source: branches/2839_HiveProjectManagement/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Analyzers/SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer.cs @ 16057

Last change on this file since 16057 was 16057, checked in by jkarder, 4 years ago

#2839:

File size: 11.0 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2018 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.Common;
26using HeuristicLab.Core;
27using HeuristicLab.Data;
28using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
29using HeuristicLab.Optimization;
30using HeuristicLab.Parameters;
31using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
32
33namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
34  /// <summary>
35  /// An operator that collects the Pareto-best symbolic data analysis solutions for single objective symbolic data analysis problems.
36  /// </summary>
37  [Item("SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer", "An operator that analyzes the Pareto-best symbolic data analysis solution for single objective symbolic data analysis problems.")]
38  [StorableClass]
39  public abstract class SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer<S, T, U> : SymbolicDataAnalysisSingleObjectiveValidationAnalyzer<T, U>, ISymbolicDataAnalysisBoundedOperator
40    where S : class, ISymbolicDataAnalysisSolution
41    where T : class, ISymbolicDataAnalysisSingleObjectiveEvaluator<U>
42    where U : class, IDataAnalysisProblemData {
43    private const string ValidationBestSolutionsParameterName = "Best validation solutions";
44    private const string ValidationBestSolutionQualitiesParameterName = "Best validation solution qualities";
45    private const string ComplexityParameterName = "Complexity";
46    private const string EstimationLimitsParameterName = "EstimationLimits";
47
48    public override bool EnabledByDefault {
49      get { return false; }
50    }
51
52    #region parameter properties
53    public ILookupParameter<ItemList<S>> ValidationBestSolutionsParameter {
54      get { return (ILookupParameter<ItemList<S>>)Parameters[ValidationBestSolutionsParameterName]; }
55    }
56    public ILookupParameter<ItemList<DoubleArray>> ValidationBestSolutionQualitiesParameter {
57      get { return (ILookupParameter<ItemList<DoubleArray>>)Parameters[ValidationBestSolutionQualitiesParameterName]; }
58    }
59    public IScopeTreeLookupParameter<DoubleValue> ComplexityParameter {
60      get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters[ComplexityParameterName]; }
61    }
62    public IValueLookupParameter<DoubleLimit> EstimationLimitsParameter {
63      get { return (IValueLookupParameter<DoubleLimit>)Parameters[EstimationLimitsParameterName]; }
64    }
65
66    #endregion
67    #region properties
68    public ItemList<S> ValidationBestSolutions {
69      get { return ValidationBestSolutionsParameter.ActualValue; }
70      set { ValidationBestSolutionsParameter.ActualValue = value; }
71    }
72    public ItemList<DoubleArray> ValidationBestSolutionQualities {
73      get { return ValidationBestSolutionQualitiesParameter.ActualValue; }
74      set { ValidationBestSolutionQualitiesParameter.ActualValue = value; }
75    }
76    #endregion
77
78    [StorableConstructor]
79    protected SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer(bool deserializing) : base(deserializing) { }
80    protected SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer(SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer<S, T, U> original, Cloner cloner) : base(original, cloner) { }
81    public SymbolicDataAnalysisSingleObjectiveValidationParetoBestSolutionAnalyzer()
82      : base() {
83      Parameters.Add(new LookupParameter<ItemList<S>>(ValidationBestSolutionsParameterName, "The validation best (Pareto-optimal) symbolic data analysis solutions."));
84      Parameters.Add(new LookupParameter<ItemList<DoubleArray>>(ValidationBestSolutionQualitiesParameterName, "The qualities of the validation best (Pareto-optimal) solutions."));
85      Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>(ComplexityParameterName, "The complexity of each tree."));
86      Parameters.Add(new ValueLookupParameter<DoubleLimit>(EstimationLimitsParameterName, "The lower and upper limit for the estimated values produced by the symbolic classification model."));
87    }
88
89    public override IOperation Apply() {
90      IEnumerable<int> rows = GenerateRowsToEvaluate();
91      if (!rows.Any()) return base.Apply();
92
93      #region find best tree
94      var evaluator = EvaluatorParameter.ActualValue;
95      var problemData = ProblemDataParameter.ActualValue;
96      ISymbolicExpressionTree[] tree = SymbolicExpressionTree.ToArray();
97
98      // sort is ascending and we take the first n% => order so that best solutions are smallest
99      // sort order is determined by maximization parameter
100      double[] trainingQuality;
101      if (Maximization.Value) {
102        // largest values must be sorted first
103        trainingQuality = Quality.Select(x => -x.Value).ToArray();
104      } else {
105        // smallest values must be sorted first
106        trainingQuality = Quality.Select(x => x.Value).ToArray();
107      }
108
109      int[] treeIndexes = Enumerable.Range(0, tree.Length).ToArray();
110
111      // sort trees by training qualities
112      Array.Sort(trainingQuality, treeIndexes);
113
114      // number of best training solutions to validate (at least 1)
115      int topN = (int)Math.Max(tree.Length * PercentageOfBestSolutionsParameter.ActualValue.Value, 1);
116
117      IExecutionContext childContext = (IExecutionContext)ExecutionContext.CreateChildOperation(evaluator);
118      // evaluate best n training trees on validiation set
119      var qualities = treeIndexes
120        .Select(i => tree[i])
121        .Take(topN)
122        .Select(t => evaluator.Evaluate(childContext, t, problemData, rows))
123        .ToArray();
124      #endregion
125
126      var results = ResultCollection;
127      // create empty parameter and result values
128      if (ValidationBestSolutions == null) {
129        ValidationBestSolutions = new ItemList<S>();
130        ValidationBestSolutionQualities = new ItemList<DoubleArray>();
131        results.Add(new Result(ValidationBestSolutionQualitiesParameter.Name, ValidationBestSolutionQualitiesParameter.Description, ValidationBestSolutionQualities));
132        results.Add(new Result(ValidationBestSolutionsParameter.Name, ValidationBestSolutionsParameter.Description, ValidationBestSolutions));
133      }
134
135      IList<Tuple<double, double>> validationBestQualities = ValidationBestSolutionQualities
136        .Select(x => Tuple.Create(x[0], x[1]))
137        .ToList();
138
139      #region find best trees
140      IList<int> nonDominatedIndexes = new List<int>();
141
142      List<double> complexities;
143      if (ComplexityParameter.ActualValue != null && ComplexityParameter.ActualValue.Length == trainingQuality.Length) {
144        complexities = ComplexityParameter.ActualValue.Select(x => x.Value).ToList();
145      } else {
146        complexities = tree.Select(t => (double)t.Length).ToList();
147      }
148      List<Tuple<double, double>> fitness = new List<Tuple<double, double>>();
149      for (int i = 0; i < qualities.Length; i++)
150        fitness.Add(Tuple.Create(qualities[i], complexities[treeIndexes[i]]));
151
152      var maximization = Tuple.Create(Maximization.Value, false); // complexity must be minimized
153      List<Tuple<double, double>> newNonDominatedQualities = new List<Tuple<double, double>>();
154      for (int i = 0; i < fitness.Count; i++) {
155        if (IsNonDominated(fitness[i], validationBestQualities, maximization) &&
156          IsNonDominated(fitness[i], newNonDominatedQualities, maximization) &&
157          IsNonDominated(fitness[i], fitness.Skip(i + 1), maximization)) {
158          if (!newNonDominatedQualities.Contains(fitness[i])) {
159            newNonDominatedQualities.Add(fitness[i]);
160            nonDominatedIndexes.Add(i);
161          }
162        }
163      }
164      #endregion
165
166      #region update Pareto-optimal solution archive
167      if (nonDominatedIndexes.Count > 0) {
168        ItemList<DoubleArray> nonDominatedQualities = new ItemList<DoubleArray>();
169        ItemList<S> nonDominatedSolutions = new ItemList<S>();
170        // add all new non-dominated solutions to the archive
171        foreach (var index in nonDominatedIndexes) {
172          S solution = CreateSolution(tree[treeIndexes[index]]);
173          nonDominatedSolutions.Add(solution);
174          nonDominatedQualities.Add(new DoubleArray(new double[] { fitness[index].Item1, fitness[index].Item2 }));
175        }
176        // add old non-dominated solutions only if they are not dominated by one of the new solutions
177        for (int i = 0; i < validationBestQualities.Count; i++) {
178          if (IsNonDominated(validationBestQualities[i], newNonDominatedQualities, maximization)) {
179            if (!newNonDominatedQualities.Contains(validationBestQualities[i])) {
180              nonDominatedSolutions.Add(ValidationBestSolutions[i]);
181              nonDominatedQualities.Add(ValidationBestSolutionQualities[i]);
182            }
183          }
184        }
185
186        // make sure solutions and qualities are ordered in the results
187        var orderedIndexes =
188          nonDominatedSolutions.Select((s, i) => i).OrderBy(i => nonDominatedQualities[i][0]).ToArray();
189
190        var orderedNonDominatedSolutions = new ItemList<S>();
191        var orderedNonDominatedQualities = new ItemList<DoubleArray>();
192        foreach (var i in orderedIndexes) {
193          orderedNonDominatedQualities.Add(nonDominatedQualities[i]);
194          orderedNonDominatedSolutions.Add(nonDominatedSolutions[i]);
195        }
196
197        ValidationBestSolutions = orderedNonDominatedSolutions;
198        ValidationBestSolutionQualities = orderedNonDominatedQualities;
199
200        results[ValidationBestSolutionsParameter.Name].Value = orderedNonDominatedSolutions;
201        results[ValidationBestSolutionQualitiesParameter.Name].Value = orderedNonDominatedQualities;
202      }
203      #endregion
204      return base.Apply();
205    }
206
207    protected abstract S CreateSolution(ISymbolicExpressionTree bestTree);
208
209    private bool IsNonDominated(Tuple<double, double> point, IEnumerable<Tuple<double, double>> points, Tuple<bool, bool> maximization) {
210      return !points.Any(p => IsBetterOrEqual(p.Item1, point.Item1, maximization.Item1) &&
211                             IsBetterOrEqual(p.Item2, point.Item2, maximization.Item2));
212    }
213    private bool IsBetterOrEqual(double lhs, double rhs, bool maximization) {
214      if (maximization) return lhs >= rhs;
215      else return lhs <= rhs;
216    }
217  }
218}
Note: See TracBrowser for help on using the repository browser.