1 | using System;
|
---|
2 | using System.Diagnostics;
|
---|
3 | using HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration.GrammarEnumeration;
|
---|
4 | using HeuristicLab.Analysis;
|
---|
5 | using HeuristicLab.Common;
|
---|
6 | using HeuristicLab.Core;
|
---|
7 | using HeuristicLab.Data;
|
---|
8 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
9 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
10 | using HeuristicLab.Problems.DataAnalysis;
|
---|
11 | using HeuristicLab.Problems.DataAnalysis.Symbolic;
|
---|
12 | using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
|
---|
13 |
|
---|
14 | namespace HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration {
|
---|
15 | [Item("RSquaredEvaluator", "")]
|
---|
16 | [StorableClass]
|
---|
17 | public class RSquaredEvaluator : Item, IGrammarEnumerationAnalyzer {
|
---|
18 | public static readonly string BestTrainingQualityResultName = "Best R² (Training)";
|
---|
19 | public static readonly string BestTestQualityResultName = "Best R² (Test)";
|
---|
20 | public static readonly string BestTrainingModelResultName = "Best model (Training)";
|
---|
21 | public static readonly string BestTrainingSolutionResultName = "Best solution (Training)";
|
---|
22 | public static readonly string BestComplexityResultName = "Best solution complexity";
|
---|
23 | public static readonly string BestSolutions = "Best solutions";
|
---|
24 |
|
---|
25 | private static readonly ISymbolicDataAnalysisExpressionTreeInterpreter expressionTreeLinearInterpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
|
---|
26 |
|
---|
27 | public RSquaredEvaluator() { }
|
---|
28 |
|
---|
29 | [StorableConstructor]
|
---|
30 | protected RSquaredEvaluator(bool deserializing) : base(deserializing) { }
|
---|
31 |
|
---|
32 | protected RSquaredEvaluator(RSquaredEvaluator original, Cloner cloner) : base(original, cloner) {
|
---|
33 | }
|
---|
34 |
|
---|
35 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
36 | return new RSquaredEvaluator(this, cloner);
|
---|
37 | }
|
---|
38 |
|
---|
39 | public void Register(GrammarEnumerationAlgorithm algorithm) {
|
---|
40 | algorithm.Started += OnStarted;
|
---|
41 | algorithm.Stopped += OnStopped;
|
---|
42 | algorithm.DistinctSentenceGenerated += AlgorithmOnDistinctSentenceGenerated;
|
---|
43 | }
|
---|
44 |
|
---|
45 | public void Deregister(GrammarEnumerationAlgorithm algorithm) {
|
---|
46 | algorithm.Started -= OnStarted;
|
---|
47 | algorithm.Stopped -= OnStopped;
|
---|
48 | algorithm.DistinctSentenceGenerated -= AlgorithmOnDistinctSentenceGenerated;
|
---|
49 | }
|
---|
50 |
|
---|
51 | private void AlgorithmOnDistinctSentenceGenerated(object sender, PhraseAddedEventArgs phraseAddedEventArgs) {
|
---|
52 | GrammarEnumerationAlgorithm algorithm = (GrammarEnumerationAlgorithm)sender;
|
---|
53 | EvaluateSentence(algorithm, phraseAddedEventArgs.NewPhrase, algorithm.OptimizeConstants);
|
---|
54 | }
|
---|
55 |
|
---|
56 | private void OnStarted(object sender, EventArgs eventArgs) {
|
---|
57 | GrammarEnumerationAlgorithm algorithm = (GrammarEnumerationAlgorithm)sender;
|
---|
58 |
|
---|
59 | algorithm.BestTrainingSentence = null;
|
---|
60 | }
|
---|
61 |
|
---|
62 | private void OnStopped(object sender, EventArgs eventArgs) { }
|
---|
63 |
|
---|
64 | private T GetValue<T>(IItem value) where T : struct {
|
---|
65 | var v = value as ValueTypeValue<T>;
|
---|
66 | if (v == null)
|
---|
67 | throw new ArgumentException(string.Format("Item is not of type {0}", typeof(ValueTypeValue<T>)));
|
---|
68 | return v.Value;
|
---|
69 | }
|
---|
70 |
|
---|
71 | private void EvaluateSentence(GrammarEnumerationAlgorithm algorithm, SymbolString symbolString, bool optimizeConstants) {
|
---|
72 | var results = algorithm.Results;
|
---|
73 | var grammar = algorithm.Grammar;
|
---|
74 | var problemData = algorithm.Problem.ProblemData;
|
---|
75 |
|
---|
76 | SymbolicExpressionTree tree = algorithm.Grammar.ParseSymbolicExpressionTree(symbolString);
|
---|
77 | Debug.Assert(SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree));
|
---|
78 |
|
---|
79 | double r2 = Evaluate(problemData, tree, optimizeConstants);
|
---|
80 | double bestR2 = results.ContainsKey(BestTrainingQualityResultName) ? GetValue<double>(results[BestTrainingQualityResultName].Value) : 0.0;
|
---|
81 | if (r2 < bestR2)
|
---|
82 | return;
|
---|
83 |
|
---|
84 | var bestComplexity = results.ContainsKey(BestComplexityResultName) ? GetValue<int>(results[BestComplexityResultName].Value) : int.MaxValue;
|
---|
85 | var complexity = grammar.GetComplexity(symbolString);
|
---|
86 |
|
---|
87 | if (algorithm.BestTrainingSentence == null || r2 > bestR2 || (r2.IsAlmost(bestR2) && complexity < bestComplexity)) {
|
---|
88 | algorithm.BestTrainingSentence = symbolString;
|
---|
89 |
|
---|
90 | var model = new SymbolicRegressionModel(problemData.TargetVariable, tree, expressionTreeLinearInterpreter);
|
---|
91 | model.Scale(problemData);
|
---|
92 | var bestSolution = model.CreateRegressionSolution(problemData);
|
---|
93 |
|
---|
94 | results.AddOrUpdateResult(BestTrainingQualityResultName, new DoubleValue(bestSolution.TrainingRSquared));
|
---|
95 | results.AddOrUpdateResult(BestTestQualityResultName, new DoubleValue(bestSolution.TestRSquared));
|
---|
96 | results.AddOrUpdateResult(BestTrainingModelResultName, bestSolution.Model);
|
---|
97 | results.AddOrUpdateResult(BestTrainingSolutionResultName, bestSolution);
|
---|
98 | results.AddOrUpdateResult(BestComplexityResultName, new IntValue(complexity));
|
---|
99 |
|
---|
100 | // record best sentence quality & length
|
---|
101 | DataTable dt;
|
---|
102 | if (!results.ContainsKey(BestSolutions)) {
|
---|
103 | var names = new[] { "Quality", "Relative Length", "Complexity", "Timestamp" };
|
---|
104 | dt = new DataTable();
|
---|
105 | foreach (var name in names) {
|
---|
106 | dt.Rows.Add(new DataRow(name) { VisualProperties = { StartIndexZero = true } });
|
---|
107 | }
|
---|
108 | results.AddOrUpdateResult(BestSolutions, dt);
|
---|
109 | }
|
---|
110 | dt = (DataTable)results[BestSolutions].Value;
|
---|
111 | dt.Rows["Quality"].Values.Add(r2);
|
---|
112 | dt.Rows["Relative Length"].Values.Add((double)symbolString.Count() / algorithm.MaxSentenceLength);
|
---|
113 | dt.Rows["Complexity"].Values.Add(complexity);
|
---|
114 | dt.Rows["Timestamp"].Values.Add(algorithm.ExecutionTime.TotalMilliseconds / 1000d);
|
---|
115 | }
|
---|
116 | }
|
---|
117 |
|
---|
118 | public static double Evaluate(IRegressionProblemData problemData, SymbolicExpressionTree tree, bool optimizeConstants = true) {
|
---|
119 | double r2;
|
---|
120 |
|
---|
121 | // TODO: Initialize constant values randomly
|
---|
122 | // TODO: Restarts
|
---|
123 | if (optimizeConstants) {
|
---|
124 | r2 = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(expressionTreeLinearInterpreter,
|
---|
125 | tree,
|
---|
126 | problemData,
|
---|
127 | problemData.TrainingIndices,
|
---|
128 | applyLinearScaling: true,
|
---|
129 | maxIterations: 10,
|
---|
130 | updateVariableWeights: false,
|
---|
131 | updateConstantsInTree: true);
|
---|
132 |
|
---|
133 | foreach (var symbolicExpressionTreeNode in tree.IterateNodesPostfix()) {
|
---|
134 | ConstantTreeNode constTreeNode = symbolicExpressionTreeNode as ConstantTreeNode;
|
---|
135 | if (constTreeNode != null && constTreeNode.Value.IsAlmost(0.0)) {
|
---|
136 | constTreeNode.Value = 0.0;
|
---|
137 | }
|
---|
138 | }
|
---|
139 | } else {
|
---|
140 | r2 = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(expressionTreeLinearInterpreter,
|
---|
141 | tree,
|
---|
142 | double.MinValue,
|
---|
143 | double.MaxValue,
|
---|
144 | problemData,
|
---|
145 | problemData.TrainingIndices,
|
---|
146 | applyLinearScaling: true);
|
---|
147 | }
|
---|
148 | return double.IsNaN(r2) ? 0.0 : r2;
|
---|
149 | }
|
---|
150 | }
|
---|
151 | }
|
---|