Free cookie consent management tool by TermsFeed Policy Generator

source: trunk/sources/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/Analyzers/SymbolicRegressionModelQualityAnalyzer.cs @ 4468

Last change on this file since 4468 was 4468, checked in by mkommend, 14 years ago

Preparation for cross validation - removed the test samples from the trainining samples and added ValidationPercentage parameter (ticket #1199).

File size: 15.3 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2010 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 HeuristicLab.Analysis;
26using HeuristicLab.Core;
27using HeuristicLab.Data;
28using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
29using HeuristicLab.Operators;
30using HeuristicLab.Optimization;
31using HeuristicLab.Parameters;
32using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
33using HeuristicLab.Problems.DataAnalysis.Evaluators;
34using HeuristicLab.Problems.DataAnalysis.Symbolic;
35
36namespace HeuristicLab.Problems.DataAnalysis.Regression.Symbolic.Analyzers {
37  /// <summary>
38  /// "An operator for analyzing the quality of symbolic regression solutions symbolic expression tree encoding."
39  /// </summary>
40  [Item("SymbolicRegressionModelQualityAnalyzer", "An operator for analyzing the quality of symbolic regression solutions symbolic expression tree encoding.")]
41  [StorableClass]
42  public sealed class SymbolicRegressionModelQualityAnalyzer : SingleSuccessorOperator, ISymbolicRegressionAnalyzer {
43    private const string SymbolicExpressionTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
44    private const string SymbolicExpressionTreeParameterName = "SymbolicExpressionTree";
45    private const string ProblemDataParameterName = "ProblemData";
46    private const string ResultsParameterName = "Results";
47
48    private const string TrainingMeanSquaredErrorQualityParameterName = "Mean squared error (training)";
49    private const string MinTrainingMeanSquaredErrorQualityParameterName = "Min mean squared error (training)";
50    private const string MaxTrainingMeanSquaredErrorQualityParameterName = "Max mean squared error (training)";
51    private const string AverageTrainingMeanSquaredErrorQualityParameterName = "Average mean squared error (training)";
52    private const string BestTrainingMeanSquaredErrorQualityParameterName = "Best mean squared error (training)";
53
54    private const string TrainingAverageRelativeErrorQualityParameterName = "Average relative error (training)";
55    private const string MinTrainingAverageRelativeErrorQualityParameterName = "Min average relative error (training)";
56    private const string MaxTrainingAverageRelativeErrorQualityParameterName = "Max average relative error (training)";
57    private const string AverageTrainingAverageRelativeErrorQualityParameterName = "Average average relative error (training)";
58    private const string BestTrainingAverageRelativeErrorQualityParameterName = "Best average relative error (training)";
59
60    private const string TrainingRSquaredQualityParameterName = "R² (training)";
61    private const string MinTrainingRSquaredQualityParameterName = "Min R² (training)";
62    private const string MaxTrainingRSquaredQualityParameterName = "Max R² (training)";
63    private const string AverageTrainingRSquaredQualityParameterName = "Average R² (training)";
64    private const string BestTrainingRSquaredQualityParameterName = "Best R² (training)";
65
66    private const string TestMeanSquaredErrorQualityParameterName = "Mean squared error (test)";
67    private const string MinTestMeanSquaredErrorQualityParameterName = "Min mean squared error (test)";
68    private const string MaxTestMeanSquaredErrorQualityParameterName = "Max mean squared error (test)";
69    private const string AverageTestMeanSquaredErrorQualityParameterName = "Average mean squared error (test)";
70    private const string BestTestMeanSquaredErrorQualityParameterName = "Best mean squared error (test)";
71
72    private const string TestAverageRelativeErrorQualityParameterName = "Average relative error (test)";
73    private const string MinTestAverageRelativeErrorQualityParameterName = "Min average relative error (test)";
74    private const string MaxTestAverageRelativeErrorQualityParameterName = "Max average relative error (test)";
75    private const string AverageTestAverageRelativeErrorQualityParameterName = "Average average relative error (test)";
76    private const string BestTestAverageRelativeErrorQualityParameterName = "Best average relative error (test)";
77
78    private const string TestRSquaredQualityParameterName = "R² (test)";
79    private const string MinTestRSquaredQualityParameterName = "Min R² (test)";
80    private const string MaxTestRSquaredQualityParameterName = "Max R² (test)";
81    private const string AverageTestRSquaredQualityParameterName = "Average R² (test)";
82    private const string BestTestRSquaredQualityParameterName = "Best R² (test)";
83
84    private const string RSquaredValuesParameterName = "R²";
85    private const string MeanSquaredErrorValuesParameterName = "Mean squared error";
86    private const string RelativeErrorValuesParameterName = "Average relative error";
87
88    private const string UpperEstimationLimitParameterName = "UpperEstimationLimit";
89    private const string LowerEstimationLimitParameterName = "LowerEstimationLimit";
90
91    #region parameter properties
92    public ScopeTreeLookupParameter<SymbolicExpressionTree> SymbolicExpressionTreeParameter {
93      get { return (ScopeTreeLookupParameter<SymbolicExpressionTree>)Parameters[SymbolicExpressionTreeParameterName]; }
94    }
95    public IValueLookupParameter<ISymbolicExpressionTreeInterpreter> SymbolicExpressionTreeInterpreterParameter {
96      get { return (IValueLookupParameter<ISymbolicExpressionTreeInterpreter>)Parameters[SymbolicExpressionTreeInterpreterParameterName]; }
97    }
98    public IValueLookupParameter<DataAnalysisProblemData> ProblemDataParameter {
99      get { return (IValueLookupParameter<DataAnalysisProblemData>)Parameters[ProblemDataParameterName]; }
100    }
101    public IValueLookupParameter<DoubleValue> UpperEstimationLimitParameter {
102      get { return (IValueLookupParameter<DoubleValue>)Parameters[UpperEstimationLimitParameterName]; }
103    }
104    public IValueLookupParameter<DoubleValue> LowerEstimationLimitParameter {
105      get { return (IValueLookupParameter<DoubleValue>)Parameters[LowerEstimationLimitParameterName]; }
106    }
107    public ILookupParameter<ResultCollection> ResultsParameter {
108      get { return (ILookupParameter<ResultCollection>)Parameters[ResultsParameterName]; }
109    }
110    #endregion
111    #region properties
112    public DoubleValue UpperEstimationLimit {
113      get { return UpperEstimationLimitParameter.ActualValue; }
114    }
115    public DoubleValue LowerEstimationLimit {
116      get { return LowerEstimationLimitParameter.ActualValue; }
117    }
118    #endregion
119
120    public SymbolicRegressionModelQualityAnalyzer()
121      : base() {
122      Parameters.Add(new ScopeTreeLookupParameter<SymbolicExpressionTree>(SymbolicExpressionTreeParameterName, "The symbolic expression trees to analyze."));
123      Parameters.Add(new ValueLookupParameter<ISymbolicExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, "The interpreter that should be used to calculate the output values of the symbolic expression tree."));
124      Parameters.Add(new ValueLookupParameter<DataAnalysisProblemData>(ProblemDataParameterName, "The problem data containing the input varaibles for the symbolic regression problem."));
125      Parameters.Add(new ValueLookupParameter<DoubleValue>(UpperEstimationLimitParameterName, "The upper limit that should be used as cut off value for the output values of symbolic expression trees."));
126      Parameters.Add(new ValueLookupParameter<DoubleValue>(LowerEstimationLimitParameterName, "The lower limit that should be used as cut off value for the output values of symbolic expression trees."));
127      Parameters.Add(new ValueLookupParameter<DataTable>(MeanSquaredErrorValuesParameterName, "The data table to collect mean squared error values."));
128      Parameters.Add(new ValueLookupParameter<DataTable>(RSquaredValuesParameterName, "The data table to collect R² correlation coefficient values."));
129      Parameters.Add(new ValueLookupParameter<DataTable>(RelativeErrorValuesParameterName, "The data table to collect relative error values."));
130      Parameters.Add(new LookupParameter<ResultCollection>(ResultsParameterName, "The result collection where the best symbolic regression solution should be stored."));
131    }
132
133    [StorableConstructor]
134    private SymbolicRegressionModelQualityAnalyzer(bool deserializing) : base() { }
135
136    public override IOperation Apply() {
137      Analyze(SymbolicExpressionTreeParameter.ActualValue, SymbolicExpressionTreeInterpreterParameter.ActualValue,
138        UpperEstimationLimit.Value, LowerEstimationLimit.Value, ProblemDataParameter.ActualValue,
139        ResultsParameter.ActualValue);
140      return base.Apply();
141    }
142
143    public static void Analyze(IEnumerable<SymbolicExpressionTree> trees, ISymbolicExpressionTreeInterpreter interpreter,
144      double upperEstimationLimit, double lowerEstimationLimit,
145      DataAnalysisProblemData problemData, ResultCollection results) {
146      int targetVariableIndex = problemData.Dataset.GetVariableIndex(problemData.TargetVariable.Value);
147      IEnumerable<double> originalTrainingValues = problemData.Dataset.GetEnumeratedVariableValues(targetVariableIndex, problemData.TrainingIndizes);
148      IEnumerable<double> originalTestValues = problemData.Dataset.GetEnumeratedVariableValues(targetVariableIndex, problemData.TestIndizes);
149      List<double> trainingMse = new List<double>();
150      List<double> trainingR2 = new List<double>();
151      List<double> trainingRelErr = new List<double>();
152      List<double> testMse = new List<double>();
153      List<double> testR2 = new List<double>();
154      List<double> testRelErr = new List<double>();
155
156      OnlineMeanSquaredErrorEvaluator mseEvaluator = new OnlineMeanSquaredErrorEvaluator();
157      OnlineMeanAbsolutePercentageErrorEvaluator relErrEvaluator = new OnlineMeanAbsolutePercentageErrorEvaluator();
158      OnlinePearsonsRSquaredEvaluator r2Evaluator = new OnlinePearsonsRSquaredEvaluator();
159
160      foreach (var tree in trees) {
161        #region training
162        var estimatedTrainingValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, problemData.TrainingIndizes);
163        mseEvaluator.Reset();
164        r2Evaluator.Reset();
165        relErrEvaluator.Reset();
166        var estimatedEnumerator = estimatedTrainingValues.GetEnumerator();
167        var originalEnumerator = originalTrainingValues.GetEnumerator();
168        while (estimatedEnumerator.MoveNext() & originalEnumerator.MoveNext()) {
169          double estimated = estimatedEnumerator.Current;
170          if (double.IsNaN(estimated)) estimated = upperEstimationLimit;
171          else estimated = Math.Min(upperEstimationLimit, Math.Max(lowerEstimationLimit, estimated));
172          mseEvaluator.Add(originalEnumerator.Current, estimated);
173          r2Evaluator.Add(originalEnumerator.Current, estimated);
174          relErrEvaluator.Add(originalEnumerator.Current, estimated);
175        }
176        if (estimatedEnumerator.MoveNext() || originalEnumerator.MoveNext()) {
177          throw new InvalidOperationException("Number of elements in estimated and original enumeration doesn't match.");
178        }
179        trainingMse.Add(mseEvaluator.MeanSquaredError);
180        trainingR2.Add(r2Evaluator.RSquared);
181        trainingRelErr.Add(relErrEvaluator.MeanAbsolutePercentageError);
182        #endregion
183        #region test
184        var estimatedTestValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, problemData.TestIndizes);
185
186        mseEvaluator.Reset();
187        r2Evaluator.Reset();
188        relErrEvaluator.Reset();
189        estimatedEnumerator = estimatedTestValues.GetEnumerator();
190        originalEnumerator = originalTestValues.GetEnumerator();
191        while (estimatedEnumerator.MoveNext() & originalEnumerator.MoveNext()) {
192          double estimated = estimatedEnumerator.Current;
193          if (double.IsNaN(estimated)) estimated = upperEstimationLimit;
194          else estimated = Math.Min(upperEstimationLimit, Math.Max(lowerEstimationLimit, estimated));
195          mseEvaluator.Add(originalEnumerator.Current, estimated);
196          r2Evaluator.Add(originalEnumerator.Current, estimated);
197          relErrEvaluator.Add(originalEnumerator.Current, estimated);
198        }
199        if (estimatedEnumerator.MoveNext() || originalEnumerator.MoveNext()) {
200          throw new InvalidOperationException("Number of elements in estimated and original enumeration doesn't match.");
201        }
202        testMse.Add(mseEvaluator.MeanSquaredError);
203        testR2.Add(r2Evaluator.RSquared);
204        testRelErr.Add(relErrEvaluator.MeanAbsolutePercentageError);
205        #endregion
206      }
207
208      AddResultTableValues(results, MeanSquaredErrorValuesParameterName, "mean squared error (training)", trainingMse.Min(), trainingMse.Average(), trainingMse.Max());
209      AddResultTableValues(results, MeanSquaredErrorValuesParameterName, "mean squared error (test)", testMse.Min(), testMse.Average(), testMse.Max());
210      AddResultTableValues(results, RelativeErrorValuesParameterName, "mean relative error (training)", trainingRelErr.Min(), trainingRelErr.Average(), trainingRelErr.Max());
211      AddResultTableValues(results, RelativeErrorValuesParameterName, "mean relative error (test)", testRelErr.Min(), testRelErr.Average(), testRelErr.Max());
212      AddResultTableValues(results, RSquaredValuesParameterName, "Pearson's R² (training)", trainingR2.Min(), trainingR2.Average(), trainingR2.Max());
213      AddResultTableValues(results, RSquaredValuesParameterName, "Pearson's R² (test)", testR2.Min(), testR2.Average(), testR2.Max());
214    }
215
216    private static void AddResultTableValues(ResultCollection results, string tableName, string valueName, double minValue, double avgValue, double maxValue) {
217      if (!results.ContainsKey(tableName)) {
218        results.Add(new Result(tableName, new DataTable(tableName)));
219      }
220      DataTable table = (DataTable)results[tableName].Value;
221      AddValue(table, minValue, "Min. " + valueName, string.Empty);
222      AddValue(table, avgValue, "Avg. " + valueName, string.Empty);
223      AddValue(table, maxValue, "Max. " + valueName, string.Empty);
224    }
225
226    private static void AddValue(DataTable table, double data, string name, string description) {
227      DataRow row;
228      table.Rows.TryGetValue(name, out row);
229      if (row == null) {
230        row = new DataRow(name, description);
231        row.Values.Add(data);
232        table.Rows.Add(row);
233      } else {
234        row.Values.Add(data);
235      }
236    }
237
238
239    private static void SetResultValue(ResultCollection results, string name, double value) {
240      if (results.ContainsKey(name))
241        results[name].Value = new DoubleValue(value);
242      else
243        results.Add(new Result(name, new DoubleValue(value)));
244    }
245  }
246}
Note: See TracBrowser for help on using the repository browser.