Free cookie consent management tool by TermsFeed Policy Generator

source: branches/HeuristicLab.EvolutionaryTracking/HeuristicLab.EvolutionaryTracking/3.4/Analyzers/SymbolicExpressionTreeFragmentsAnalyzer.cs @ 9082

Last change on this file since 9082 was 9082, checked in by bburlacu, 11 years ago

#1772: Renamed and refactored genealogy graph components. Added SymbolGraph and FPGraph (frequent pattern graph). Added evolvability and genetic operator average improvement analyzer.

File size: 16.4 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2012 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.Collections.Generic;
23using System.Linq;
24using HeuristicLab.Common;
25using HeuristicLab.Core;
26using HeuristicLab.Data;
27using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
28using HeuristicLab.Operators;
29using HeuristicLab.Optimization;
30using HeuristicLab.Parameters;
31using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
32using HeuristicLab.Problems.DataAnalysis;
33using HeuristicLab.Problems.DataAnalysis.Symbolic;
34// type definitions for convenience
35using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
36using CloneMapType = HeuristicLab.Core.ItemDictionary<HeuristicLab.Core.IItem, HeuristicLab.Core.IItem>;
37using TraceMapType = HeuristicLab.Core.ItemDictionary<HeuristicLab.Core.IItem, HeuristicLab.Core.IItemList<HeuristicLab.Core.IItem>>;
38
39namespace HeuristicLab.EvolutionaryTracking {
40  /// <summary>
41  /// An operator that gathers information on tree fragments that get passed from one generation to the next via genetic operators
42  /// (Tracking of genetic variance and heredity)
43  /// </summary>
44  [Item("SymbolicExpressionTreeFragmentsAnalyzer", "An operator that provides statistics about crossover fragments")]
45  [StorableClass]
46  public abstract class SymbolicExpressionTreeFragmentsAnalyzer : SingleSuccessorOperator, IAnalyzer {
47    #region Parameter names
48    protected const string UpdateIntervalParameterName = "UpdateInterval";
49    protected const string UpdateCounterParameterName = "UpdateCounter";
50    protected const string ResultsParameterName = "Results";
51    protected const string SecondaryTraceMapParameterName = "SecondaryTraceMap";
52    protected const string SecondaryFragmentMapParameterName = "SecondaryFragmentMap";
53    protected const string SecondaryCloneMapParameterName = "SecondaryCloneMap";
54    protected const string GlobalTraceMapParameterName = "GlobalTraceMap";
55    protected const string GlobalTraceMapParameterDescription = "Global map of trees and their parents.";
56    protected const string GlobalCloneMapParameterName = "GlobalCloneMap";
57    protected const string GlobalCloneMapParameterDescription = "Global map of trees and their clones";
58    protected const string GlobalFragmentMapParameterName = "GlobalFragmentMap";
59    protected const string GlobalFragmentMapParameterDescription = "Global map of trees and their respective fragments.";
60    protected const string GenerationsParameterName = "Generations";
61    protected const string SymbolicExpressionInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
62    protected const string SymbolicRegressionProblemDataParameterName = "ProblemData";
63    #endregion
64
65    // impact values calculator
66    private readonly SymbolicRegressionSolutionImpactValuesCalculator calculator = new SymbolicRegressionSolutionImpactValuesCalculator();
67
68    #region Parameter properties
69    public ValueParameter<IntValue> UpdateIntervalParameter {
70      get { return (ValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
71    }
72    public ValueParameter<IntValue> UpdateCounterParameter {
73      get { return (ValueParameter<IntValue>)Parameters[UpdateCounterParameterName]; }
74    }
75    public LookupParameter<ResultCollection> ResultsParameter {
76      get { return (LookupParameter<ResultCollection>)Parameters[ResultsParameterName]; }
77    }
78    public LookupParameter<IntValue> GenerationsParameter {
79      get { return (LookupParameter<IntValue>)Parameters[GenerationsParameterName]; }
80    }
81    // tracking structures
82    public LookupParameter<TraceMapType> GlobalTraceMapParameter {
83      get { return (LookupParameter<TraceMapType>)Parameters[GlobalTraceMapParameterName]; }
84    }
85    public LookupParameter<CloneMapType> GlobalCloneMapParameter {
86      get { return (LookupParameter<CloneMapType>)Parameters[GlobalCloneMapParameterName]; }
87    }
88    public LookupParameter<CloneMapType> GlobalFragmentMapParameter {
89      get { return (LookupParameter<CloneMapType>)Parameters[GlobalFragmentMapParameterName]; }
90    }
91    // auxiliary structures for tracking fragments and individuals from the previous generation
92    // (this is needed because tracking is done in connection to the current generation, i.e., for
93    // counting "successful" offspring and fragments
94    public LookupParameter<TraceMapType> SecondaryTraceMapParameter {
95      get { return (LookupParameter<TraceMapType>)Parameters[SecondaryTraceMapParameterName]; }
96    }
97    public LookupParameter<CloneMapType> SecondaryFragmentMapParameter {
98      get { return (LookupParameter<CloneMapType>)Parameters[SecondaryFragmentMapParameterName]; }
99    }
100    public LookupParameter<CloneMapType> SecondaryCloneMapParameter {
101      get { return (LookupParameter<CloneMapType>)Parameters[SecondaryCloneMapParameterName]; }
102    }
103    // problem data, interpreter and evaluator
104    public LookupParameter<SymbolicDataAnalysisExpressionTreeInterpreter> SymbolicExpressionInterpreterParameter {
105      get { return (LookupParameter<SymbolicDataAnalysisExpressionTreeInterpreter>)Parameters[SymbolicExpressionInterpreterParameterName]; }
106    }
107    public LookupParameter<RegressionProblemData> SymbolicRegressionProblemDataParameter {
108      get { return (LookupParameter<RegressionProblemData>)Parameters[SymbolicRegressionProblemDataParameterName]; }
109    }
110    #endregion
111
112    #region Parameters
113    public bool EnabledByDefault {
114      get { return true; }
115    }
116    public IntValue UpdateInterval {
117      get { return UpdateIntervalParameter.Value; }
118    }
119    public IntValue UpdateCounter {
120      get { return UpdateCounterParameter.Value; }
121    }
122    public ResultCollection Results {
123      get { return ResultsParameter.ActualValue; }
124    }
125    public CloneMapType GlobalCloneMap {
126      get { return GlobalCloneMapParameter.ActualValue; }
127    }
128    public TraceMapType GlobalTraceMap {
129      get { return GlobalTraceMapParameter.ActualValue; }
130    }
131    public TraceMapType SecondaryTraceMap {
132      get { return SecondaryTraceMapParameter.ActualValue; }
133      set { SecondaryTraceMapParameter.ActualValue = value; }
134    }
135    public CloneMapType SecondaryFragmentMap {
136      get { return SecondaryFragmentMapParameter.ActualValue; }
137      set { SecondaryFragmentMapParameter.ActualValue = value; }
138    }
139    public CloneMapType SecondaryCloneMap {
140      get { return SecondaryCloneMapParameter.ActualValue; }
141      set { SecondaryCloneMapParameter.ActualValue = value; }
142    }
143    public CloneMapType GlobalFragmentMap {
144      get { return GlobalFragmentMapParameter.ActualValue; }
145    }
146    public IntValue Generations {
147      get { return GenerationsParameter.ActualValue; }
148    }
149    public SymbolicDataAnalysisExpressionTreeInterpreter SymbolicExpressionInterpreter {
150      get { return SymbolicExpressionInterpreterParameter.ActualValue; }
151    }
152    public RegressionProblemData SymbolicRegressionProblemData {
153      get { return SymbolicRegressionProblemDataParameter.ActualValue; }
154    }
155    #endregion
156
157    [StorableConstructor]
158    protected SymbolicExpressionTreeFragmentsAnalyzer(bool deserializing) : base(deserializing) { }
159
160    protected SymbolicExpressionTreeFragmentsAnalyzer(SymbolicExpressionTreeFragmentsAnalyzer original, Cloner cloner) : base(original, cloner) { }
161
162    protected SymbolicExpressionTreeFragmentsAnalyzer()
163      : base() {
164      #region Add parameters
165      // analyzer update counter and update interval
166      Parameters.Add(new ValueParameter<IntValue>(UpdateIntervalParameterName, "The interval in which the tree length analysis should be applied.", new IntValue(1)));
167      Parameters.Add(new ValueParameter<IntValue>(UpdateCounterParameterName, "The value which counts how many times the operator was called since the last update", new IntValue(0)));
168      // tracking
169      Parameters.Add(new LookupParameter<TraceMapType>(GlobalTraceMapParameterName, GlobalTraceMapParameterDescription));
170      Parameters.Add(new LookupParameter<CloneMapType>(GlobalCloneMapParameterName, GlobalCloneMapParameterDescription));
171      Parameters.Add(new LookupParameter<CloneMapType>(GlobalFragmentMapParameterName, GlobalFragmentMapParameterDescription));
172      // secondary tracking
173      Parameters.Add(new LookupParameter<TraceMapType>(SecondaryTraceMapParameterName, GlobalTraceMapParameterDescription));
174      Parameters.Add(new LookupParameter<CloneMapType>(SecondaryCloneMapParameterName, GlobalCloneMapParameterDescription));
175      Parameters.Add(new LookupParameter<CloneMapType>(SecondaryFragmentMapParameterName, GlobalFragmentMapParameterDescription));
176      // impact calculation
177      Parameters.Add(new LookupParameter<SymbolicDataAnalysisExpressionTreeInterpreter>(SymbolicExpressionInterpreterParameterName, "Interpreter for symbolic expression trees"));
178      Parameters.Add(new LookupParameter<RegressionProblemData>(SymbolicRegressionProblemDataParameterName, "The symbolic data analysis problem."));
179      // fragment statistics parameters
180      Parameters.Add(new ValueLookupParameter<ResultCollection>(ResultsParameterName, "The results collection where the analysis values should be stored."));
181      Parameters.Add(new LookupParameter<IntValue>(GenerationsParameterName, "The number of generations so far."));
182      #endregion
183
184      UpdateCounterParameter.Hidden = true;
185      UpdateIntervalParameter.Hidden = true;
186
187    }
188    #region After deserialization code
189    [StorableHook(HookType.AfterDeserialization)]
190    private void AfterDeserialization() {
191      // check if all the parameters are present and accounted for
192      if (!Parameters.ContainsKey(UpdateIntervalParameterName)) {
193        Parameters.Add(new ValueParameter<IntValue>(UpdateIntervalParameterName, "The interval in which the tree length analysis should be applied.", new IntValue(1)));
194      }
195      if (!Parameters.ContainsKey(UpdateCounterParameterName)) {
196        Parameters.Add(new ValueParameter<IntValue>(UpdateCounterParameterName, "The value which counts how many times the operator was called since the last update", new IntValue(0)));
197        UpdateCounterParameter.Hidden = true;
198      }
199    }
200    #endregion
201    #region IStatefulItem members
202    public override void InitializeState() {
203      base.InitializeState();
204      UpdateCounter.Value = 0;
205    }
206
207    public override void ClearState() {
208      base.ClearState();
209      UpdateCounter.Value = 0;
210    }
211    #endregion
212
213    /* A small description of the way this analyzer works
214     *
215     * We keep 3 maps in the global scope: the GlobalTraceMap, the GlobalFragmentMap and the GlobalCloneMap that are updated on each step: selection, crossover, mutation
216     * These maps provide information about the children, parents and transferred fragments as follows:
217     *
218     * 1) GlobalCloneMap
219     *    This data structure provides the basis for tracking. In the selection step, every selected individual gets cloned and transferred into the recombination pool.
220     *    The GlobalCloneMap keeps track of clones by mapping them to the original individuals. This is an essential step for building the genealogy graph. Using this
221     *    structure we can find out how many times each individual was selected.
222     *    The GlobalCloneMap consists of Key-Value pairs Clone->Original
223     *
224     * 2) GlobalTraceMap
225     *    Keeps information in the form of key-value pairs Child->{Parents}, where {Parents} is a set consisting of
226     *      - one parent individual (in the case of mutation), by language abuse we say that the individual that mutation acted on is the parent,
227     *        and the resulting individual is the child
228     *      - two parent individuals (in the case of crossover), named Parent0 and Parent1. Parent0 is the individual that crossover acts on by replacing
229     *        one of its subtrees with a compatible subtree, randomly selected from Parent1
230     *
231     *    CROSSOVER
232     *        Parent0     Parent1                                                                    {Parents}
233     *           \           /                                                                           ^
234     *            \       fragment (inserted from Parent1 into Parent0 to create the Child)              ^    [Mapping provided by the GlobalTraceMap]
235     *             \       /                                                                             ^
236     *              \     /                                                                              ^
237     *               Child                                                                             Child
238     *               
239     *    MUTATION
240     *        Parent0
241     *           |
242     *           |    [mutation acts on a random node/subtree in the 'parent' individual
243     *           |
244     *         Child
245     *         
246     *    Since crossover and mutation act on individuals in the recombination pool (which in fact are clones of individuals from the previous generation), the GlobalTraceMap
247     *    is populated with values given by the mappings in the GlobalCloneMap, so that the original parent for each child is known.
248     *   
249     *  3) GlobalFragmentMap
250     *     Keeps information in the form of Key-Value pairs about an individual and its respective fragment
251     *     (ie., the subtree received via crossover or created/altered by mutation)
252     * */
253
254    public override IOperation Apply() {
255      // save the current generation so it can be compared to the following one, next time the analyzer is applied
256      SecondaryTraceMap = SecondaryTraceMap ?? new TraceMapType();
257      SecondaryCloneMap = SecondaryCloneMap ?? new CloneMapType();
258      SecondaryFragmentMap = SecondaryFragmentMap ?? new CloneMapType();
259
260      SecondaryTraceMap.Clear();
261      if (GlobalTraceMap != null)
262        foreach (var m in GlobalTraceMap)
263          SecondaryTraceMap.Add(m.Key, m.Value);
264
265      SecondaryCloneMap.Clear();
266      if (GlobalCloneMap != null)
267        foreach (var m in GlobalCloneMap)
268          SecondaryCloneMap.Add(m.Key, m.Value);
269
270      SecondaryFragmentMap.Clear();
271      if (GlobalFragmentMap != null)
272        foreach (var m in GlobalFragmentMap)
273          SecondaryFragmentMap.Add(m.Key, m.Value);
274      return base.Apply();
275    }
276
277    protected virtual void InitializeParameters() {
278      #region Init parameters
279
280      #endregion
281    }
282
283    private static int VisitationLength(ISymbolicExpressionTree tree) {
284      return VisitationLength(tree.Root);
285    }
286
287    private static int VisitationLength(ISymbolicExpressionTreeNode node) {
288      return node.IterateNodesBreadth().Sum(n => n.GetLength());
289    }
290
291    #region Similarity computations
292    /// <summary>
293    /// Provide a measure of how similar the fragments that get passed by crossover from one generation to the next are.
294    /// </summary>
295    /// <param name="fragments">The symbolic expression tree fragments</param>
296    /// <param name="mode">The similarity mode (0 - exact, 1 - high, 2 - relaxed)</param>
297    /// <returns>The average number of similar fragments</returns>
298    private static double CalculateSimilarity(IList<IFragment> fragments, SymbolicExpressionTreeNodeSimilarityComparer comparer) {
299      var visited = new bool[fragments.Count];
300      int groups = 0;
301      for (int i = 0; i != fragments.Count - 1; ++i) {
302        if (visited[i]) continue;
303        for (int j = i + 1; j != fragments.Count; ++j) {
304          if (visited[j]) continue;
305          if (fragments[i].Root.IsSimilarTo(fragments[j].Root, comparer))
306            visited[j] = true;
307        }
308        ++groups;
309      }
310      return (double)fragments.Count / groups;
311    }
312    #endregion
313  } //SymbolicExpressionTreeFragmentsAnalyzer
314}
Note: See TracBrowser for help on using the repository browser.