17982 was 17180, checked in by swagner, 5 years ago

#2875: Removed years in copyrights

File size: 17.1 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 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 HeuristicLab.Common;
25using HeuristicLab.Data;
26using HeuristicLab.Optimization;
27using HEAL.Attic;
28
29namespace HeuristicLab.Problems.DataAnalysis {
30  [StorableType("03116F5E-ABBF-4966-9428-E3DC67D1D03D")]
31  public abstract class RegressionSolutionBase : DataAnalysisSolution, IRegressionSolution {
32    protected const string TrainingMeanSquaredErrorResultName = "Mean squared error (training)";
33    protected const string TestMeanSquaredErrorResultName = "Mean squared error (test)";
34    protected const string TrainingMeanAbsoluteErrorResultName = "Mean absolute error (training)";
35    protected const string TestMeanAbsoluteErrorResultName = "Mean absolute error (test)";
36    protected const string TrainingSquaredCorrelationResultName = "Pearson's R² (training)";
37    protected const string TestSquaredCorrelationResultName = "Pearson's R² (test)";
38    protected const string TrainingRelativeErrorResultName = "Average relative error (training)";
39    protected const string TestRelativeErrorResultName = "Average relative error (test)";
40    protected const string TrainingNormalizedMeanSquaredErrorResultName = "Normalized mean squared error (training)";
41    protected const string TestNormalizedMeanSquaredErrorResultName = "Normalized mean squared error (test)";
42    protected const string TrainingRootMeanSquaredErrorResultName = "Root mean squared error (training)";
43    protected const string TestRootMeanSquaredErrorResultName = "Root mean squared error (test)";
44
45    // BackwardsCompatibility3.3
46    #region Backwards compatible code, remove with 3.5
47    private const string TrainingMeanErrorResultName = "Mean error (training)";
48    private const string TestMeanErrorResultName = "Mean error (test)";
49    #endregion
50
51
52    protected const string TrainingMeanSquaredErrorResultDescription = "Mean of squared errors of the model on the training partition";
53    protected const string TestMeanSquaredErrorResultDescription = "Mean of squared errors of the model on the test partition";
54    protected const string TrainingMeanAbsoluteErrorResultDescription = "Mean of absolute errors of the model on the training partition";
55    protected const string TestMeanAbsoluteErrorResultDescription = "Mean of absolute errors of the model on the test partition";
56    protected const string TrainingSquaredCorrelationResultDescription = "Squared Pearson's correlation coefficient of the model output and the actual values on the training partition";
57    protected const string TestSquaredCorrelationResultDescription = "Squared Pearson's correlation coefficient of the model output and the actual values on the test partition";
58    protected const string TrainingRelativeErrorResultDescription = "Average of the relative errors of the model output and the actual values on the training partition";
59    protected const string TestRelativeErrorResultDescription = "Average of the relative errors of the model output and the actual values on the test partition";
60    protected const string TrainingNormalizedMeanSquaredErrorResultDescription = "Normalized mean of squared errors of the model on the training partition";
61    protected const string TestNormalizedMeanSquaredErrorResultDescription = "Normalized mean of squared errors of the model on the test partition";
62    protected const string TrainingRootMeanSquaredErrorResultDescription = "Root mean of squared errors of the model on the training partition";
63    protected const string TestRootMeanSquaredErrorResultDescription = "Root mean of squared errors of the model on the test partition";
64
65    public new IRegressionModel Model {
66      get { return (IRegressionModel)base.Model; }
67      protected set { base.Model = value; }
68    }
69
70    public new IRegressionProblemData ProblemData {
71      get { return (IRegressionProblemData)base.ProblemData; }
72      set {
73        if (value == null) throw new ArgumentNullException("The problemData must not be null.");
74        string errorMessage = string.Empty;
75        if (!Model.IsProblemDataCompatible(value, out errorMessage)) throw new ArgumentException(errorMessage);
76
77        base.ProblemData = value;
78      }
79    }
80
81    public abstract IEnumerable<double> EstimatedValues { get; }
82    public abstract IEnumerable<double> EstimatedTrainingValues { get; }
83    public abstract IEnumerable<double> EstimatedTestValues { get; }
84    public abstract IEnumerable<double> GetEstimatedValues(IEnumerable<int> rows);
85
86    #region Results
87    public double TrainingMeanSquaredError {
88      get { return ((DoubleValue)this[TrainingMeanSquaredErrorResultName].Value).Value; }
89      private set { ((DoubleValue)this[TrainingMeanSquaredErrorResultName].Value).Value = value; }
90    }
91    public double TestMeanSquaredError {
92      get { return ((DoubleValue)this[TestMeanSquaredErrorResultName].Value).Value; }
93      private set { ((DoubleValue)this[TestMeanSquaredErrorResultName].Value).Value = value; }
94    }
95    public double TrainingMeanAbsoluteError {
96      get { return ((DoubleValue)this[TrainingMeanAbsoluteErrorResultName].Value).Value; }
97      private set { ((DoubleValue)this[TrainingMeanAbsoluteErrorResultName].Value).Value = value; }
98    }
99    public double TestMeanAbsoluteError {
100      get { return ((DoubleValue)this[TestMeanAbsoluteErrorResultName].Value).Value; }
101      private set { ((DoubleValue)this[TestMeanAbsoluteErrorResultName].Value).Value = value; }
102    }
103    public double TrainingRSquared {
104      get { return ((DoubleValue)this[TrainingSquaredCorrelationResultName].Value).Value; }
105      private set { ((DoubleValue)this[TrainingSquaredCorrelationResultName].Value).Value = value; }
106    }
107    public double TestRSquared {
108      get { return ((DoubleValue)this[TestSquaredCorrelationResultName].Value).Value; }
109      private set { ((DoubleValue)this[TestSquaredCorrelationResultName].Value).Value = value; }
110    }
111    public double TrainingRelativeError {
112      get { return ((DoubleValue)this[TrainingRelativeErrorResultName].Value).Value; }
113      private set { ((DoubleValue)this[TrainingRelativeErrorResultName].Value).Value = value; }
114    }
115    public double TestRelativeError {
116      get { return ((DoubleValue)this[TestRelativeErrorResultName].Value).Value; }
117      private set { ((DoubleValue)this[TestRelativeErrorResultName].Value).Value = value; }
118    }
119    public double TrainingNormalizedMeanSquaredError {
120      get { return ((DoubleValue)this[TrainingNormalizedMeanSquaredErrorResultName].Value).Value; }
121      private set { ((DoubleValue)this[TrainingNormalizedMeanSquaredErrorResultName].Value).Value = value; }
122    }
123    public double TestNormalizedMeanSquaredError {
124      get { return ((DoubleValue)this[TestNormalizedMeanSquaredErrorResultName].Value).Value; }
125      private set { ((DoubleValue)this[TestNormalizedMeanSquaredErrorResultName].Value).Value = value; }
126    }
127    public double TrainingRootMeanSquaredError {
128      get { return ((DoubleValue)this[TrainingRootMeanSquaredErrorResultName].Value).Value; }
129      private set { ((DoubleValue)this[TrainingRootMeanSquaredErrorResultName].Value).Value = value; }
130    }
131    public double TestRootMeanSquaredError {
132      get { return ((DoubleValue)this[TestRootMeanSquaredErrorResultName].Value).Value; }
133      private set { ((DoubleValue)this[TestRootMeanSquaredErrorResultName].Value).Value = value; }
134    }
135
136    // BackwardsCompatibility3.3
137    #region Backwards compatible code, remove with 3.5
138    private double TrainingMeanError {
139      get {
140        if (!ContainsKey(TrainingMeanErrorResultName)) return double.NaN;
141        return ((DoubleValue)this[TrainingMeanErrorResultName].Value).Value;
142      }
143      set {
144        if (ContainsKey(TrainingMeanErrorResultName))
145          ((DoubleValue)this[TrainingMeanErrorResultName].Value).Value = value;
146      }
147    }
148    private double TestMeanError {
149      get {
150        if (!ContainsKey(TestMeanErrorResultName)) return double.NaN;
151        return ((DoubleValue)this[TestMeanErrorResultName].Value).Value;
152      }
153      set {
154        if (ContainsKey(TestMeanErrorResultName))
155          ((DoubleValue)this[TestMeanErrorResultName].Value).Value = value;
156      }
157    }
158    #endregion
159    #endregion
160
161    [StorableConstructor]
162    protected RegressionSolutionBase(StorableConstructorFlag _) : base(_) { }
163    protected RegressionSolutionBase(RegressionSolutionBase original, Cloner cloner)
164      : base(original, cloner) {
165    }
166    protected RegressionSolutionBase(IRegressionModel model, IRegressionProblemData problemData)
167      : base(model, problemData) {
168      Add(new Result(TrainingMeanSquaredErrorResultName, TrainingMeanSquaredErrorResultDescription, new DoubleValue()));
169      Add(new Result(TestMeanSquaredErrorResultName, TestMeanSquaredErrorResultDescription, new DoubleValue()));
170      Add(new Result(TrainingMeanAbsoluteErrorResultName, TrainingMeanAbsoluteErrorResultDescription, new DoubleValue()));
171      Add(new Result(TestMeanAbsoluteErrorResultName, TestMeanAbsoluteErrorResultDescription, new DoubleValue()));
172      Add(new Result(TrainingSquaredCorrelationResultName, TrainingSquaredCorrelationResultDescription, new DoubleValue()));
173      Add(new Result(TestSquaredCorrelationResultName, TestSquaredCorrelationResultDescription, new DoubleValue()));
174      Add(new Result(TrainingRelativeErrorResultName, TrainingRelativeErrorResultDescription, new PercentValue()));
175      Add(new Result(TestRelativeErrorResultName, TestRelativeErrorResultDescription, new PercentValue()));
176      Add(new Result(TrainingNormalizedMeanSquaredErrorResultName, TrainingNormalizedMeanSquaredErrorResultDescription, new DoubleValue()));
177      Add(new Result(TestNormalizedMeanSquaredErrorResultName, TestNormalizedMeanSquaredErrorResultDescription, new DoubleValue()));
178      Add(new Result(TrainingRootMeanSquaredErrorResultName, TrainingRootMeanSquaredErrorResultDescription, new DoubleValue()));
179      Add(new Result(TestRootMeanSquaredErrorResultName, TestRootMeanSquaredErrorResultDescription, new DoubleValue()));
180    }
181
182    [StorableHook(HookType.AfterDeserialization)]
183    private void AfterDeserialization() {
184      if (string.IsNullOrEmpty(Model.TargetVariable))
185        Model.TargetVariable = this.ProblemData.TargetVariable;
186
187      // BackwardsCompatibility3.4
188      #region Backwards compatible code, remove with 3.5
189      if (!ContainsKey(TrainingMeanAbsoluteErrorResultName)) {
190        OnlineCalculatorError errorState;
191        Add(new Result(TrainingMeanAbsoluteErrorResultName, "Mean of absolute errors of the model on the training partition", new DoubleValue()));
192        double trainingMAE = OnlineMeanAbsoluteErrorCalculator.Calculate(EstimatedTrainingValues, ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices), out errorState);
193        TrainingMeanAbsoluteError = errorState == OnlineCalculatorError.None ? trainingMAE : double.NaN;
194      }
195
196      if (!ContainsKey(TestMeanAbsoluteErrorResultName)) {
197        OnlineCalculatorError errorState;
198        Add(new Result(TestMeanAbsoluteErrorResultName, "Mean of absolute errors of the model on the test partition", new DoubleValue()));
199        double testMAE = OnlineMeanAbsoluteErrorCalculator.Calculate(EstimatedTestValues, ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndices), out errorState);
200        TestMeanAbsoluteError = errorState == OnlineCalculatorError.None ? testMAE : double.NaN;
201      }
202
203      if (!ContainsKey(TrainingRootMeanSquaredErrorResultName)) {
204        OnlineCalculatorError errorState;
205        Add(new Result(TrainingRootMeanSquaredErrorResultName, TrainingRootMeanSquaredErrorResultDescription, new DoubleValue()));
206        double trainingMSE = OnlineMeanSquaredErrorCalculator.Calculate(EstimatedTrainingValues, ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices), out errorState);
207        TrainingRootMeanSquaredError = errorState == OnlineCalculatorError.None ? Math.Sqrt(trainingMSE) : double.NaN;
208      }
209
210      if (!ContainsKey(TestRootMeanSquaredErrorResultName)) {
211        OnlineCalculatorError errorState;
212        Add(new Result(TestRootMeanSquaredErrorResultName, TestRootMeanSquaredErrorResultDescription, new DoubleValue()));
213        double testMSE = OnlineMeanSquaredErrorCalculator.Calculate(EstimatedTestValues, ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndices), out errorState);
214        TestRootMeanSquaredError = errorState == OnlineCalculatorError.None ? Math.Sqrt(testMSE) : double.NaN;
215      }
216      #endregion
217    }
218
219    protected override void RecalculateResults() {
220      CalculateRegressionResults();
221    }
222
223    protected void CalculateRegressionResults() {
224      IEnumerable<double> estimatedTrainingValues = EstimatedTrainingValues; // cache values
225      IEnumerable<double> originalTrainingValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices);
226      IEnumerable<double> estimatedTestValues = EstimatedTestValues; // cache values
227      IEnumerable<double> originalTestValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndices);
228
229      OnlineCalculatorError errorState;
230      double trainingMSE = OnlineMeanSquaredErrorCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
231      TrainingMeanSquaredError = errorState == OnlineCalculatorError.None ? trainingMSE : double.NaN;
232      double testMSE = OnlineMeanSquaredErrorCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
233      TestMeanSquaredError = errorState == OnlineCalculatorError.None ? testMSE : double.NaN;
234
235      double trainingMAE = OnlineMeanAbsoluteErrorCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
236      TrainingMeanAbsoluteError = errorState == OnlineCalculatorError.None ? trainingMAE : double.NaN;
237      double testMAE = OnlineMeanAbsoluteErrorCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
238      TestMeanAbsoluteError = errorState == OnlineCalculatorError.None ? testMAE : double.NaN;
239
240      double trainingR = OnlinePearsonsRCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
241      TrainingRSquared = errorState == OnlineCalculatorError.None ? trainingR * trainingR : double.NaN;
242      double testR = OnlinePearsonsRCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
243      TestRSquared = errorState == OnlineCalculatorError.None ? testR * testR : double.NaN;
244
245      double trainingRelError = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
246      TrainingRelativeError = errorState == OnlineCalculatorError.None ? trainingRelError : double.NaN;
247      double testRelError = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
248      TestRelativeError = errorState == OnlineCalculatorError.None ? testRelError : double.NaN;
249
250      double trainingNMSE = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
251      TrainingNormalizedMeanSquaredError = errorState == OnlineCalculatorError.None ? trainingNMSE : double.NaN;
252      double testNMSE = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
253      TestNormalizedMeanSquaredError = errorState == OnlineCalculatorError.None ? testNMSE : double.NaN;
254
255      TrainingRootMeanSquaredError = Math.Sqrt(TrainingMeanSquaredError);
256      TestRootMeanSquaredError = Math.Sqrt(TestMeanSquaredError);
257
258      // BackwardsCompatibility3.3
259      #region Backwards compatible code, remove with 3.5
260      if (ContainsKey(TrainingMeanErrorResultName)) {
261        double trainingME = OnlineMeanErrorCalculator.Calculate(originalTrainingValues, estimatedTrainingValues, out errorState);
262        TrainingMeanError = errorState == OnlineCalculatorError.None ? trainingME : double.NaN;
263      }
264      if (ContainsKey(TestMeanErrorResultName)) {
265        double testME = OnlineMeanErrorCalculator.Calculate(originalTestValues, estimatedTestValues, out errorState);
266        TestMeanError = errorState == OnlineCalculatorError.None ? testME : double.NaN;
267      }
268      #endregion
269    }
270  }
271}
Note: See TracBrowser for help on using the repository browser.