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 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
33 |
|
---|
34 | namespace HeuristicLab.Problems.DataAnalysis.Regression.Symbolic.Analyzers {
|
---|
35 | public class SymbolicRegressionTournamentPruning : SingleSuccessorOperator, ISymbolicRegressionAnalyzer {
|
---|
36 | private const string RandomParameterName = "Random";
|
---|
37 | private const string SymbolicExpressionTreeParameterName = "SymbolicExpressionTree";
|
---|
38 | private const string DataAnalysisProblemDataParameterName = "DataAnalysisProblemData";
|
---|
39 | private const string SamplesStartParameterName = "SamplesStart";
|
---|
40 | private const string SamplesEndParameterName = "SamplesEnd";
|
---|
41 | private const string EvaluatorParameterName = "Evaluator";
|
---|
42 | private const string MaximizationParameterName = "Maximization";
|
---|
43 | private const string SymbolicExpressionTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
|
---|
44 | private const string UpperEstimationLimitParameterName = "UpperEstimationLimit";
|
---|
45 | private const string LowerEstimationLimitParameterName = "LowerEstimationLimit";
|
---|
46 | private const string MaxPruningRatioParameterName = "MaxPruningRatio";
|
---|
47 | private const string TournamentSizeParameterName = "TournamentSize";
|
---|
48 | private const string PopulationPercentileStartParameterName = "PopulationPercentileStart";
|
---|
49 | private const string PopulationPercentileEndParameterName = "PopulationPercentileEnd";
|
---|
50 | private const string QualityGainWeightParameterName = "QualityGainWeight";
|
---|
51 | private const string IterationsParameterName = "Iterations";
|
---|
52 | private const string FirstPruningGenerationParameterName = "FirstPruningGeneration";
|
---|
53 | private const string PruningFrequencyParameterName = "PruningFrequency";
|
---|
54 | private const string GenerationParameterName = "Generations";
|
---|
55 | private const string ResultsParameterName = "Results";
|
---|
56 |
|
---|
57 | #region parameter properties
|
---|
58 | public ILookupParameter<IRandom> RandomParameter {
|
---|
59 | get { return (ILookupParameter<IRandom>)Parameters[RandomParameterName]; }
|
---|
60 | }
|
---|
61 | public ScopeTreeLookupParameter<SymbolicExpressionTree> SymbolicExpressionTreeParameter {
|
---|
62 | get { return (ScopeTreeLookupParameter<SymbolicExpressionTree>)Parameters[SymbolicExpressionTreeParameterName]; }
|
---|
63 | }
|
---|
64 | public ILookupParameter<DataAnalysisProblemData> DataAnalysisProblemDataParameter {
|
---|
65 | get { return (ILookupParameter<DataAnalysisProblemData>)Parameters[DataAnalysisProblemDataParameterName]; }
|
---|
66 | }
|
---|
67 | public ILookupParameter<ISymbolicExpressionTreeInterpreter> SymbolicExpressionTreeInterpreterParameter {
|
---|
68 | get { return (ILookupParameter<ISymbolicExpressionTreeInterpreter>)Parameters[SymbolicExpressionTreeInterpreterParameterName]; }
|
---|
69 | }
|
---|
70 | public IValueLookupParameter<DoubleValue> UpperEstimationLimitParameter {
|
---|
71 | get { return (IValueLookupParameter<DoubleValue>)Parameters[UpperEstimationLimitParameterName]; }
|
---|
72 | }
|
---|
73 | public IValueLookupParameter<DoubleValue> LowerEstimationLimitParameter {
|
---|
74 | get { return (IValueLookupParameter<DoubleValue>)Parameters[LowerEstimationLimitParameterName]; }
|
---|
75 | }
|
---|
76 | public IValueLookupParameter<IntValue> SamplesStartParameter {
|
---|
77 | get { return (IValueLookupParameter<IntValue>)Parameters[SamplesStartParameterName]; }
|
---|
78 | }
|
---|
79 | public IValueLookupParameter<IntValue> SamplesEndParameter {
|
---|
80 | get { return (IValueLookupParameter<IntValue>)Parameters[SamplesEndParameterName]; }
|
---|
81 | }
|
---|
82 | public ILookupParameter<ISymbolicRegressionEvaluator> EvaluatorParameter {
|
---|
83 | get { return (ILookupParameter<ISymbolicRegressionEvaluator>)Parameters[EvaluatorParameterName]; }
|
---|
84 | }
|
---|
85 | public ILookupParameter<BoolValue> MaximizationParameter {
|
---|
86 | get { return (ILookupParameter<BoolValue>)Parameters[MaximizationParameterName]; }
|
---|
87 | }
|
---|
88 | public IValueLookupParameter<DoubleValue> MaxPruningRatioParameter {
|
---|
89 | get { return (IValueLookupParameter<DoubleValue>)Parameters[MaxPruningRatioParameterName]; }
|
---|
90 | }
|
---|
91 | public IValueLookupParameter<IntValue> TournamentSizeParameter {
|
---|
92 | get { return (IValueLookupParameter<IntValue>)Parameters[TournamentSizeParameterName]; }
|
---|
93 | }
|
---|
94 | public IValueLookupParameter<DoubleValue> PopulationPercentileStartParameter {
|
---|
95 | get { return (IValueLookupParameter<DoubleValue>)Parameters[PopulationPercentileStartParameterName]; }
|
---|
96 | }
|
---|
97 | public IValueLookupParameter<DoubleValue> PopulationPercentileEndParameter {
|
---|
98 | get { return (IValueLookupParameter<DoubleValue>)Parameters[PopulationPercentileEndParameterName]; }
|
---|
99 | }
|
---|
100 | public IValueLookupParameter<DoubleValue> QualityGainWeightParameter {
|
---|
101 | get { return (IValueLookupParameter<DoubleValue>)Parameters[QualityGainWeightParameterName]; }
|
---|
102 | }
|
---|
103 | public IValueLookupParameter<IntValue> IterationsParameter {
|
---|
104 | get { return (IValueLookupParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
105 | }
|
---|
106 | public IValueLookupParameter<IntValue> FirstPruningGenerationParameter {
|
---|
107 | get { return (IValueLookupParameter<IntValue>)Parameters[FirstPruningGenerationParameterName]; }
|
---|
108 | }
|
---|
109 | public IValueLookupParameter<IntValue> PruningFrequencyParameter {
|
---|
110 | get { return (IValueLookupParameter<IntValue>)Parameters[PruningFrequencyParameterName]; }
|
---|
111 | }
|
---|
112 | public ILookupParameter<IntValue> GenerationParameter {
|
---|
113 | get { return (ILookupParameter<IntValue>)Parameters[GenerationParameterName]; }
|
---|
114 | }
|
---|
115 | public ILookupParameter<ResultCollection> ResultsParameter {
|
---|
116 | get { return (ILookupParameter<ResultCollection>)Parameters[ResultsParameterName]; }
|
---|
117 | }
|
---|
118 | #endregion
|
---|
119 | #region properties
|
---|
120 | public IRandom Random {
|
---|
121 | get { return RandomParameter.ActualValue; }
|
---|
122 | }
|
---|
123 | public ItemArray<SymbolicExpressionTree> SymbolicExpressionTree {
|
---|
124 | get { return SymbolicExpressionTreeParameter.ActualValue; }
|
---|
125 | }
|
---|
126 | public DataAnalysisProblemData DataAnalysisProblemData {
|
---|
127 | get { return DataAnalysisProblemDataParameter.ActualValue; }
|
---|
128 | }
|
---|
129 | public ISymbolicExpressionTreeInterpreter SymbolicExpressionTreeInterpreter {
|
---|
130 | get { return SymbolicExpressionTreeInterpreterParameter.ActualValue; }
|
---|
131 | }
|
---|
132 | public DoubleValue UpperEstimationLimit {
|
---|
133 | get { return UpperEstimationLimitParameter.ActualValue; }
|
---|
134 | }
|
---|
135 | public DoubleValue LowerEstimationLimit {
|
---|
136 | get { return LowerEstimationLimitParameter.ActualValue; }
|
---|
137 | }
|
---|
138 | public IntValue SamplesStart {
|
---|
139 | get { return SamplesStartParameter.ActualValue; }
|
---|
140 | }
|
---|
141 | public IntValue SamplesEnd {
|
---|
142 | get { return SamplesEndParameter.ActualValue; }
|
---|
143 | }
|
---|
144 | public ISymbolicRegressionEvaluator Evaluator {
|
---|
145 | get { return EvaluatorParameter.ActualValue; }
|
---|
146 | }
|
---|
147 | public BoolValue Maximization {
|
---|
148 | get { return MaximizationParameter.ActualValue; }
|
---|
149 | }
|
---|
150 | public DoubleValue MaxPruningRatio {
|
---|
151 | get { return MaxPruningRatioParameter.ActualValue; }
|
---|
152 | }
|
---|
153 | public IntValue TournamentSize {
|
---|
154 | get { return TournamentSizeParameter.ActualValue; }
|
---|
155 | }
|
---|
156 | public DoubleValue PopulationPercentileStart {
|
---|
157 | get { return PopulationPercentileStartParameter.ActualValue; }
|
---|
158 | }
|
---|
159 | public DoubleValue PopulationPercentileEnd {
|
---|
160 | get { return PopulationPercentileEndParameter.ActualValue; }
|
---|
161 | }
|
---|
162 | public DoubleValue QualityGainWeight {
|
---|
163 | get { return QualityGainWeightParameter.ActualValue; }
|
---|
164 | }
|
---|
165 | public IntValue Iterations {
|
---|
166 | get { return IterationsParameter.ActualValue; }
|
---|
167 | }
|
---|
168 | public IntValue PruningFrequency {
|
---|
169 | get { return PruningFrequencyParameter.ActualValue; }
|
---|
170 | }
|
---|
171 | public IntValue FirstPruningGeneration {
|
---|
172 | get { return FirstPruningGenerationParameter.ActualValue; }
|
---|
173 | }
|
---|
174 | public IntValue Generation {
|
---|
175 | get { return GenerationParameter.ActualValue; }
|
---|
176 | }
|
---|
177 | #endregion
|
---|
178 | protected SymbolicRegressionTournamentPruning(bool deserializing) : base(deserializing) { }
|
---|
179 | public SymbolicRegressionTournamentPruning()
|
---|
180 | : base() {
|
---|
181 | Parameters.Add(new LookupParameter<IRandom>(RandomParameterName, "A random number generator."));
|
---|
182 | Parameters.Add(new ScopeTreeLookupParameter<SymbolicExpressionTree>(SymbolicExpressionTreeParameterName, "The symbolic expression trees to prune."));
|
---|
183 | Parameters.Add(new LookupParameter<DataAnalysisProblemData>(DataAnalysisProblemDataParameterName, "The data analysis problem data to use for branch impact evaluation."));
|
---|
184 | Parameters.Add(new LookupParameter<ISymbolicExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, "The interpreter to use for node impact evaluation"));
|
---|
185 | Parameters.Add(new ValueLookupParameter<IntValue>(SamplesStartParameterName, "The first row index of the dataset partition to use for branch impact evaluation."));
|
---|
186 | Parameters.Add(new ValueLookupParameter<IntValue>(SamplesEndParameterName, "The last row index of the dataset partition to use for branch impact evaluation."));
|
---|
187 | Parameters.Add(new LookupParameter<ISymbolicRegressionEvaluator>(EvaluatorParameterName, "The evaluator that should be used to determine which branches are not relevant."));
|
---|
188 | Parameters.Add(new LookupParameter<BoolValue>(MaximizationParameterName, "The direction of optimization."));
|
---|
189 | Parameters.Add(new ValueLookupParameter<DoubleValue>(MaxPruningRatioParameterName, "The maximal relative size of the pruned branch.", new DoubleValue(0.5)));
|
---|
190 | Parameters.Add(new ValueLookupParameter<IntValue>(TournamentSizeParameterName, "The number of branches to compare for pruning", new IntValue(10)));
|
---|
191 | Parameters.Add(new ValueLookupParameter<DoubleValue>(PopulationPercentileStartParameterName, "The start of the population percentile to consider for pruning.", new DoubleValue(0.25)));
|
---|
192 | Parameters.Add(new ValueLookupParameter<DoubleValue>(PopulationPercentileEndParameterName, "The end of the population percentile to consider for pruning.", new DoubleValue(0.75)));
|
---|
193 | Parameters.Add(new ValueLookupParameter<DoubleValue>(QualityGainWeightParameterName, "The weight of the quality gain relative to the size gain.", new DoubleValue(1.0)));
|
---|
194 | Parameters.Add(new ValueLookupParameter<DoubleValue>(UpperEstimationLimitParameterName, "The upper estimation limit to use for evaluation."));
|
---|
195 | Parameters.Add(new ValueLookupParameter<DoubleValue>(LowerEstimationLimitParameterName, "The lower estimation limit to use for evaluation."));
|
---|
196 | Parameters.Add(new ValueLookupParameter<IntValue>(IterationsParameterName, "The number of pruning iterations to apply for each tree.", new IntValue(1)));
|
---|
197 | Parameters.Add(new ValueLookupParameter<IntValue>(FirstPruningGenerationParameterName, "The first generation when pruning should be applied.", new IntValue(1)));
|
---|
198 | Parameters.Add(new ValueLookupParameter<IntValue>(PruningFrequencyParameterName, "The frequency of pruning operations (1: every generation, 2: every second generation...)", new IntValue(1)));
|
---|
199 | Parameters.Add(new LookupParameter<IntValue>(GenerationParameterName, "The current generation."));
|
---|
200 | Parameters.Add(new LookupParameter<ResultCollection>(ResultsParameterName, "The results collection."));
|
---|
201 | }
|
---|
202 |
|
---|
203 | [StorableHook(HookType.AfterDeserialization)]
|
---|
204 | private void AfterDeserialization() {
|
---|
205 | #region compatibility remove before releasing 3.3.1
|
---|
206 | if (!Parameters.ContainsKey(EvaluatorParameterName)) {
|
---|
207 | Parameters.Add(new LookupParameter<ISymbolicRegressionEvaluator>(EvaluatorParameterName, "The evaluator which should be used to evaluate the solution on the validation set."));
|
---|
208 | }
|
---|
209 | if (!Parameters.ContainsKey(MaximizationParameterName)) {
|
---|
210 | Parameters.Add(new LookupParameter<BoolValue>(MaximizationParameterName, "The direction of optimization."));
|
---|
211 | }
|
---|
212 | #endregion
|
---|
213 | }
|
---|
214 |
|
---|
215 | public override IOperation Apply() {
|
---|
216 | bool pruningCondition =
|
---|
217 | (Generation.Value >= FirstPruningGeneration.Value) &&
|
---|
218 | ((Generation.Value - FirstPruningGeneration.Value) % PruningFrequency.Value == 0);
|
---|
219 | if (pruningCondition) {
|
---|
220 | int n = SymbolicExpressionTree.Length;
|
---|
221 | double percentileStart = PopulationPercentileStart.Value;
|
---|
222 | double percentileEnd = PopulationPercentileEnd.Value;
|
---|
223 | // for each tree in the given percentile
|
---|
224 | var trees = SymbolicExpressionTree
|
---|
225 | .Skip((int)(n * percentileStart))
|
---|
226 | .Take((int)(n * (percentileEnd - percentileStart)));
|
---|
227 | foreach (var tree in trees) {
|
---|
228 | Prune(Random, tree, Iterations.Value, TournamentSize.Value,
|
---|
229 | DataAnalysisProblemData, SamplesStart.Value, SamplesEnd.Value,
|
---|
230 | SymbolicExpressionTreeInterpreter, Evaluator, Maximization.Value,
|
---|
231 | LowerEstimationLimit.Value, UpperEstimationLimit.Value,
|
---|
232 | MaxPruningRatio.Value, QualityGainWeight.Value);
|
---|
233 | }
|
---|
234 | }
|
---|
235 | return base.Apply();
|
---|
236 | }
|
---|
237 |
|
---|
238 | public static void Prune(IRandom random, SymbolicExpressionTree tree, int iterations, int tournamentSize,
|
---|
239 | DataAnalysisProblemData problemData, int samplesStart, int samplesEnd,
|
---|
240 | ISymbolicExpressionTreeInterpreter interpreter, ISymbolicRegressionEvaluator evaluator, bool maximization,
|
---|
241 | double lowerEstimationLimit, double upperEstimationLimit,
|
---|
242 | double maxPruningRatio, double qualityGainWeight) {
|
---|
243 | IEnumerable<int> rows = Enumerable.Range(samplesStart, samplesEnd - samplesStart);
|
---|
244 | int originalSize = tree.Size;
|
---|
245 | double originalQuality = evaluator.Evaluate(interpreter, tree,
|
---|
246 | lowerEstimationLimit, upperEstimationLimit, problemData.Dataset, problemData.TargetVariable.Value, rows);
|
---|
247 |
|
---|
248 | int minPrunedSize = (int)(originalSize * (1 - maxPruningRatio));
|
---|
249 |
|
---|
250 | // tree for branch evaluation
|
---|
251 | SymbolicExpressionTree templateTree = (SymbolicExpressionTree)tree.Clone();
|
---|
252 | while (templateTree.Root.SubTrees[0].SubTrees.Count > 0) templateTree.Root.SubTrees[0].RemoveSubTree(0);
|
---|
253 |
|
---|
254 | SymbolicExpressionTree prunedTree = tree;
|
---|
255 | for (int iteration = 0; iteration < iterations; iteration++) {
|
---|
256 | SymbolicExpressionTree iterationBestTree = prunedTree;
|
---|
257 | double bestGain = double.PositiveInfinity;
|
---|
258 | int maxPrunedBranchSize = (int)(prunedTree.Size * maxPruningRatio);
|
---|
259 |
|
---|
260 | for (int i = 0; i < tournamentSize; i++) {
|
---|
261 | var clonedTree = (SymbolicExpressionTree)prunedTree.Clone();
|
---|
262 | int clonedTreeSize = clonedTree.Size;
|
---|
263 | var prunePoints = (from node in clonedTree.IterateNodesPostfix()
|
---|
264 | from subTree in node.SubTrees
|
---|
265 | let subTreeSize = subTree.GetSize()
|
---|
266 | where subTreeSize <= maxPrunedBranchSize
|
---|
267 | where clonedTreeSize - subTreeSize >= minPrunedSize
|
---|
268 | select new { Parent = node, Branch = subTree, SubTreeIndex = node.SubTrees.IndexOf(subTree) })
|
---|
269 | .ToList();
|
---|
270 | if (prunePoints.Count > 0) {
|
---|
271 | var selectedPrunePoint = prunePoints.SelectRandom(random);
|
---|
272 | templateTree.Root.SubTrees[0].AddSubTree(selectedPrunePoint.Branch);
|
---|
273 | IEnumerable<double> branchValues = interpreter.GetSymbolicExpressionTreeValues(templateTree, problemData.Dataset, rows);
|
---|
274 | double branchMean = branchValues.Average();
|
---|
275 | templateTree.Root.SubTrees[0].RemoveSubTree(0);
|
---|
276 |
|
---|
277 | selectedPrunePoint.Parent.RemoveSubTree(selectedPrunePoint.SubTreeIndex);
|
---|
278 | var constNode = CreateConstant(branchMean);
|
---|
279 | selectedPrunePoint.Parent.InsertSubTree(selectedPrunePoint.SubTreeIndex, constNode);
|
---|
280 |
|
---|
281 | double prunedQuality = evaluator.Evaluate(interpreter, clonedTree,
|
---|
282 | lowerEstimationLimit, upperEstimationLimit, problemData.Dataset, problemData.TargetVariable.Value, Enumerable.Range(samplesStart, samplesEnd - samplesStart));
|
---|
283 | double prunedSize = clonedTree.Size;
|
---|
284 | // deteriation in quality:
|
---|
285 | // exp: MSE : newMse < origMse (improvement) => prefer the larger improvement
|
---|
286 | // MSE : newMse > origMse (deteriation) => prefer the smaller deteriation
|
---|
287 | // MSE : minimize: newMse / origMse
|
---|
288 | // R² : newR² > origR² (improvment) => prefer the larger improvment
|
---|
289 | // R² : newR² < origR² (deteriation) => prefer smaller deteriation
|
---|
290 | // R² : minimize: origR² / newR²
|
---|
291 | double qualityDeteriation = maximization ? originalQuality / prunedQuality : prunedQuality / originalQuality;
|
---|
292 | // size of the pruned tree is always smaller than the size of the original tree
|
---|
293 | // same change in quality => prefer pruning operation that removes a larger tree
|
---|
294 | double gain = (qualityDeteriation * qualityGainWeight) /
|
---|
295 | (originalSize / prunedSize);
|
---|
296 | if (gain < bestGain) {
|
---|
297 | bestGain = gain;
|
---|
298 | iterationBestTree = clonedTree;
|
---|
299 | }
|
---|
300 | }
|
---|
301 | }
|
---|
302 | prunedTree = iterationBestTree;
|
---|
303 | }
|
---|
304 | tree.Root = prunedTree.Root;
|
---|
305 | }
|
---|
306 |
|
---|
307 | private static SymbolicExpressionTreeNode CreateConstant(double constantValue) {
|
---|
308 | var node = (ConstantTreeNode)(new Constant()).CreateTreeNode();
|
---|
309 | node.Value = constantValue;
|
---|
310 | return node;
|
---|
311 | }
|
---|
312 | }
|
---|
313 | }
|
---|