[18061] | 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;
|
---|
[18062] | 13 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
[18075] | 14 | using HeuristicLab.PluginInfrastructure;
|
---|
[18061] | 15 |
|
---|
[18063] | 16 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
|
---|
[18061] | 17 | [StorableType("7464E84B-65CC-440A-91F0-9FA920D730F9")]
|
---|
[18063] | 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)]
|
---|
[18075] | 20 | public class StructuredSymbolicRegressionSingleObjectiveProblem : SingleObjectiveBasicProblem<MultiEncoding>, IRegressionProblem, IProblemInstanceConsumer<IRegressionProblemData> {
|
---|
[18061] | 21 |
|
---|
| 22 | #region Constants
|
---|
| 23 | private const string ProblemDataParameterName = "ProblemData";
|
---|
[18063] | 24 | private const string StructureTemplateParameterName = "Structure Template";
|
---|
[18075] | 25 | private const string InterpreterParameterName = "Interpreter";
|
---|
[18076] | 26 | private const string EstimationLimitsParameterName = "EstimationLimits";
|
---|
| 27 | private const string BestTrainingSolutionParameterName = "Best Training Solution";
|
---|
[18072] | 28 |
|
---|
[18076] | 29 | private const string SymbolicExpressionTreeName = "SymbolicExpressionTree";
|
---|
| 30 |
|
---|
| 31 | private const string StructureTemplateDescriptionText =
|
---|
[18072] | 32 | "Enter your expression as string in infix format into the empty input field.\n" +
|
---|
| 33 | "By checking the \"Apply Linear Scaling\" checkbox you can add the relevant scaling terms to your expression.\n" +
|
---|
| 34 | "After entering the expression click parse to build the tree.\n" +
|
---|
| 35 | "To edit the defined sub-functions, click on the coressponding colored node in the tree view.";
|
---|
[18061] | 36 | #endregion
|
---|
| 37 |
|
---|
[18072] | 38 | #region Parameters
|
---|
[18061] | 39 | public IValueParameter<IRegressionProblemData> ProblemDataParameter => (IValueParameter<IRegressionProblemData>)Parameters[ProblemDataParameterName];
|
---|
[18063] | 40 | public IFixedValueParameter<StructureTemplate> StructureTemplateParameter => (IFixedValueParameter<StructureTemplate>)Parameters[StructureTemplateParameterName];
|
---|
[18075] | 41 | public IValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter> InterpreterParameter => (IValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>)Parameters[InterpreterParameterName];
|
---|
[18076] | 42 | public IFixedValueParameter<DoubleLimit> EstimationLimitsParameter => (IFixedValueParameter<DoubleLimit>)Parameters[EstimationLimitsParameterName];
|
---|
| 43 | public IResultParameter<ISymbolicRegressionSolution> BestTrainingSolutionParameter => (IResultParameter<ISymbolicRegressionSolution>)Parameters[BestTrainingSolutionParameterName];
|
---|
[18061] | 44 | #endregion
|
---|
| 45 |
|
---|
| 46 | #region Properties
|
---|
[18076] | 47 | public IRegressionProblemData ProblemData {
|
---|
| 48 | get => ProblemDataParameter.Value;
|
---|
[18061] | 49 | set {
|
---|
| 50 | ProblemDataParameter.Value = value;
|
---|
| 51 | ProblemDataChanged?.Invoke(this, EventArgs.Empty);
|
---|
| 52 | }
|
---|
| 53 | }
|
---|
| 54 |
|
---|
[18075] | 55 | public StructureTemplate StructureTemplate => StructureTemplateParameter.Value;
|
---|
[18061] | 56 |
|
---|
[18075] | 57 | public ISymbolicDataAnalysisExpressionTreeInterpreter Interpreter => InterpreterParameter.Value;
|
---|
[18063] | 58 |
|
---|
[18061] | 59 | IParameter IDataAnalysisProblem.ProblemDataParameter => ProblemDataParameter;
|
---|
| 60 | IDataAnalysisProblemData IDataAnalysisProblem.ProblemData => ProblemData;
|
---|
| 61 |
|
---|
[18076] | 62 | public DoubleLimit EstimationLimits => EstimationLimitsParameter.Value;
|
---|
| 63 |
|
---|
[18066] | 64 | public override bool Maximization => true;
|
---|
[18061] | 65 | #endregion
|
---|
| 66 |
|
---|
| 67 | #region EventHandlers
|
---|
| 68 | public event EventHandler ProblemDataChanged;
|
---|
| 69 | #endregion
|
---|
| 70 |
|
---|
| 71 | #region Constructors & Cloning
|
---|
| 72 | public StructuredSymbolicRegressionSingleObjectiveProblem() {
|
---|
[18062] | 73 | var problemData = new ShapeConstrainedRegressionProblemData();
|
---|
[18076] | 74 | var targetInterval = problemData.VariableRanges.GetInterval(problemData.TargetVariable);
|
---|
| 75 | var estimationWidth = targetInterval.Width * 10;
|
---|
[18062] | 76 |
|
---|
[18076] | 77 |
|
---|
[18065] | 78 | var structureTemplate = new StructureTemplate();
|
---|
[18066] | 79 | structureTemplate.Changed += OnTemplateChanged;
|
---|
[18065] | 80 |
|
---|
[18075] | 81 | Parameters.Add(new ValueParameter<IRegressionProblemData>(
|
---|
[18076] | 82 | ProblemDataParameterName,
|
---|
[18075] | 83 | problemData));
|
---|
| 84 | Parameters.Add(new FixedValueParameter<StructureTemplate>(
|
---|
[18076] | 85 | StructureTemplateParameterName,
|
---|
| 86 | StructureTemplateDescriptionText,
|
---|
[18075] | 87 | structureTemplate));
|
---|
| 88 | Parameters.Add(new ValueParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>(
|
---|
[18076] | 89 | InterpreterParameterName,
|
---|
| 90 | new SymbolicDataAnalysisExpressionTreeInterpreter()) { Hidden = true });
|
---|
| 91 | Parameters.Add(new FixedValueParameter<DoubleLimit>(
|
---|
| 92 | EstimationLimitsParameterName,
|
---|
| 93 | new DoubleLimit(targetInterval.LowerBound - estimationWidth, targetInterval.UpperBound + estimationWidth)));
|
---|
| 94 | Parameters.Add(new ResultParameter<ISymbolicRegressionSolution>(BestTrainingSolutionParameterName, ""));
|
---|
[18075] | 95 |
|
---|
| 96 | ProblemDataParameter.ValueChanged += ProblemDataParameterValueChanged;
|
---|
[18076] | 97 |
|
---|
| 98 | Operators.Add(new SymbolicDataAnalysisVariableFrequencyAnalyzer());
|
---|
| 99 | Operators.Add(new MinAverageMaxSymbolicExpressionTreeLengthAnalyzer());
|
---|
| 100 | //TODO change to value lookup
|
---|
| 101 | //Operators.Add(new SymbolicExpressionTreeLengthAnalyzer());
|
---|
| 102 | Operators.Add(new SymbolicExpressionSymbolFrequencyAnalyzer());
|
---|
| 103 |
|
---|
[18061] | 104 | }
|
---|
| 105 |
|
---|
[18076] | 106 | public StructuredSymbolicRegressionSingleObjectiveProblem(StructuredSymbolicRegressionSingleObjectiveProblem original,
|
---|
| 107 | Cloner cloner) : base(original, cloner) { }
|
---|
[18061] | 108 |
|
---|
| 109 | [StorableConstructor]
|
---|
[18063] | 110 | protected StructuredSymbolicRegressionSingleObjectiveProblem(StorableConstructorFlag _) : base(_) { }
|
---|
[18065] | 111 | #endregion
|
---|
[18061] | 112 |
|
---|
[18065] | 113 | #region Cloning
|
---|
[18061] | 114 | public override IDeepCloneable Clone(Cloner cloner) =>
|
---|
| 115 | new StructuredSymbolicRegressionSingleObjectiveProblem(this, cloner);
|
---|
| 116 | #endregion
|
---|
| 117 |
|
---|
[18075] | 118 | private void ProblemDataParameterValueChanged(object sender, EventArgs e) {
|
---|
| 119 | StructureTemplate.Reset();
|
---|
| 120 | // InfoBox for Reset?
|
---|
| 121 | }
|
---|
| 122 |
|
---|
[18066] | 123 | private void OnTemplateChanged(object sender, EventArgs args) {
|
---|
[18068] | 124 | SetupStructureTemplate();
|
---|
| 125 | }
|
---|
| 126 |
|
---|
| 127 | private void SetupStructureTemplate() {
|
---|
[18066] | 128 | foreach (var e in Encoding.Encodings.ToArray())
|
---|
| 129 | Encoding.Remove(e);
|
---|
| 130 |
|
---|
[18068] | 131 | foreach (var f in StructureTemplate.SubFunctions.Values) {
|
---|
| 132 | SetupVariables(f);
|
---|
[18075] | 133 | if (!Encoding.Encodings.Any(x => x.Name == f.Name)) // to prevent the same encoding twice
|
---|
| 134 | Encoding.Add(new SymbolicExpressionTreeEncoding(
|
---|
[18076] | 135 | f.Name,
|
---|
| 136 | f.Grammar,
|
---|
| 137 | f.MaximumSymbolicExpressionTreeLength,
|
---|
[18075] | 138 | f.MaximumSymbolicExpressionTreeDepth));
|
---|
[18066] | 139 | }
|
---|
| 140 | }
|
---|
| 141 |
|
---|
| 142 | public override void Analyze(Individual[] individuals, double[] qualities, ResultCollection results, IRandom random) {
|
---|
| 143 | base.Analyze(individuals, qualities, results, random);
|
---|
| 144 |
|
---|
[18076] | 145 | var orderedIndividuals = individuals.Zip(qualities, (i, q) => new { Individual = i, Quality = q }).OrderBy(z => z.Quality);
|
---|
| 146 | var best = Maximization ? orderedIndividuals.Last().Individual : orderedIndividuals.First().Individual;
|
---|
| 147 |
|
---|
| 148 | if (!results.ContainsKey(BestTrainingSolutionParameter.ActualName)) {
|
---|
| 149 | results.Add(new Result(BestTrainingSolutionParameter.ActualName, typeof(SymbolicRegressionSolution)));
|
---|
[18066] | 150 | }
|
---|
| 151 |
|
---|
[18076] | 152 | var tree = (ISymbolicExpressionTree)best[SymbolicExpressionTreeName];
|
---|
| 153 |
|
---|
| 154 | var model = new SymbolicRegressionModel(ProblemData.TargetVariable, tree, Interpreter);
|
---|
| 155 | var solution = model.CreateRegressionSolution(ProblemData);
|
---|
| 156 |
|
---|
| 157 | results[BestTrainingSolutionParameter.ActualName].Value = solution;
|
---|
[18066] | 158 | }
|
---|
| 159 |
|
---|
[18076] | 160 |
|
---|
[18065] | 161 | public override double Evaluate(Individual individual, IRandom random) {
|
---|
[18066] | 162 | var tree = BuildTree(individual);
|
---|
[18071] | 163 |
|
---|
[18072] | 164 | if (StructureTemplate.ApplyLinearScaling)
|
---|
[18076] | 165 | AdjustLinearScalingParams(ProblemData, tree, Interpreter);
|
---|
| 166 |
|
---|
| 167 | individual[SymbolicExpressionTreeName] = tree;
|
---|
| 168 |
|
---|
[18066] | 169 | var quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(
|
---|
[18076] | 170 | Interpreter, tree,
|
---|
| 171 | EstimationLimits.Lower, EstimationLimits.Upper,
|
---|
[18066] | 172 | ProblemData, ProblemData.TrainingIndices, false);
|
---|
[18076] | 173 |
|
---|
[18066] | 174 | return quality;
|
---|
| 175 | }
|
---|
| 176 |
|
---|
[18076] | 177 | private static void AdjustLinearScalingParams(IRegressionProblemData problemData, ISymbolicExpressionTree tree, ISymbolicDataAnalysisExpressionTreeInterpreter interpreter) {
|
---|
[18071] | 178 | var offsetNode = tree.Root.GetSubtree(0).GetSubtree(0);
|
---|
| 179 | var scalingNode = offsetNode.Subtrees.Where(x => !(x is ConstantTreeNode)).First();
|
---|
| 180 |
|
---|
| 181 | var offsetConstantNode = (ConstantTreeNode)offsetNode.Subtrees.Where(x => x is ConstantTreeNode).First();
|
---|
| 182 | var scalingConstantNode = (ConstantTreeNode)scalingNode.Subtrees.Where(x => x is ConstantTreeNode).First();
|
---|
| 183 |
|
---|
[18076] | 184 | var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, problemData.TrainingIndices);
|
---|
| 185 | var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices);
|
---|
[18071] | 186 |
|
---|
| 187 | OnlineLinearScalingParameterCalculator.Calculate(estimatedValues, targetValues, out double a, out double b, out OnlineCalculatorError error);
|
---|
[18076] | 188 | if (error == OnlineCalculatorError.None) {
|
---|
[18071] | 189 | offsetConstantNode.Value = a;
|
---|
| 190 | scalingConstantNode.Value = b;
|
---|
| 191 | }
|
---|
| 192 | }
|
---|
| 193 |
|
---|
[18066] | 194 | private ISymbolicExpressionTree BuildTree(Individual individual) {
|
---|
[18075] | 195 | if (StructureTemplate.Tree == null)
|
---|
| 196 | throw new ArgumentException("No structure template defined!");
|
---|
| 197 |
|
---|
[18065] | 198 | var templateTree = (ISymbolicExpressionTree)StructureTemplate.Tree.Clone();
|
---|
[18066] | 199 |
|
---|
[18065] | 200 | // build main tree
|
---|
[18076] | 201 | foreach (var subFunctionTreeNode in templateTree.IterateNodesPrefix().OfType<SubFunctionTreeNode>()) {
|
---|
| 202 | var subFunctionTree = individual.SymbolicExpressionTree(subFunctionTreeNode.Name);
|
---|
[18062] | 203 |
|
---|
[18076] | 204 | // add new tree
|
---|
| 205 | var subTree = subFunctionTree.Root.GetSubtree(0) // Start
|
---|
| 206 | .GetSubtree(0); // Offset
|
---|
| 207 | subTree = (ISymbolicExpressionTreeNode)subTree.Clone();
|
---|
| 208 | subFunctionTreeNode.AddSubtree(subTree);
|
---|
| 209 |
|
---|
[18065] | 210 | }
|
---|
[18066] | 211 | return templateTree;
|
---|
[18061] | 212 | }
|
---|
| 213 |
|
---|
[18068] | 214 | private void SetupVariables(SubFunction subFunction) {
|
---|
| 215 | var varSym = (Variable)subFunction.Grammar.GetSymbol("Variable");
|
---|
| 216 | if (varSym == null) {
|
---|
| 217 | varSym = new Variable();
|
---|
| 218 | subFunction.Grammar.AddSymbol(varSym);
|
---|
| 219 | }
|
---|
| 220 |
|
---|
| 221 | var allVariables = ProblemData.InputVariables.Select(x => x.Value);
|
---|
| 222 | var allInputs = allVariables.Where(x => x != ProblemData.TargetVariable);
|
---|
| 223 |
|
---|
| 224 | // set all variables
|
---|
| 225 | varSym.AllVariableNames = allVariables;
|
---|
| 226 |
|
---|
| 227 | // set all allowed variables
|
---|
| 228 | if (subFunction.Arguments.Contains("_")) {
|
---|
| 229 | varSym.VariableNames = allInputs;
|
---|
| 230 | } else {
|
---|
| 231 | var vars = new List<string>();
|
---|
| 232 | var exceptions = new List<Exception>();
|
---|
| 233 | foreach (var arg in subFunction.Arguments) {
|
---|
| 234 | if (allInputs.Contains(arg))
|
---|
| 235 | vars.Add(arg);
|
---|
| 236 | else
|
---|
| 237 | exceptions.Add(new ArgumentException($"The argument '{arg}' for sub-function '{subFunction.Name}' is not a valid variable."));
|
---|
| 238 | }
|
---|
| 239 | if (exceptions.Any())
|
---|
| 240 | throw new AggregateException(exceptions);
|
---|
| 241 | varSym.VariableNames = vars;
|
---|
| 242 | }
|
---|
| 243 |
|
---|
| 244 | varSym.Enabled = true;
|
---|
| 245 | }
|
---|
| 246 |
|
---|
[18075] | 247 | public void Load(IRegressionProblemData data) => ProblemData = data;
|
---|
[18061] | 248 | }
|
---|
| 249 | }
|
---|