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.Collections.Generic;
|
---|
23 | using System.Linq;
|
---|
24 | using HeuristicLab.Core;
|
---|
25 | using HeuristicLab.Data;
|
---|
26 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
27 | using HeuristicLab.Operators;
|
---|
28 | using HeuristicLab.Optimization;
|
---|
29 | using HeuristicLab.Parameters;
|
---|
30 | using HeuristicLab.Problems.DataAnalysis.Symbolic;
|
---|
31 | using HeuristicLab.Problems.DataAnalysis.Symbolic.Symbols;
|
---|
32 |
|
---|
33 | namespace HeuristicLab.Problems.DataAnalysis.Regression.Symbolic.Analyzers {
|
---|
34 | public class SymbolicRegressionTournamentPruning : SingleSuccessorOperator, ISymbolicRegressionAnalyzer {
|
---|
35 | private const string RandomParameterName = "Random";
|
---|
36 | private const string SymbolicExpressionTreeParameterName = "SymbolicExpressionTree";
|
---|
37 | private const string DataAnalysisProblemDataParameterName = "DataAnalysisProblemData";
|
---|
38 | private const string SamplesStartParameterName = "SamplesStart";
|
---|
39 | private const string SamplesEndParameterName = "SamplesEnd";
|
---|
40 | private const string SymbolicExpressionTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
|
---|
41 | private const string UpperEstimationLimitParameterName = "UpperEstimationLimit";
|
---|
42 | private const string LowerEstimationLimitParameterName = "LowerEstimationLimit";
|
---|
43 | private const string MaxPruningRatioParameterName = "MaxPruningRatio";
|
---|
44 | private const string TournamentSizeParameterName = "TournamentSize";
|
---|
45 | private const string PopulationPercentileStartParameterName = "PopulationPercentileStart";
|
---|
46 | private const string PopulationPercentileEndParameterName = "PopulationPercentileEnd";
|
---|
47 | private const string QualityGainWeightParameterName = "QualityGainWeight";
|
---|
48 | private const string IterationsParameterName = "Iterations";
|
---|
49 | private const string FirstPruningGenerationParameterName = "FirstPruningGeneration";
|
---|
50 | private const string PruningFrequencyParameterName = "PruningFrequency";
|
---|
51 | private const string GenerationParameterName = "Generations";
|
---|
52 | private const string ResultsParameterName = "Results";
|
---|
53 |
|
---|
54 | #region parameter properties
|
---|
55 | public ILookupParameter<IRandom> RandomParameter {
|
---|
56 | get { return (ILookupParameter<IRandom>)Parameters[RandomParameterName]; }
|
---|
57 | }
|
---|
58 | public ScopeTreeLookupParameter<SymbolicExpressionTree> SymbolicExpressionTreeParameter {
|
---|
59 | get { return (ScopeTreeLookupParameter<SymbolicExpressionTree>)Parameters[SymbolicExpressionTreeParameterName]; }
|
---|
60 | }
|
---|
61 | public ILookupParameter<DataAnalysisProblemData> DataAnalysisProblemDataParameter {
|
---|
62 | get { return (ILookupParameter<DataAnalysisProblemData>)Parameters[DataAnalysisProblemDataParameterName]; }
|
---|
63 | }
|
---|
64 | public ILookupParameter<ISymbolicExpressionTreeInterpreter> SymbolicExpressionTreeInterpreterParameter {
|
---|
65 | get { return (ILookupParameter<ISymbolicExpressionTreeInterpreter>)Parameters[SymbolicExpressionTreeInterpreterParameterName]; }
|
---|
66 | }
|
---|
67 | public IValueLookupParameter<DoubleValue> UpperEstimationLimitParameter {
|
---|
68 | get { return (IValueLookupParameter<DoubleValue>)Parameters[UpperEstimationLimitParameterName]; }
|
---|
69 | }
|
---|
70 | public IValueLookupParameter<DoubleValue> LowerEstimationLimitParameter {
|
---|
71 | get { return (IValueLookupParameter<DoubleValue>)Parameters[LowerEstimationLimitParameterName]; }
|
---|
72 | }
|
---|
73 | public IValueLookupParameter<IntValue> SamplesStartParameter {
|
---|
74 | get { return (IValueLookupParameter<IntValue>)Parameters[SamplesStartParameterName]; }
|
---|
75 | }
|
---|
76 | public IValueLookupParameter<IntValue> SamplesEndParameter {
|
---|
77 | get { return (IValueLookupParameter<IntValue>)Parameters[SamplesEndParameterName]; }
|
---|
78 | }
|
---|
79 | public IValueLookupParameter<DoubleValue> MaxPruningRatioParameter {
|
---|
80 | get { return (IValueLookupParameter<DoubleValue>)Parameters[MaxPruningRatioParameterName]; }
|
---|
81 | }
|
---|
82 | public IValueLookupParameter<IntValue> TournamentSizeParameter {
|
---|
83 | get { return (IValueLookupParameter<IntValue>)Parameters[TournamentSizeParameterName]; }
|
---|
84 | }
|
---|
85 | public IValueLookupParameter<DoubleValue> PopulationPercentileStartParameter {
|
---|
86 | get { return (IValueLookupParameter<DoubleValue>)Parameters[PopulationPercentileStartParameterName]; }
|
---|
87 | }
|
---|
88 | public IValueLookupParameter<DoubleValue> PopulationPercentileEndParameter {
|
---|
89 | get { return (IValueLookupParameter<DoubleValue>)Parameters[PopulationPercentileEndParameterName]; }
|
---|
90 | }
|
---|
91 | public IValueLookupParameter<DoubleValue> QualityGainWeightParameter {
|
---|
92 | get { return (IValueLookupParameter<DoubleValue>)Parameters[QualityGainWeightParameterName]; }
|
---|
93 | }
|
---|
94 | public IValueLookupParameter<IntValue> IterationsParameter {
|
---|
95 | get { return (IValueLookupParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
96 | }
|
---|
97 | public IValueLookupParameter<IntValue> FirstPruningGenerationParameter {
|
---|
98 | get { return (IValueLookupParameter<IntValue>)Parameters[FirstPruningGenerationParameterName]; }
|
---|
99 | }
|
---|
100 | public IValueLookupParameter<IntValue> PruningFrequencyParameter {
|
---|
101 | get { return (IValueLookupParameter<IntValue>)Parameters[PruningFrequencyParameterName]; }
|
---|
102 | }
|
---|
103 | public ILookupParameter<IntValue> GenerationParameter {
|
---|
104 | get { return (ILookupParameter<IntValue>)Parameters[GenerationParameterName]; }
|
---|
105 | }
|
---|
106 | public ILookupParameter<ResultCollection> ResultsParameter {
|
---|
107 | get { return (ILookupParameter<ResultCollection>)Parameters[ResultsParameterName]; }
|
---|
108 | }
|
---|
109 | #endregion
|
---|
110 | #region properties
|
---|
111 | public IRandom Random {
|
---|
112 | get { return RandomParameter.ActualValue; }
|
---|
113 | }
|
---|
114 | public ItemArray<SymbolicExpressionTree> SymbolicExpressionTree {
|
---|
115 | get { return SymbolicExpressionTreeParameter.ActualValue; }
|
---|
116 | }
|
---|
117 | public DataAnalysisProblemData DataAnalysisProblemData {
|
---|
118 | get { return DataAnalysisProblemDataParameter.ActualValue; }
|
---|
119 | }
|
---|
120 | public ISymbolicExpressionTreeInterpreter SymbolicExpressionTreeInterpreter {
|
---|
121 | get { return SymbolicExpressionTreeInterpreterParameter.ActualValue; }
|
---|
122 | }
|
---|
123 | public DoubleValue UpperEstimationLimit {
|
---|
124 | get { return UpperEstimationLimitParameter.ActualValue; }
|
---|
125 | }
|
---|
126 | public DoubleValue LowerEstimationLimit {
|
---|
127 | get { return LowerEstimationLimitParameter.ActualValue; }
|
---|
128 | }
|
---|
129 | public IntValue SamplesStart {
|
---|
130 | get { return SamplesStartParameter.ActualValue; }
|
---|
131 | }
|
---|
132 | public IntValue SamplesEnd {
|
---|
133 | get { return SamplesEndParameter.ActualValue; }
|
---|
134 | }
|
---|
135 | public DoubleValue MaxPruningRatio {
|
---|
136 | get { return MaxPruningRatioParameter.ActualValue; }
|
---|
137 | }
|
---|
138 | public IntValue TournamentSize {
|
---|
139 | get { return TournamentSizeParameter.ActualValue; }
|
---|
140 | }
|
---|
141 | public DoubleValue PopulationPercentileStart {
|
---|
142 | get { return PopulationPercentileStartParameter.ActualValue; }
|
---|
143 | }
|
---|
144 | public DoubleValue PopulationPercentileEnd {
|
---|
145 | get { return PopulationPercentileEndParameter.ActualValue; }
|
---|
146 | }
|
---|
147 | public DoubleValue QualityGainWeight {
|
---|
148 | get { return QualityGainWeightParameter.ActualValue; }
|
---|
149 | }
|
---|
150 | public IntValue Iterations {
|
---|
151 | get { return IterationsParameter.ActualValue; }
|
---|
152 | }
|
---|
153 | public IntValue PruningFrequency {
|
---|
154 | get { return PruningFrequencyParameter.ActualValue; }
|
---|
155 | }
|
---|
156 | public IntValue FirstPruningGeneration {
|
---|
157 | get { return FirstPruningGenerationParameter.ActualValue; }
|
---|
158 | }
|
---|
159 | public IntValue Generation {
|
---|
160 | get { return GenerationParameter.ActualValue; }
|
---|
161 | }
|
---|
162 | #endregion
|
---|
163 | public SymbolicRegressionTournamentPruning()
|
---|
164 | : base() {
|
---|
165 | Parameters.Add(new LookupParameter<IRandom>(RandomParameterName, "A random number generator."));
|
---|
166 | Parameters.Add(new ScopeTreeLookupParameter<SymbolicExpressionTree>(SymbolicExpressionTreeParameterName, "The symbolic expression trees to prune."));
|
---|
167 | Parameters.Add(new LookupParameter<DataAnalysisProblemData>(DataAnalysisProblemDataParameterName, "The data analysis problem data to use for branch impact evaluation."));
|
---|
168 | Parameters.Add(new LookupParameter<ISymbolicExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, "The interpreter to use for node impact evaluation"));
|
---|
169 | Parameters.Add(new ValueLookupParameter<IntValue>(SamplesStartParameterName, "The first row index of the dataset partition to use for branch impact evaluation."));
|
---|
170 | Parameters.Add(new ValueLookupParameter<IntValue>(SamplesEndParameterName, "The last row index of the dataset partition to use for branch impact evaluation."));
|
---|
171 | Parameters.Add(new ValueLookupParameter<DoubleValue>(MaxPruningRatioParameterName, "The maximal relative size of the pruned branch.", new DoubleValue(0.5)));
|
---|
172 | Parameters.Add(new ValueLookupParameter<IntValue>(TournamentSizeParameterName, "The number of branches to compare for pruning", new IntValue(10)));
|
---|
173 | Parameters.Add(new ValueLookupParameter<DoubleValue>(PopulationPercentileStartParameterName, "The start of the population percentile to consider for pruning.", new DoubleValue(0.25)));
|
---|
174 | Parameters.Add(new ValueLookupParameter<DoubleValue>(PopulationPercentileEndParameterName, "The end of the population percentile to consider for pruning.", new DoubleValue(0.75)));
|
---|
175 | Parameters.Add(new ValueLookupParameter<DoubleValue>(QualityGainWeightParameterName, "The weight of the quality gain relative to the size gain.", new DoubleValue(1.0)));
|
---|
176 | Parameters.Add(new ValueLookupParameter<DoubleValue>(UpperEstimationLimitParameterName, "The upper estimation limit to use for evaluation."));
|
---|
177 | Parameters.Add(new ValueLookupParameter<DoubleValue>(LowerEstimationLimitParameterName, "The lower estimation limit to use for evaluation."));
|
---|
178 | Parameters.Add(new ValueLookupParameter<IntValue>(IterationsParameterName, "The number of pruning iterations to apply for each tree.", new IntValue(1)));
|
---|
179 | Parameters.Add(new ValueLookupParameter<IntValue>(FirstPruningGenerationParameterName, "The first generation when pruning should be applied.", new IntValue(1)));
|
---|
180 | Parameters.Add(new ValueLookupParameter<IntValue>(PruningFrequencyParameterName, "The frequency of pruning operations (1: every generation, 2: every second generation...)", new IntValue(1)));
|
---|
181 | Parameters.Add(new LookupParameter<IntValue>(GenerationParameterName, "The current generation."));
|
---|
182 | Parameters.Add(new LookupParameter<ResultCollection>(ResultsParameterName, "The results collection."));
|
---|
183 | }
|
---|
184 |
|
---|
185 | public override IOperation Apply() {
|
---|
186 | bool pruningCondition =
|
---|
187 | (Generation.Value >= FirstPruningGeneration.Value) &&
|
---|
188 | ((Generation.Value - FirstPruningGeneration.Value) % PruningFrequency.Value == 0);
|
---|
189 | if (pruningCondition) {
|
---|
190 | int n = SymbolicExpressionTree.Length;
|
---|
191 | double percentileStart = PopulationPercentileStart.Value;
|
---|
192 | double percentileEnd = PopulationPercentileEnd.Value;
|
---|
193 | // for each tree in the given percentile
|
---|
194 | var trees = SymbolicExpressionTree
|
---|
195 | .Skip((int)(n * percentileStart))
|
---|
196 | .Take((int)(n * (percentileEnd - percentileStart)));
|
---|
197 | foreach (var tree in trees) {
|
---|
198 | Prune(Random, tree, Iterations.Value, TournamentSize.Value,
|
---|
199 | DataAnalysisProblemData, SamplesStart.Value, SamplesEnd.Value,
|
---|
200 | SymbolicExpressionTreeInterpreter,
|
---|
201 | LowerEstimationLimit.Value, UpperEstimationLimit.Value,
|
---|
202 | MaxPruningRatio.Value, QualityGainWeight.Value);
|
---|
203 | }
|
---|
204 | }
|
---|
205 | return base.Apply();
|
---|
206 | }
|
---|
207 |
|
---|
208 | public static void Prune(IRandom random, SymbolicExpressionTree tree, int iterations, int tournamentSize,
|
---|
209 | DataAnalysisProblemData problemData, int samplesStart, int samplesEnd,
|
---|
210 | ISymbolicExpressionTreeInterpreter interpreter,
|
---|
211 | double lowerEstimationLimit, double upperEstimationLimit,
|
---|
212 | double maxPruningRatio, double qualityGainWeight) {
|
---|
213 | IEnumerable<int> rows = Enumerable.Range(samplesStart, samplesEnd - samplesStart);
|
---|
214 | int originalSize = tree.Size;
|
---|
215 | double originalMse = SymbolicRegressionScaledMeanSquaredErrorEvaluator.Calculate(interpreter, tree,
|
---|
216 | lowerEstimationLimit, upperEstimationLimit, problemData.Dataset, problemData.TargetVariable.Value, Enumerable.Range(samplesStart, samplesEnd - samplesStart));
|
---|
217 |
|
---|
218 | int minPrunedSize = (int)(originalSize * (1 - maxPruningRatio));
|
---|
219 |
|
---|
220 | // tree for branch evaluation
|
---|
221 | SymbolicExpressionTree templateTree = (SymbolicExpressionTree)tree.Clone();
|
---|
222 | while (templateTree.Root.SubTrees[0].SubTrees.Count > 0) templateTree.Root.SubTrees[0].RemoveSubTree(0);
|
---|
223 |
|
---|
224 | SymbolicExpressionTree prunedTree = tree;
|
---|
225 | for (int iteration = 0; iteration < iterations; iteration++) {
|
---|
226 | SymbolicExpressionTree iterationBestTree = prunedTree;
|
---|
227 | double bestGain = double.PositiveInfinity;
|
---|
228 | int maxPrunedBranchSize = (int)(prunedTree.Size * maxPruningRatio);
|
---|
229 |
|
---|
230 | for (int i = 0; i < tournamentSize; i++) {
|
---|
231 | var clonedTree = (SymbolicExpressionTree)prunedTree.Clone();
|
---|
232 | int clonedTreeSize = clonedTree.Size;
|
---|
233 | var prunePoints = (from node in clonedTree.IterateNodesPostfix()
|
---|
234 | from subTree in node.SubTrees
|
---|
235 | let subTreeSize = subTree.GetSize()
|
---|
236 | where subTreeSize <= maxPrunedBranchSize
|
---|
237 | where clonedTreeSize - subTreeSize >= minPrunedSize
|
---|
238 | select new { Parent = node, Branch = subTree, SubTreeIndex = node.SubTrees.IndexOf(subTree) })
|
---|
239 | .ToList();
|
---|
240 | if (prunePoints.Count > 0) {
|
---|
241 | var selectedPrunePoint = prunePoints.SelectRandom(random);
|
---|
242 | templateTree.Root.SubTrees[0].AddSubTree(selectedPrunePoint.Branch);
|
---|
243 | IEnumerable<double> branchValues = interpreter.GetSymbolicExpressionTreeValues(templateTree, problemData.Dataset, rows);
|
---|
244 | double branchMean = branchValues.Average();
|
---|
245 | templateTree.Root.SubTrees[0].RemoveSubTree(0);
|
---|
246 |
|
---|
247 | selectedPrunePoint.Parent.RemoveSubTree(selectedPrunePoint.SubTreeIndex);
|
---|
248 | var constNode = CreateConstant(branchMean);
|
---|
249 | selectedPrunePoint.Parent.InsertSubTree(selectedPrunePoint.SubTreeIndex, constNode);
|
---|
250 |
|
---|
251 | double prunedMse = SymbolicRegressionScaledMeanSquaredErrorEvaluator.Calculate(interpreter, clonedTree,
|
---|
252 | lowerEstimationLimit, upperEstimationLimit, problemData.Dataset, problemData.TargetVariable.Value, Enumerable.Range(samplesStart, samplesEnd - samplesStart));
|
---|
253 | double prunedSize = clonedTree.Size;
|
---|
254 | // MSE of the pruned tree is larger than the original tree in most cases
|
---|
255 | // size of the pruned tree is always smaller than the size of the original tree
|
---|
256 | // same change in quality => prefer pruning operation that removes a larger tree
|
---|
257 | double gain = ((prunedMse / originalMse) * qualityGainWeight) /
|
---|
258 | (originalSize / prunedSize);
|
---|
259 | if (gain < bestGain) {
|
---|
260 | bestGain = gain;
|
---|
261 | iterationBestTree = clonedTree;
|
---|
262 | }
|
---|
263 | }
|
---|
264 | }
|
---|
265 | prunedTree = iterationBestTree;
|
---|
266 | }
|
---|
267 | tree.Root = prunedTree.Root;
|
---|
268 | }
|
---|
269 |
|
---|
270 | private static SymbolicExpressionTreeNode CreateConstant(double constantValue) {
|
---|
271 | var node = (ConstantTreeNode)(new Constant()).CreateTreeNode();
|
---|
272 | node.Value = constantValue;
|
---|
273 | return node;
|
---|
274 | }
|
---|
275 | }
|
---|
276 | }
|
---|