1 | using System;
|
---|
2 | using System.Collections.Generic;
|
---|
3 | using System.Linq;
|
---|
4 | using System.Text;
|
---|
5 | using System.Threading.Tasks;
|
---|
6 | using HeuristicLab.Core;
|
---|
7 | using HeuristicLab.Optimization;
|
---|
8 | using HEAL.Attic;
|
---|
9 | using HeuristicLab.Common;
|
---|
10 | using HeuristicLab.Problems.Instances;
|
---|
11 | using HeuristicLab.Parameters;
|
---|
12 | using HeuristicLab.Data;
|
---|
13 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
14 | using HeuristicLab.PluginInfrastructure;
|
---|
15 |
|
---|
16 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
|
---|
17 | [StorableType("7464E84B-65CC-440A-91F0-9FA920D730F9")]
|
---|
18 | [Item(Name = "Structured Symbolic Regression Single Objective Problem (single-objective)", Description = "A problem with a structural definition and unfixed subfunctions.")]
|
---|
19 | [Creatable(CreatableAttribute.Categories.GeneticProgrammingProblems, Priority = 150)]
|
---|
20 | public class StructuredSymbolicRegressionSingleObjectiveProblem : SingleObjectiveBasicProblem<MultiEncoding>, IRegressionProblem, IProblemInstanceConsumer<IRegressionProblemData> {
|
---|
21 |
|
---|
22 | #region Constants
|
---|
23 | private const string ProblemDataParameterName = "ProblemData";
|
---|
24 | private const string StructureTemplateParameterName = "Structure Template";
|
---|
25 | private const string InterpreterParameterName = "Interpreter";
|
---|
26 |
|
---|
27 | private const string StructureTemplateDescriptionText =
|
---|
28 | "Enter your expression as string in infix format into the empty input field.\n" +
|
---|
29 | "By checking the \"Apply Linear Scaling\" checkbox you can add the relevant scaling terms to your expression.\n" +
|
---|
30 | "After entering the expression click parse to build the tree.\n" +
|
---|
31 | "To edit the defined sub-functions, click on the coressponding colored node in the tree view.";
|
---|
32 | #endregion
|
---|
33 |
|
---|
34 | #region Parameters
|
---|
35 | public IValueParameter<IRegressionProblemData> ProblemDataParameter => (IValueParameter<IRegressionProblemData>)Parameters[ProblemDataParameterName];
|
---|
36 | public IFixedValueParameter<StructureTemplate> StructureTemplateParameter => (IFixedValueParameter<StructureTemplate>)Parameters[StructureTemplateParameterName];
|
---|
37 | public IValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter> InterpreterParameter => (IValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>)Parameters[InterpreterParameterName];
|
---|
38 | #endregion
|
---|
39 |
|
---|
40 | #region Properties
|
---|
41 | public IRegressionProblemData ProblemData {
|
---|
42 | get => ProblemDataParameter.Value;
|
---|
43 | set {
|
---|
44 | ProblemDataParameter.Value = value;
|
---|
45 | ProblemDataChanged?.Invoke(this, EventArgs.Empty);
|
---|
46 | }
|
---|
47 | }
|
---|
48 |
|
---|
49 | public StructureTemplate StructureTemplate => StructureTemplateParameter.Value;
|
---|
50 |
|
---|
51 | public ISymbolicDataAnalysisExpressionTreeInterpreter Interpreter => InterpreterParameter.Value;
|
---|
52 |
|
---|
53 | IParameter IDataAnalysisProblem.ProblemDataParameter => ProblemDataParameter;
|
---|
54 | IDataAnalysisProblemData IDataAnalysisProblem.ProblemData => ProblemData;
|
---|
55 |
|
---|
56 | public override bool Maximization => true;
|
---|
57 | #endregion
|
---|
58 |
|
---|
59 | #region EventHandlers
|
---|
60 | public event EventHandler ProblemDataChanged;
|
---|
61 | #endregion
|
---|
62 |
|
---|
63 | #region Constructors & Cloning
|
---|
64 | public StructuredSymbolicRegressionSingleObjectiveProblem() {
|
---|
65 | var problemData = new ShapeConstrainedRegressionProblemData();
|
---|
66 |
|
---|
67 | var structureTemplate = new StructureTemplate();
|
---|
68 | structureTemplate.Changed += OnTemplateChanged;
|
---|
69 |
|
---|
70 | Parameters.Add(new ValueParameter<IRegressionProblemData>(
|
---|
71 | ProblemDataParameterName,
|
---|
72 | problemData));
|
---|
73 |
|
---|
74 | Parameters.Add(new FixedValueParameter<StructureTemplate>(
|
---|
75 | StructureTemplateParameterName,
|
---|
76 | StructureTemplateDescriptionText,
|
---|
77 | structureTemplate));
|
---|
78 |
|
---|
79 | Parameters.Add(new ValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>(
|
---|
80 | InterpreterParameterName,
|
---|
81 | new SymbolicDataAnalysisExpressionTreeInterpreter())
|
---|
82 | { Hidden = true });
|
---|
83 |
|
---|
84 | ProblemDataParameter.ValueChanged += ProblemDataParameterValueChanged;
|
---|
85 | }
|
---|
86 |
|
---|
87 | public StructuredSymbolicRegressionSingleObjectiveProblem(StructuredSymbolicRegressionSingleObjectiveProblem original,
|
---|
88 | Cloner cloner) : base(original, cloner){ }
|
---|
89 |
|
---|
90 | [StorableConstructor]
|
---|
91 | protected StructuredSymbolicRegressionSingleObjectiveProblem(StorableConstructorFlag _) : base(_) { }
|
---|
92 | #endregion
|
---|
93 |
|
---|
94 | #region Cloning
|
---|
95 | public override IDeepCloneable Clone(Cloner cloner) =>
|
---|
96 | new StructuredSymbolicRegressionSingleObjectiveProblem(this, cloner);
|
---|
97 | #endregion
|
---|
98 |
|
---|
99 | private void ProblemDataParameterValueChanged(object sender, EventArgs e) {
|
---|
100 | StructureTemplate.Reset();
|
---|
101 | // InfoBox for Reset?
|
---|
102 | }
|
---|
103 |
|
---|
104 | private void OnTemplateChanged(object sender, EventArgs args) {
|
---|
105 | SetupStructureTemplate();
|
---|
106 | }
|
---|
107 |
|
---|
108 | private void SetupStructureTemplate() {
|
---|
109 | foreach (var e in Encoding.Encodings.ToArray())
|
---|
110 | Encoding.Remove(e);
|
---|
111 |
|
---|
112 | foreach (var f in StructureTemplate.SubFunctions.Values) {
|
---|
113 | SetupVariables(f);
|
---|
114 | if (!Encoding.Encodings.Any(x => x.Name == f.Name)) // to prevent the same encoding twice
|
---|
115 | Encoding.Add(new SymbolicExpressionTreeEncoding(
|
---|
116 | f.Name,
|
---|
117 | f.Grammar,
|
---|
118 | f.MaximumSymbolicExpressionTreeLength,
|
---|
119 | f.MaximumSymbolicExpressionTreeDepth));
|
---|
120 | }
|
---|
121 | }
|
---|
122 |
|
---|
123 | public override void Analyze(Individual[] individuals, double[] qualities, ResultCollection results, IRandom random) {
|
---|
124 | base.Analyze(individuals, qualities, results, random);
|
---|
125 |
|
---|
126 | int bestIdx = 0;
|
---|
127 | double bestQuality = Maximization ? double.MinValue : double.MaxValue;
|
---|
128 | for(int idx = 0; idx < qualities.Length; ++idx) {
|
---|
129 | if((Maximization && qualities[idx] > bestQuality) ||
|
---|
130 | (!Maximization && qualities[idx] < bestQuality)) {
|
---|
131 | bestQuality = qualities[idx];
|
---|
132 | bestIdx = idx;
|
---|
133 | }
|
---|
134 | }
|
---|
135 |
|
---|
136 | if (results.TryGetValue("Best Tree", out IResult result)) {
|
---|
137 | var tree = BuildTree(individuals[bestIdx]);
|
---|
138 | if (StructureTemplate.ApplyLinearScaling)
|
---|
139 | AdjustLinearScalingParams(tree, Interpreter);
|
---|
140 | result.Value = tree;
|
---|
141 | }
|
---|
142 | else {
|
---|
143 | var tree = BuildTree(individuals[bestIdx]);
|
---|
144 | if (StructureTemplate.ApplyLinearScaling)
|
---|
145 | AdjustLinearScalingParams(tree, Interpreter);
|
---|
146 | results.Add(new Result("Best Tree", tree));
|
---|
147 | }
|
---|
148 | }
|
---|
149 |
|
---|
150 | public override double Evaluate(Individual individual, IRandom random) {
|
---|
151 | var tree = BuildTree(individual);
|
---|
152 |
|
---|
153 | if (StructureTemplate.ApplyLinearScaling)
|
---|
154 | AdjustLinearScalingParams(tree, Interpreter);
|
---|
155 | var estimationInterval = ProblemData.VariableRanges.GetInterval(ProblemData.TargetVariable);
|
---|
156 | var quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(
|
---|
157 | Interpreter, tree,
|
---|
158 | estimationInterval.LowerBound, estimationInterval.UpperBound,
|
---|
159 | ProblemData, ProblemData.TrainingIndices, false);
|
---|
160 |
|
---|
161 | return quality;
|
---|
162 | }
|
---|
163 |
|
---|
164 | private void AdjustLinearScalingParams(ISymbolicExpressionTree tree, ISymbolicDataAnalysisExpressionTreeInterpreter interpreter) {
|
---|
165 | var offsetNode = tree.Root.GetSubtree(0).GetSubtree(0);
|
---|
166 | var scalingNode = offsetNode.Subtrees.Where(x => !(x is ConstantTreeNode)).First();
|
---|
167 |
|
---|
168 | var offsetConstantNode = (ConstantTreeNode)offsetNode.Subtrees.Where(x => x is ConstantTreeNode).First();
|
---|
169 | var scalingConstantNode = (ConstantTreeNode)scalingNode.Subtrees.Where(x => x is ConstantTreeNode).First();
|
---|
170 |
|
---|
171 | var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(tree, ProblemData.Dataset, ProblemData.TrainingIndices);
|
---|
172 | var targetValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices);
|
---|
173 |
|
---|
174 | OnlineLinearScalingParameterCalculator.Calculate(estimatedValues, targetValues, out double a, out double b, out OnlineCalculatorError error);
|
---|
175 | if(error == OnlineCalculatorError.None) {
|
---|
176 | offsetConstantNode.Value = a;
|
---|
177 | scalingConstantNode.Value = b;
|
---|
178 | }
|
---|
179 | }
|
---|
180 |
|
---|
181 | private ISymbolicExpressionTree BuildTree(Individual individual) {
|
---|
182 | if (StructureTemplate.Tree == null)
|
---|
183 | throw new ArgumentException("No structure template defined!");
|
---|
184 |
|
---|
185 | var templateTree = (ISymbolicExpressionTree)StructureTemplate.Tree.Clone();
|
---|
186 |
|
---|
187 | // build main tree
|
---|
188 | foreach (var n in templateTree.IterateNodesPrefix()) {
|
---|
189 | if (n.Symbol is SubFunctionSymbol) {
|
---|
190 | var subFunctionTreeNode = n as SubFunctionTreeNode;
|
---|
191 | var subFunctionTree = individual.SymbolicExpressionTree(subFunctionTreeNode.Name);
|
---|
192 |
|
---|
193 | // add new tree
|
---|
194 | var subTree = subFunctionTree.Root.GetSubtree(0) // Start
|
---|
195 | .GetSubtree(0); // Offset
|
---|
196 | subFunctionTreeNode.AddSubtree(subTree);
|
---|
197 | }
|
---|
198 | }
|
---|
199 | return templateTree;
|
---|
200 | }
|
---|
201 |
|
---|
202 | private void SetupVariables(SubFunction subFunction) {
|
---|
203 | var varSym = (Variable)subFunction.Grammar.GetSymbol("Variable");
|
---|
204 | if (varSym == null) {
|
---|
205 | varSym = new Variable();
|
---|
206 | subFunction.Grammar.AddSymbol(varSym);
|
---|
207 | }
|
---|
208 |
|
---|
209 | var allVariables = ProblemData.InputVariables.Select(x => x.Value);
|
---|
210 | var allInputs = allVariables.Where(x => x != ProblemData.TargetVariable);
|
---|
211 |
|
---|
212 | // set all variables
|
---|
213 | varSym.AllVariableNames = allVariables;
|
---|
214 |
|
---|
215 | // set all allowed variables
|
---|
216 | if (subFunction.Arguments.Contains("_")) {
|
---|
217 | varSym.VariableNames = allInputs;
|
---|
218 | } else {
|
---|
219 | var vars = new List<string>();
|
---|
220 | var exceptions = new List<Exception>();
|
---|
221 | foreach (var arg in subFunction.Arguments) {
|
---|
222 | if (allInputs.Contains(arg))
|
---|
223 | vars.Add(arg);
|
---|
224 | else
|
---|
225 | exceptions.Add(new ArgumentException($"The argument '{arg}' for sub-function '{subFunction.Name}' is not a valid variable."));
|
---|
226 | }
|
---|
227 | if (exceptions.Any())
|
---|
228 | throw new AggregateException(exceptions);
|
---|
229 | varSym.VariableNames = vars;
|
---|
230 | }
|
---|
231 |
|
---|
232 | varSym.Enabled = true;
|
---|
233 | }
|
---|
234 |
|
---|
235 | public void Load(IRegressionProblemData data) => ProblemData = data;
|
---|
236 | }
|
---|
237 | }
|
---|