[645] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
[5445] | 3 | * Copyright (C) 2002-2011 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[645] | 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 |
|
---|
[3462] | 22 | using System;
|
---|
[4068] | 23 | using System.Collections.Generic;
|
---|
[3462] | 24 | using System.Linq;
|
---|
[4722] | 25 | using HeuristicLab.Common;
|
---|
[645] | 26 | using HeuristicLab.Core;
|
---|
| 27 | using HeuristicLab.Data;
|
---|
[5521] | 28 | using HeuristicLab.Parameters;
|
---|
[4068] | 29 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
[645] | 30 |
|
---|
[5499] | 31 | namespace HeuristicLab.Encodings.SymbolicExpressionTreeEncoding {
|
---|
[3223] | 32 | [StorableClass]
|
---|
[5549] | 33 | [Item("ProbabilisticTreeCreator", "An operator that creates new symbolic expression trees with uniformly distributed length")]
|
---|
[5618] | 34 | public class ProbabilisticTreeCreator : SymbolicExpressionTreeCreator,
|
---|
[5686] | 35 | ISymbolicExpressionTreeSizeConstraintOperator, ISymbolicExpressionTreeGrammarBasedOperator {
|
---|
[3369] | 36 | private const int MAX_TRIES = 100;
|
---|
[5499] | 37 | private const string MaximumSymbolicExpressionTreeLengthParameterName = "MaximumSymbolicExpressionTreeLength";
|
---|
| 38 | private const string MaximumSymbolicExpressionTreeDepthParameterName = "MaximumSymbolicExpressionTreeDepth";
|
---|
| 39 | private const string SymbolicExpressionTreeGrammarParameterName = "SymbolicExpressionTreeGrammar";
|
---|
[6233] | 40 | private const string ClonedSymbolicExpressionTreeGrammarParameterName = "ClonedSymbolicExpressionTreeGrammar";
|
---|
[5499] | 41 | #region Parameter Properties
|
---|
| 42 | public IValueLookupParameter<IntValue> MaximumSymbolicExpressionTreeLengthParameter {
|
---|
| 43 | get { return (IValueLookupParameter<IntValue>)Parameters[MaximumSymbolicExpressionTreeLengthParameterName]; }
|
---|
| 44 | }
|
---|
| 45 | public IValueLookupParameter<IntValue> MaximumSymbolicExpressionTreeDepthParameter {
|
---|
| 46 | get { return (IValueLookupParameter<IntValue>)Parameters[MaximumSymbolicExpressionTreeDepthParameterName]; }
|
---|
| 47 | }
|
---|
[5686] | 48 | public IValueLookupParameter<ISymbolicExpressionGrammar> SymbolicExpressionTreeGrammarParameter {
|
---|
| 49 | get { return (IValueLookupParameter<ISymbolicExpressionGrammar>)Parameters[SymbolicExpressionTreeGrammarParameterName]; }
|
---|
[5499] | 50 | }
|
---|
[6233] | 51 | public ILookupParameter<ISymbolicExpressionGrammar> ClonedSymbolicExpressionTreeGrammarParameter {
|
---|
| 52 | get { return (ILookupParameter<ISymbolicExpressionGrammar>)Parameters[ClonedSymbolicExpressionTreeGrammarParameterName]; }
|
---|
| 53 | }
|
---|
[5499] | 54 | #endregion
|
---|
| 55 | #region Properties
|
---|
| 56 | public IntValue MaximumSymbolicExpressionTreeLength {
|
---|
| 57 | get { return MaximumSymbolicExpressionTreeLengthParameter.ActualValue; }
|
---|
| 58 | }
|
---|
| 59 | public IntValue MaximumSymbolicExpressionTreeDepth {
|
---|
| 60 | get { return MaximumSymbolicExpressionTreeDepthParameter.ActualValue; }
|
---|
| 61 | }
|
---|
[5686] | 62 | public ISymbolicExpressionGrammar SymbolicExpressionTreeGrammar {
|
---|
[6233] | 63 | get { return ClonedSymbolicExpressionTreeGrammarParameter.ActualValue; }
|
---|
[5499] | 64 | }
|
---|
| 65 | #endregion
|
---|
| 66 |
|
---|
[4722] | 67 | [StorableConstructor]
|
---|
[5618] | 68 | protected ProbabilisticTreeCreator(bool deserializing) : base(deserializing) { }
|
---|
| 69 | protected ProbabilisticTreeCreator(ProbabilisticTreeCreator original, Cloner cloner) : base(original, cloner) { }
|
---|
[5499] | 70 | public ProbabilisticTreeCreator()
|
---|
| 71 | : base() {
|
---|
| 72 | Parameters.Add(new ValueLookupParameter<IntValue>(MaximumSymbolicExpressionTreeLengthParameterName, "The maximal length (number of nodes) of the symbolic expression tree."));
|
---|
| 73 | Parameters.Add(new ValueLookupParameter<IntValue>(MaximumSymbolicExpressionTreeDepthParameterName, "The maximal depth of the symbolic expression tree (a tree with one node has depth = 0)."));
|
---|
[5686] | 74 | Parameters.Add(new ValueLookupParameter<ISymbolicExpressionGrammar>(SymbolicExpressionTreeGrammarParameterName, "The tree grammar that defines the correct syntax of symbolic expression trees that should be created."));
|
---|
[6233] | 75 | Parameters.Add(new LookupParameter<ISymbolicExpressionGrammar>(ClonedSymbolicExpressionTreeGrammarParameterName, "An immutable clone of the concrete grammar that is actually used to create and manipulate trees."));
|
---|
[5499] | 76 | }
|
---|
[3338] | 77 |
|
---|
[4722] | 78 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
| 79 | return new ProbabilisticTreeCreator(this, cloner);
|
---|
[645] | 80 | }
|
---|
[6233] | 81 | [StorableHook(HookType.AfterDeserialization)]
|
---|
| 82 | private void AfterDeserialization() {
|
---|
| 83 | if (!Parameters.ContainsKey(ClonedSymbolicExpressionTreeGrammarParameterName))
|
---|
| 84 | Parameters.Add(new LookupParameter<ISymbolicExpressionGrammar>(ClonedSymbolicExpressionTreeGrammarParameterName, "An immutable clone of the concrete grammar that is actually used to create and manipulate trees."));
|
---|
| 85 | }
|
---|
[645] | 86 |
|
---|
[6233] | 87 | public override IOperation Apply() {
|
---|
| 88 | if (ClonedSymbolicExpressionTreeGrammarParameter.ActualValue == null) {
|
---|
| 89 | SymbolicExpressionTreeGrammarParameter.ActualValue.ReadOnly = true;
|
---|
| 90 | IScope globalScope = ExecutionContext.Scope;
|
---|
| 91 | while (globalScope.Parent != null)
|
---|
| 92 | globalScope = globalScope.Parent;
|
---|
| 93 |
|
---|
| 94 | globalScope.Variables.Add(new Variable(ClonedSymbolicExpressionTreeGrammarParameterName, (ISymbolicExpressionGrammar)SymbolicExpressionTreeGrammarParameter.ActualValue.Clone()));
|
---|
| 95 | }
|
---|
| 96 | return base.Apply();
|
---|
| 97 | }
|
---|
| 98 |
|
---|
[5510] | 99 | protected override ISymbolicExpressionTree Create(IRandom random) {
|
---|
[5686] | 100 | return Create(random, SymbolicExpressionTreeGrammar, MaximumSymbolicExpressionTreeLength.Value, MaximumSymbolicExpressionTreeDepth.Value);
|
---|
[2447] | 101 | }
|
---|
| 102 |
|
---|
[5686] | 103 | public static ISymbolicExpressionTree Create(IRandom random, ISymbolicExpressionGrammar grammar,
|
---|
| 104 | int maxTreeLength, int maxTreeDepth) {
|
---|
[3223] | 105 | SymbolicExpressionTree tree = new SymbolicExpressionTree();
|
---|
[5686] | 106 | var rootNode = (SymbolicExpressionTreeTopLevelNode)grammar.ProgramRootSymbol.CreateTreeNode();
|
---|
[3442] | 107 | if (rootNode.HasLocalParameters) rootNode.ResetLocalParameters(random);
|
---|
[4249] | 108 | rootNode.SetGrammar(new SymbolicExpressionTreeGrammar(grammar));
|
---|
[5686] | 109 | var startNode = (SymbolicExpressionTreeTopLevelNode)grammar.StartSymbol.CreateTreeNode();
|
---|
| 110 | startNode.SetGrammar(new SymbolicExpressionTreeGrammar(grammar));
|
---|
| 111 | if (startNode.HasLocalParameters) startNode.ResetLocalParameters(random);
|
---|
[5733] | 112 | rootNode.AddSubtree(startNode);
|
---|
[5686] | 113 | PTC2(random, startNode, maxTreeLength, maxTreeDepth);
|
---|
| 114 | tree.Root = rootNode;
|
---|
[3223] | 115 | return tree;
|
---|
| 116 | }
|
---|
| 117 |
|
---|
[3338] | 118 | private class TreeExtensionPoint {
|
---|
[5499] | 119 | public ISymbolicExpressionTreeNode Parent { get; set; }
|
---|
[3338] | 120 | public int ChildIndex { get; set; }
|
---|
| 121 | public int ExtensionPointDepth { get; set; }
|
---|
[3223] | 122 | }
|
---|
[3360] | 123 |
|
---|
[5686] | 124 | public static void PTC2(IRandom random, ISymbolicExpressionTreeNode seedNode,
|
---|
| 125 | int maxLength, int maxDepth) {
|
---|
[6009] | 126 | // make sure it is possible to create a trees smaller than maxLength and maxDepth
|
---|
| 127 | if (seedNode.Grammar.GetMinimumExpressionLength(seedNode.Symbol) > maxLength)
|
---|
| 128 | throw new ArgumentException("Cannot create trees of length " + maxLength + " or shorter because of grammar constraints.", "maxLength");
|
---|
| 129 | if (seedNode.Grammar.GetMinimumExpressionDepth(seedNode.Symbol) > maxDepth)
|
---|
| 130 | throw new ArgumentException("Cannot create trees of depth " + maxDepth + " or smaller because of grammar constraints.", "maxDepth");
|
---|
| 131 |
|
---|
[5549] | 132 | // tree length is limited by the grammar and by the explicit size constraints
|
---|
[5686] | 133 | int allowedMinLength = seedNode.Grammar.GetMinimumExpressionLength(seedNode.Symbol);
|
---|
| 134 | int allowedMaxLength = Math.Min(maxLength, seedNode.Grammar.GetMaximumExpressionLength(seedNode.Symbol));
|
---|
[3369] | 135 | int tries = 0;
|
---|
| 136 | while (tries++ < MAX_TRIES) {
|
---|
[5549] | 137 | // select a target tree length uniformly in the possible range (as determined by explicit limits and limits of the grammar)
|
---|
| 138 | int targetTreeLength;
|
---|
| 139 | targetTreeLength = random.Next(allowedMinLength, allowedMaxLength + 1);
|
---|
[5686] | 140 | if (targetTreeLength <= 1 || maxDepth <= 1) return;
|
---|
[3369] | 141 |
|
---|
[5727] | 142 | bool success = TryCreateFullTreeFromSeed(random, seedNode, seedNode.Grammar, targetTreeLength, maxDepth);
|
---|
[3369] | 143 |
|
---|
[4189] | 144 | // if successful => check constraints and return the tree if everything looks ok
|
---|
[5549] | 145 | if (success && seedNode.GetLength() <= maxLength && seedNode.GetDepth() <= maxDepth) {
|
---|
[5686] | 146 | return;
|
---|
[3369] | 147 | } else {
|
---|
| 148 | // clean seedNode
|
---|
[5733] | 149 | while (seedNode.Subtrees.Count() > 0) seedNode.RemoveSubtree(0);
|
---|
[3360] | 150 | }
|
---|
[5549] | 151 | // try a different length MAX_TRIES times
|
---|
[3369] | 152 | }
|
---|
[3825] | 153 | throw new ArgumentException("Couldn't create a random valid tree.");
|
---|
[3338] | 154 | }
|
---|
| 155 |
|
---|
[5727] | 156 | private static bool TryCreateFullTreeFromSeed(IRandom random, ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeGrammar globalGrammar,
|
---|
[5686] | 157 | int targetLength, int maxDepth) {
|
---|
[3338] | 158 | List<TreeExtensionPoint> extensionPoints = new List<TreeExtensionPoint>();
|
---|
[5549] | 159 | int currentLength = 1;
|
---|
[5686] | 160 | int totalListMinLength = globalGrammar.GetMinimumExpressionLength(root.Symbol) - 1;
|
---|
[5549] | 161 | int actualArity = SampleArity(random, root, targetLength);
|
---|
[5727] | 162 | if (actualArity < 0) return false;
|
---|
| 163 |
|
---|
[3223] | 164 | for (int i = 0; i < actualArity; i++) {
|
---|
| 165 | // insert a dummy sub-tree and add the pending extension to the list
|
---|
[3338] | 166 | var dummy = new SymbolicExpressionTreeNode();
|
---|
[5733] | 167 | root.AddSubtree(dummy);
|
---|
[4219] | 168 | extensionPoints.Add(new TreeExtensionPoint { Parent = root, ChildIndex = i, ExtensionPointDepth = 0 });
|
---|
[3223] | 169 | }
|
---|
| 170 | // while there are pending extension points and we have not reached the limit of adding new extension points
|
---|
[5549] | 171 | while (extensionPoints.Count > 0 && totalListMinLength + currentLength < targetLength) {
|
---|
[3294] | 172 | int randomIndex = random.Next(extensionPoints.Count);
|
---|
[3338] | 173 | TreeExtensionPoint nextExtension = extensionPoints[randomIndex];
|
---|
[3294] | 174 | extensionPoints.RemoveAt(randomIndex);
|
---|
[5499] | 175 | ISymbolicExpressionTreeNode parent = nextExtension.Parent;
|
---|
[3338] | 176 | int argumentIndex = nextExtension.ChildIndex;
|
---|
| 177 | int extensionDepth = nextExtension.ExtensionPointDepth;
|
---|
[6009] | 178 | if (parent.Grammar.GetMinimumExpressionDepth(parent.Symbol) >= maxDepth - extensionDepth) {
|
---|
[5686] | 179 | ReplaceWithMinimalTree(random, root, parent, argumentIndex);
|
---|
[3223] | 180 | } else {
|
---|
[6803] | 181 | var allowedSymbols = (from s in parent.Grammar.GetAllowedChildSymbols(parent.Symbol, argumentIndex)
|
---|
[5925] | 182 | where s.InitialFrequency > 0.0
|
---|
[6009] | 183 | where parent.Grammar.GetMinimumExpressionDepth(s) < maxDepth - extensionDepth + 1
|
---|
[5686] | 184 | where parent.Grammar.GetMaximumExpressionLength(s) > targetLength - totalListMinLength - currentLength
|
---|
[5499] | 185 | select s)
|
---|
| 186 | .ToList();
|
---|
[6233] | 187 | if (allowedSymbols.Count == 0) return false;
|
---|
[5499] | 188 | var weights = allowedSymbols.Select(x => x.InitialFrequency).ToList();
|
---|
| 189 | var selectedSymbol = allowedSymbols.SelectRandom(weights, random);
|
---|
| 190 | ISymbolicExpressionTreeNode newTree = selectedSymbol.CreateTreeNode();
|
---|
[3442] | 191 | if (newTree.HasLocalParameters) newTree.ResetLocalParameters(random);
|
---|
[5733] | 192 | parent.RemoveSubtree(argumentIndex);
|
---|
| 193 | parent.InsertSubtree(argumentIndex, newTree);
|
---|
[3338] | 194 |
|
---|
[5686] | 195 | var topLevelNode = newTree as SymbolicExpressionTreeTopLevelNode;
|
---|
| 196 | if (topLevelNode != null)
|
---|
| 197 | topLevelNode.SetGrammar((ISymbolicExpressionTreeGrammar)root.Grammar.Clone());
|
---|
[3338] | 198 |
|
---|
[5549] | 199 | currentLength++;
|
---|
| 200 | totalListMinLength--;
|
---|
[3223] | 201 |
|
---|
[5549] | 202 | actualArity = SampleArity(random, newTree, targetLength - currentLength);
|
---|
[5727] | 203 | if (actualArity < 0) return false;
|
---|
[3223] | 204 | for (int i = 0; i < actualArity; i++) {
|
---|
| 205 | // insert a dummy sub-tree and add the pending extension to the list
|
---|
[3338] | 206 | var dummy = new SymbolicExpressionTreeNode();
|
---|
[5733] | 207 | newTree.AddSubtree(dummy);
|
---|
[3338] | 208 | extensionPoints.Add(new TreeExtensionPoint { Parent = newTree, ChildIndex = i, ExtensionPointDepth = extensionDepth + 1 });
|
---|
[3223] | 209 | }
|
---|
[5686] | 210 | totalListMinLength += newTree.Grammar.GetMinimumExpressionLength(newTree.Symbol);
|
---|
[3223] | 211 | }
|
---|
| 212 | }
|
---|
| 213 | // fill all pending extension points
|
---|
[3294] | 214 | while (extensionPoints.Count > 0) {
|
---|
| 215 | int randomIndex = random.Next(extensionPoints.Count);
|
---|
[3338] | 216 | TreeExtensionPoint nextExtension = extensionPoints[randomIndex];
|
---|
[3294] | 217 | extensionPoints.RemoveAt(randomIndex);
|
---|
[5499] | 218 | ISymbolicExpressionTreeNode parent = nextExtension.Parent;
|
---|
[3338] | 219 | int a = nextExtension.ChildIndex;
|
---|
| 220 | int d = nextExtension.ExtensionPointDepth;
|
---|
[5686] | 221 | ReplaceWithMinimalTree(random, root, parent, a);
|
---|
[3223] | 222 | }
|
---|
[5727] | 223 | return true;
|
---|
[2210] | 224 | }
|
---|
[3223] | 225 |
|
---|
[5499] | 226 | private static void ReplaceWithMinimalTree(IRandom random, ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeNode parent,
|
---|
[5686] | 227 | int childIndex) {
|
---|
[3338] | 228 | // determine possible symbols that will lead to the smallest possible tree
|
---|
[5686] | 229 | var possibleSymbols = (from s in parent.Grammar.GetAllowedChildSymbols(parent.Symbol, childIndex)
|
---|
[5925] | 230 | where s.InitialFrequency > 0.0
|
---|
[5686] | 231 | group s by parent.Grammar.GetMinimumExpressionLength(s) into g
|
---|
[3338] | 232 | orderby g.Key
|
---|
[5499] | 233 | select g).First().ToList();
|
---|
| 234 | var weights = possibleSymbols.Select(x => x.InitialFrequency).ToList();
|
---|
| 235 | var selectedSymbol = possibleSymbols.SelectRandom(weights, random);
|
---|
[3338] | 236 | var tree = selectedSymbol.CreateTreeNode();
|
---|
[3442] | 237 | if (tree.HasLocalParameters) tree.ResetLocalParameters(random);
|
---|
[5733] | 238 | parent.RemoveSubtree(childIndex);
|
---|
| 239 | parent.InsertSubtree(childIndex, tree);
|
---|
[5686] | 240 |
|
---|
| 241 | var topLevelNode = tree as SymbolicExpressionTreeTopLevelNode;
|
---|
| 242 | if (topLevelNode != null)
|
---|
| 243 | topLevelNode.SetGrammar((ISymbolicExpressionTreeGrammar)root.Grammar.Clone());
|
---|
| 244 |
|
---|
| 245 | for (int i = 0; i < tree.Grammar.GetMinimumSubtreeCount(tree.Symbol); i++) {
|
---|
[3338] | 246 | // insert a dummy sub-tree and add the pending extension to the list
|
---|
| 247 | var dummy = new SymbolicExpressionTreeNode();
|
---|
[5733] | 248 | tree.AddSubtree(dummy);
|
---|
[3338] | 249 | // replace the just inserted dummy by recursive application
|
---|
[5686] | 250 | ReplaceWithMinimalTree(random, root, tree, i);
|
---|
[3338] | 251 | }
|
---|
| 252 | }
|
---|
[3360] | 253 |
|
---|
[5499] | 254 | private static bool IsTopLevelBranch(ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeNode branch) {
|
---|
[3369] | 255 | return branch is SymbolicExpressionTreeTopLevelNode;
|
---|
[3338] | 256 | }
|
---|
| 257 |
|
---|
[5549] | 258 | private static int SampleArity(IRandom random, ISymbolicExpressionTreeNode node, int targetLength) {
|
---|
[3223] | 259 | // select actualArity randomly with the constraint that the sub-trees in the minimal arity can become large enough
|
---|
[5686] | 260 | int minArity = node.Grammar.GetMinimumSubtreeCount(node.Symbol);
|
---|
| 261 | int maxArity = node.Grammar.GetMaximumSubtreeCount(node.Symbol);
|
---|
[5549] | 262 | if (maxArity > targetLength) {
|
---|
| 263 | maxArity = targetLength;
|
---|
[3223] | 264 | }
|
---|
[5549] | 265 | // the min number of sub-trees has to be set to a value that is large enough so that the largest possible tree is at least tree length
|
---|
| 266 | // if 1..3 trees are possible and the largest possible first sub-tree is smaller larger than the target length then minArity should be at least 2
|
---|
[3237] | 267 | long aggregatedLongestExpressionLength = 0;
|
---|
[3223] | 268 | for (int i = 0; i < maxArity; i++) {
|
---|
[5686] | 269 | aggregatedLongestExpressionLength += (from s in node.Grammar.GetAllowedChildSymbols(node.Symbol, i)
|
---|
[5925] | 270 | where s.InitialFrequency > 0.0
|
---|
[5686] | 271 | select node.Grammar.GetMaximumExpressionLength(s)).Max();
|
---|
[6803] | 272 | if (i > minArity && aggregatedLongestExpressionLength < targetLength) minArity = i + 1;
|
---|
[3223] | 273 | else break;
|
---|
| 274 | }
|
---|
| 275 |
|
---|
[5549] | 276 | // the max number of sub-trees has to be set to a value that is small enough so that the smallest possible tree is at most tree length
|
---|
| 277 | // if 1..3 trees are possible and the smallest possible first sub-tree is already larger than the target length then maxArity should be at most 0
|
---|
[3237] | 278 | long aggregatedShortestExpressionLength = 0;
|
---|
[3223] | 279 | for (int i = 0; i < maxArity; i++) {
|
---|
[5686] | 280 | aggregatedShortestExpressionLength += (from s in node.Grammar.GetAllowedChildSymbols(node.Symbol, i)
|
---|
[5925] | 281 | where s.InitialFrequency > 0.0
|
---|
[5686] | 282 | select node.Grammar.GetMinimumExpressionLength(s)).Min();
|
---|
[5549] | 283 | if (aggregatedShortestExpressionLength > targetLength) {
|
---|
[3223] | 284 | maxArity = i;
|
---|
| 285 | break;
|
---|
| 286 | }
|
---|
| 287 | }
|
---|
[5727] | 288 | if (minArity > maxArity) return -1;
|
---|
[3223] | 289 | return random.Next(minArity, maxArity + 1);
|
---|
| 290 | }
|
---|
[2447] | 291 | }
|
---|
[645] | 292 | } |
---|