[4233] | 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 |
|
---|
| 22 | using System;
|
---|
| 23 | using System.Linq;
|
---|
| 24 | using alglib;
|
---|
| 25 | using HeuristicLab.Core;
|
---|
| 26 | using HeuristicLab.Data;
|
---|
| 27 | using HeuristicLab.Operators;
|
---|
| 28 | using HeuristicLab.Optimization;
|
---|
| 29 | using HeuristicLab.Parameters;
|
---|
| 30 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
| 31 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
| 32 | using System.Collections.Generic;
|
---|
| 33 | using HeuristicLab.Problems.DataAnalysis.Evaluators;
|
---|
[4255] | 34 | using HeuristicLab.Analysis;
|
---|
[4233] | 35 |
|
---|
| 36 | namespace HeuristicLab.Problems.DataAnalysis.Operators {
|
---|
| 37 | [Item("Covariant Parsimony Pressure", "Covariant Parsimony Pressure.")]
|
---|
| 38 | [StorableClass]
|
---|
| 39 | public class CovariantParsimonyPressure : SingleSuccessorOperator {
|
---|
| 40 | public IScopeTreeLookupParameter<SymbolicExpressionTree> SymbolicExpressionTreeParameter {
|
---|
| 41 | get { return (IScopeTreeLookupParameter<SymbolicExpressionTree>)Parameters["SymbolicExpressionTree"]; }
|
---|
| 42 | }
|
---|
| 43 | public IScopeTreeLookupParameter<DoubleValue> QualityParameter {
|
---|
| 44 | get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters["Quality"]; }
|
---|
| 45 | }
|
---|
[4255] | 46 | public IScopeTreeLookupParameter<DoubleValue> AdjustedQualityParameter {
|
---|
| 47 | get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters["AdjustedQuality"]; }
|
---|
| 48 | }
|
---|
[4233] | 49 | public ILookupParameter<BoolValue> MaximizationParameter {
|
---|
| 50 | get { return (ILookupParameter<BoolValue>)Parameters["Maximization"]; }
|
---|
| 51 | }
|
---|
| 52 | public IValueLookupParameter<DoubleValue> KParameter {
|
---|
| 53 | get { return (IValueLookupParameter<DoubleValue>)Parameters["K"]; }
|
---|
| 54 | }
|
---|
[4309] | 55 | public ILookupParameter<DoubleValue> CParameter {
|
---|
| 56 | get { return (ILookupParameter<DoubleValue>)Parameters["C"]; }
|
---|
| 57 | }
|
---|
[4255] | 58 | public ILookupParameter<IntValue> GenerationsParameter {
|
---|
| 59 | get { return (ILookupParameter<IntValue>)Parameters["Generations"]; }
|
---|
| 60 | }
|
---|
| 61 | public IValueLookupParameter<IntValue> FirstGenerationParameter {
|
---|
| 62 | get { return (IValueLookupParameter<IntValue>)Parameters["FirstGenerationParameter"]; }
|
---|
| 63 | }
|
---|
[4271] | 64 | public IValueLookupParameter<BoolValue> ApplyParsimonyPressureParameter {
|
---|
| 65 | get { return (IValueLookupParameter<BoolValue>)Parameters["ApplyParsimonyPressure"]; }
|
---|
[4255] | 66 | }
|
---|
| 67 | public ILookupParameter<DoubleValue> LengthCorrelationParameter {
|
---|
| 68 | get { return (ILookupParameter<DoubleValue>)Parameters["Correlation(Length, AdjustedFitness)"]; }
|
---|
| 69 | }
|
---|
| 70 | public ILookupParameter<DoubleValue> FitnessCorrelationParameter {
|
---|
| 71 | get { return (ILookupParameter<DoubleValue>)Parameters["Correlation(Fitness, AdjustedFitness)"]; }
|
---|
| 72 | }
|
---|
| 73 | public IValueLookupParameter<PercentValue> ComplexityAdaptionParameter {
|
---|
| 74 | get { return (IValueLookupParameter<PercentValue>)Parameters["ComplexityAdaption"]; }
|
---|
| 75 | }
|
---|
[4329] | 76 | public IValueLookupParameter<BoolValue> InvertComplexityAdaptionParameter {
|
---|
| 77 | get { return (IValueLookupParameter<BoolValue>)Parameters["InvertComplexityAdaption"]; }
|
---|
| 78 | }
|
---|
[4272] | 79 | public IValueLookupParameter<DoubleValue> MinAverageSizeParameter {
|
---|
| 80 | get { return (IValueLookupParameter<DoubleValue>)Parameters["MinAverageSize"]; }
|
---|
| 81 | }
|
---|
[4233] | 82 |
|
---|
| 83 | public CovariantParsimonyPressure(bool deserializing) : base(deserializing) { }
|
---|
| 84 | public CovariantParsimonyPressure()
|
---|
| 85 | : base() {
|
---|
| 86 | Parameters.Add(new ScopeTreeLookupParameter<SymbolicExpressionTree>("SymbolicExpressionTree"));
|
---|
| 87 | Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("Quality"));
|
---|
[4255] | 88 | Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("AdjustedQuality"));
|
---|
[4233] | 89 | Parameters.Add(new LookupParameter<BoolValue>("Maximization"));
|
---|
| 90 | Parameters.Add(new ValueLookupParameter<DoubleValue>("K", new DoubleValue(1.0)));
|
---|
[4255] | 91 | Parameters.Add(new LookupParameter<IntValue>("Generations"));
|
---|
[4309] | 92 | Parameters.Add(new ValueLookupParameter<IntValue>("FirstGenerationParameter", new IntValue(1)));
|
---|
[4271] | 93 | Parameters.Add(new ValueLookupParameter<BoolValue>("ApplyParsimonyPressure"));
|
---|
[4309] | 94 | Parameters.Add(new ValueLookupParameter<PercentValue>("ComplexityAdaption", new PercentValue(-0.01)));
|
---|
[4255] | 95 | Parameters.Add(new LookupParameter<DoubleValue>("Correlation(Length, AdjustedFitness)"));
|
---|
| 96 | Parameters.Add(new LookupParameter<DoubleValue>("Correlation(Fitness, AdjustedFitness)"));
|
---|
[4272] | 97 | Parameters.Add(new ValueLookupParameter<DoubleValue>("MinAverageSize", new DoubleValue(15)));
|
---|
[4309] | 98 | Parameters.Add(new LookupParameter<DoubleValue>("C"));
|
---|
[4329] | 99 | Parameters.Add(new ValueLookupParameter<BoolValue>("InvertComplexityAdaption"));
|
---|
[4233] | 100 | }
|
---|
| 101 |
|
---|
| 102 | [StorableHook(Persistence.Default.CompositeSerializers.Storable.HookType.AfterDeserialization)]
|
---|
| 103 | private void AfterDeserialization() {
|
---|
| 104 | if (!Parameters.ContainsKey("Maximization"))
|
---|
| 105 | Parameters.Add(new LookupParameter<BoolValue>("Maximization"));
|
---|
| 106 | if (!Parameters.ContainsKey("K"))
|
---|
| 107 | Parameters.Add(new ValueLookupParameter<DoubleValue>("K", new DoubleValue(1.0)));
|
---|
[4255] | 108 | if (!Parameters.ContainsKey("AdjustedQuality")) {
|
---|
| 109 | Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("AdjustedQuality"));
|
---|
| 110 | }
|
---|
| 111 | if (!Parameters.ContainsKey("Generations")) {
|
---|
| 112 | Parameters.Add(new LookupParameter<IntValue>("Generations"));
|
---|
| 113 | }
|
---|
| 114 | if (!Parameters.ContainsKey("FirstGenerationParameter")) {
|
---|
[4309] | 115 | Parameters.Add(new ValueLookupParameter<IntValue>("FirstGenerationParameter", new IntValue(1)));
|
---|
[4255] | 116 | }
|
---|
[4271] | 117 | if (!Parameters.ContainsKey("ApplyParsimonyPressure")) {
|
---|
| 118 | Parameters.Add(new ValueLookupParameter<BoolValue>("ApplyParsimonyPressure"));
|
---|
[4255] | 119 | }
|
---|
| 120 | if (!Parameters.ContainsKey("ComplexityAdaption")) {
|
---|
[4309] | 121 | Parameters.Add(new ValueLookupParameter<PercentValue>("ComplexityAdaption", new PercentValue(-0.01)));
|
---|
[4255] | 122 | }
|
---|
[4272] | 123 | if (!Parameters.ContainsKey("MinAverageSize")) {
|
---|
| 124 | Parameters.Add(new ValueLookupParameter<DoubleValue>("MinAverageSize", new DoubleValue(15)));
|
---|
| 125 | }
|
---|
[4309] | 126 | if (!Parameters.ContainsKey("C")) {
|
---|
| 127 | Parameters.Add(new LookupParameter<DoubleValue>("C"));
|
---|
| 128 | }
|
---|
[4329] | 129 | if (!Parameters.ContainsKey("InvertComplexityAdaption")) {
|
---|
| 130 | Parameters.Add(new ValueLookupParameter<BoolValue>("InvertComplexityAdaption"));
|
---|
| 131 | }
|
---|
[4233] | 132 | }
|
---|
| 133 |
|
---|
| 134 | public override IOperation Apply() {
|
---|
[4255] | 135 | ItemArray<SymbolicExpressionTree> trees = SymbolicExpressionTreeParameter.ActualValue;
|
---|
| 136 | ItemArray<DoubleValue> qualities = QualityParameter.ActualValue;
|
---|
[4271] | 137 | // always apply Parsimony pressure if overfitting has been detected
|
---|
[4255] | 138 | // otherwise appliy PP only when we are currently overfitting
|
---|
| 139 | if (GenerationsParameter.ActualValue != null && GenerationsParameter.ActualValue.Value >= FirstGenerationParameter.ActualValue.Value &&
|
---|
[4271] | 140 | ApplyParsimonyPressureParameter.ActualValue.Value == true) {
|
---|
[4255] | 141 | var lengths = from tree in trees
|
---|
| 142 | select tree.Size;
|
---|
| 143 | double k = KParameter.ActualValue.Value;
|
---|
[4233] | 144 |
|
---|
[4255] | 145 | // calculate cov(f, l) and cov(l, l^k)
|
---|
| 146 | OnlineCovarianceEvaluator lengthFitnessCovEvaluator = new OnlineCovarianceEvaluator();
|
---|
| 147 | OnlineCovarianceEvaluator lengthAdjLengthCovEvaluator = new OnlineCovarianceEvaluator();
|
---|
| 148 | OnlineMeanAndVarianceCalculator lengthMeanCalculator = new OnlineMeanAndVarianceCalculator();
|
---|
| 149 | OnlineMeanAndVarianceCalculator fitnessMeanCalculator = new OnlineMeanAndVarianceCalculator();
|
---|
| 150 | OnlineMeanAndVarianceCalculator adjLengthMeanCalculator = new OnlineMeanAndVarianceCalculator();
|
---|
| 151 | var lengthEnumerator = lengths.GetEnumerator();
|
---|
| 152 | var qualityEnumerator = qualities.GetEnumerator();
|
---|
| 153 | while (lengthEnumerator.MoveNext() & qualityEnumerator.MoveNext()) {
|
---|
| 154 | double fitness = qualityEnumerator.Current.Value;
|
---|
| 155 | if (!MaximizationParameter.ActualValue.Value) {
|
---|
| 156 | // use f = 1 / (1 + quality) for minimization problems
|
---|
| 157 | fitness = 1.0 / (1.0 + fitness);
|
---|
| 158 | }
|
---|
| 159 | lengthFitnessCovEvaluator.Add(lengthEnumerator.Current, fitness);
|
---|
| 160 | lengthAdjLengthCovEvaluator.Add(lengthEnumerator.Current, Math.Pow(lengthEnumerator.Current, k));
|
---|
| 161 | lengthMeanCalculator.Add(lengthEnumerator.Current);
|
---|
| 162 | fitnessMeanCalculator.Add(fitness);
|
---|
| 163 | adjLengthMeanCalculator.Add(Math.Pow(lengthEnumerator.Current, k));
|
---|
[4233] | 164 | }
|
---|
| 165 |
|
---|
[4329] | 166 | //double sizeAdaption = lengthMeanCalculator.Mean * ComplexityAdaptionParameter.ActualValue.Value;
|
---|
| 167 | double sizeAdaption = 100.0 * ComplexityAdaptionParameter.ActualValue.Value;
|
---|
| 168 | if (InvertComplexityAdaptionParameter.ActualValue != null && InvertComplexityAdaptionParameter.ActualValue.Value) {
|
---|
| 169 | sizeAdaption = -sizeAdaption;
|
---|
| 170 | }
|
---|
[4309] | 171 | if (lengthMeanCalculator.Mean + sizeAdaption < MinAverageSizeParameter.ActualValue.Value)
|
---|
[4350] | 172 | sizeAdaption = MinAverageSizeParameter.ActualValue.Value - lengthMeanCalculator.Mean;
|
---|
[4233] | 173 |
|
---|
[4255] | 174 | // cov(l, f) - (g(t+1) - mu(t)) avgF
|
---|
| 175 | // c(t) = --------------------------------------------
|
---|
| 176 | // cov(l, l^k) - (g(t+1) - mu(t)) E[l^k]
|
---|
[4309] | 177 | double c = lengthFitnessCovEvaluator.Covariance - sizeAdaption * fitnessMeanCalculator.Mean;
|
---|
| 178 | c /= lengthAdjLengthCovEvaluator.Covariance - sizeAdaption * adjLengthMeanCalculator.Mean;
|
---|
[4233] | 179 |
|
---|
[4309] | 180 | CParameter.ActualValue = new DoubleValue(c);
|
---|
| 181 |
|
---|
[4255] | 182 | // adjust fitness
|
---|
| 183 | bool maximization = MaximizationParameter.ActualValue.Value;
|
---|
| 184 |
|
---|
| 185 | lengthEnumerator = lengths.GetEnumerator();
|
---|
| 186 | qualityEnumerator = qualities.GetEnumerator();
|
---|
| 187 | int i = 0;
|
---|
| 188 | ItemArray<DoubleValue> adjQualities = new ItemArray<DoubleValue>(qualities.Length);
|
---|
| 189 |
|
---|
| 190 | while (lengthEnumerator.MoveNext() & qualityEnumerator.MoveNext()) {
|
---|
| 191 | adjQualities[i++] = new DoubleValue(qualityEnumerator.Current.Value - c * Math.Pow(lengthEnumerator.Current, k));
|
---|
| 192 | }
|
---|
| 193 | AdjustedQualityParameter.ActualValue = adjQualities;
|
---|
| 194 | double[] lengthArr = lengths.Select(x => (double)x).ToArray<double>();
|
---|
| 195 |
|
---|
| 196 | double[] adjFitess = (from f in AdjustedQualityParameter.ActualValue
|
---|
| 197 | select f.Value).ToArray<double>();
|
---|
| 198 | double[] fitnessArr = (from f in QualityParameter.ActualValue
|
---|
| 199 | let normFit = maximization ? f.Value : 1.0 / (1.0 + f.Value)
|
---|
| 200 | select normFit).ToArray<double>();
|
---|
| 201 |
|
---|
| 202 | LengthCorrelationParameter.ActualValue = new DoubleValue(alglib.correlation.spearmanrankcorrelation(lengthArr, adjFitess, lengthArr.Length));
|
---|
| 203 | FitnessCorrelationParameter.ActualValue = new DoubleValue(alglib.correlation.spearmanrankcorrelation(fitnessArr, adjFitess, lengthArr.Length));
|
---|
| 204 |
|
---|
| 205 | } else {
|
---|
[4309] | 206 | CParameter.ActualValue = new DoubleValue(0.0);
|
---|
[4255] | 207 | // adjusted fitness is equal to fitness
|
---|
| 208 | AdjustedQualityParameter.ActualValue = (ItemArray<DoubleValue>)QualityParameter.ActualValue.Clone();
|
---|
| 209 | FitnessCorrelationParameter.ActualValue = new DoubleValue(1.0);
|
---|
| 210 |
|
---|
| 211 | double[] lengths = (from tree in trees
|
---|
| 212 | select (double)tree.Size).ToArray<double>();
|
---|
| 213 |
|
---|
| 214 | double[] fitess = (from f in AdjustedQualityParameter.ActualValue
|
---|
| 215 | select f.Value).ToArray<double>();
|
---|
| 216 |
|
---|
| 217 | LengthCorrelationParameter.ActualValue = new DoubleValue(alglib.correlation.spearmanrankcorrelation(lengths, fitess, lengths.Length));
|
---|
[4233] | 218 | }
|
---|
| 219 | return base.Apply();
|
---|
| 220 | }
|
---|
| 221 | }
|
---|
| 222 | }
|
---|