| 1 | #region License Information
|
|---|
| 2 | /* HeuristicLab
|
|---|
| 3 | * Copyright (C) 2002-2010 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 |
|
|---|
| 22 | using System;
|
|---|
| 23 | using System.Collections.Generic;
|
|---|
| 24 | using System.Linq;
|
|---|
| 25 | using HeuristicLab.Common;
|
|---|
| 26 | using HeuristicLab.Core;
|
|---|
| 27 | using HeuristicLab.Data;
|
|---|
| 28 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
|---|
| 29 | using HeuristicLab.Parameters;
|
|---|
| 30 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
|---|
| 31 |
|
|---|
| 32 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
|
|---|
| 33 | [Item("Constant Optimization Evaluator", "Calculates Pearson R² of a symbolic regression solution and optimizes the constant used.")]
|
|---|
| 34 | [StorableClass]
|
|---|
| 35 | public class SymbolicRegressionConstantOptimizationEvaluator : SymbolicRegressionSingleObjectiveEvaluator {
|
|---|
| 36 | private const string ConstantOptimizationIterationsParameterName = "ConstantOptimizationIterations";
|
|---|
| 37 | private const string ConstantOptimizationImprovementParameterName = "ConstantOptimizationImprovement";
|
|---|
| 38 | private const string ConstantOptimizationProbabilityParameterName = "ConstantOptimizationProbability";
|
|---|
| 39 | private const string ConstantOptimizationRowsPercentageParameterName = "ConstantOptimizationRowsPercentage";
|
|---|
| 40 |
|
|---|
| 41 | private const string EvaluatedTreesResultName = "EvaluatedTrees";
|
|---|
| 42 | private const string EvaluatedTreeNodesResultName = "EvaluatedTreeNodes";
|
|---|
| 43 |
|
|---|
| 44 | public ILookupParameter<IntValue> EvaluatedTreesParameter {
|
|---|
| 45 | get { return (ILookupParameter<IntValue>)Parameters[EvaluatedTreesResultName]; }
|
|---|
| 46 | }
|
|---|
| 47 | public ILookupParameter<IntValue> EvaluatedTreeNodesParameter {
|
|---|
| 48 | get { return (ILookupParameter<IntValue>)Parameters[EvaluatedTreeNodesResultName]; }
|
|---|
| 49 | }
|
|---|
| 50 |
|
|---|
| 51 | public IFixedValueParameter<IntValue> ConstantOptimizationIterationsParameter {
|
|---|
| 52 | get { return (IFixedValueParameter<IntValue>)Parameters[ConstantOptimizationIterationsParameterName]; }
|
|---|
| 53 | }
|
|---|
| 54 | public IFixedValueParameter<DoubleValue> ConstantOptimizationImprovementParameter {
|
|---|
| 55 | get { return (IFixedValueParameter<DoubleValue>)Parameters[ConstantOptimizationImprovementParameterName]; }
|
|---|
| 56 | }
|
|---|
| 57 | public IFixedValueParameter<PercentValue> ConstantOptimizationProbabilityParameter {
|
|---|
| 58 | get { return (IFixedValueParameter<PercentValue>)Parameters[ConstantOptimizationProbabilityParameterName]; }
|
|---|
| 59 | }
|
|---|
| 60 | public IFixedValueParameter<PercentValue> ConstantOptimizationRowsPercentageParameter {
|
|---|
| 61 | get { return (IFixedValueParameter<PercentValue>)Parameters[ConstantOptimizationRowsPercentageParameterName]; }
|
|---|
| 62 | }
|
|---|
| 63 |
|
|---|
| 64 | public IntValue ConstantOptimizationIterations {
|
|---|
| 65 | get { return ConstantOptimizationIterationsParameter.Value; }
|
|---|
| 66 | }
|
|---|
| 67 | public DoubleValue ConstantOptimizationImprovement {
|
|---|
| 68 | get { return ConstantOptimizationImprovementParameter.Value; }
|
|---|
| 69 | }
|
|---|
| 70 | public PercentValue ConstantOptimizationProbability {
|
|---|
| 71 | get { return ConstantOptimizationProbabilityParameter.Value; }
|
|---|
| 72 | }
|
|---|
| 73 | public PercentValue ConstantOptimizationRowsPercentage {
|
|---|
| 74 | get { return ConstantOptimizationRowsPercentageParameter.Value; }
|
|---|
| 75 | }
|
|---|
| 76 |
|
|---|
| 77 | public override bool Maximization {
|
|---|
| 78 | get { return true; }
|
|---|
| 79 | }
|
|---|
| 80 |
|
|---|
| 81 | [StorableConstructor]
|
|---|
| 82 | protected SymbolicRegressionConstantOptimizationEvaluator(bool deserializing) : base(deserializing) { }
|
|---|
| 83 | protected SymbolicRegressionConstantOptimizationEvaluator(SymbolicRegressionConstantOptimizationEvaluator original, Cloner cloner)
|
|---|
| 84 | : base(original, cloner) {
|
|---|
| 85 | }
|
|---|
| 86 | public SymbolicRegressionConstantOptimizationEvaluator()
|
|---|
| 87 | : base() {
|
|---|
| 88 | Parameters.Add(new FixedValueParameter<IntValue>(ConstantOptimizationIterationsParameterName, "Determines how many iterations should be calculated while optimizing the constant of a symbolic expression tree (0 indicates other or default stopping criterion).", new IntValue(3), true));
|
|---|
| 89 | Parameters.Add(new FixedValueParameter<DoubleValue>(ConstantOptimizationImprovementParameterName, "Determines the relative improvement which must be achieved in the constant optimization to continue with it (0 indicates other or default stopping criterion).", new DoubleValue(0), true));
|
|---|
| 90 | Parameters.Add(new FixedValueParameter<PercentValue>(ConstantOptimizationProbabilityParameterName, "Determines the probability that the constants are optimized", new PercentValue(1), true));
|
|---|
| 91 | Parameters.Add(new FixedValueParameter<PercentValue>(ConstantOptimizationRowsPercentageParameterName, "Determines the percentage of the rows which should be used for constant optimization", new PercentValue(1), true));
|
|---|
| 92 |
|
|---|
| 93 | Parameters.Add(new LookupParameter<IntValue>(EvaluatedTreesResultName));
|
|---|
| 94 | Parameters.Add(new LookupParameter<IntValue>(EvaluatedTreeNodesResultName));
|
|---|
| 95 | }
|
|---|
| 96 |
|
|---|
| 97 | public override IDeepCloneable Clone(Cloner cloner) {
|
|---|
| 98 | return new SymbolicRegressionConstantOptimizationEvaluator(this, cloner);
|
|---|
| 99 | }
|
|---|
| 100 |
|
|---|
| 101 | public override IOperation Apply() {
|
|---|
| 102 | AddResults();
|
|---|
| 103 | int seed = RandomParameter.ActualValue.Next();
|
|---|
| 104 | var solution = SymbolicExpressionTreeParameter.ActualValue;
|
|---|
| 105 | double quality;
|
|---|
| 106 | if (RandomParameter.ActualValue.NextDouble() < ConstantOptimizationProbability.Value) {
|
|---|
| 107 | IEnumerable<int> constantOptimizationRows = GenerateRowsToEvaluate(ConstantOptimizationRowsPercentage.Value);
|
|---|
| 108 | quality = OptimizeConstants(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, ProblemDataParameter.ActualValue,
|
|---|
| 109 | constantOptimizationRows, ConstantOptimizationImprovement.Value, ConstantOptimizationIterations.Value, 0.001,
|
|---|
| 110 | EstimationLimitsParameter.ActualValue.Upper, EstimationLimitsParameter.ActualValue.Lower,
|
|---|
| 111 | EvaluatedTreesParameter.ActualValue, EvaluatedTreeNodesParameter.ActualValue);
|
|---|
| 112 | if (ConstantOptimizationRowsPercentage.Value != RelativeNumberOfEvaluatedSamplesParameter.ActualValue.Value) {
|
|---|
| 113 | var evaluationRows = GenerateRowsToEvaluate();
|
|---|
| 114 | quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, evaluationRows);
|
|---|
| 115 | }
|
|---|
| 116 | } else {
|
|---|
| 117 | var evaluationRows = GenerateRowsToEvaluate();
|
|---|
| 118 | quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, evaluationRows);
|
|---|
| 119 | }
|
|---|
| 120 | QualityParameter.ActualValue = new DoubleValue(quality);
|
|---|
| 121 | EvaluatedTreesParameter.ActualValue.Value += 1;
|
|---|
| 122 | EvaluatedTreeNodesParameter.ActualValue.Value += solution.Length;
|
|---|
| 123 |
|
|---|
| 124 | if (Successor != null)
|
|---|
| 125 | return ExecutionContext.CreateOperation(Successor);
|
|---|
| 126 | else
|
|---|
| 127 | return null;
|
|---|
| 128 | }
|
|---|
| 129 |
|
|---|
| 130 | private void AddResults() {
|
|---|
| 131 | if (EvaluatedTreesParameter.ActualValue == null) {
|
|---|
| 132 | var scope = ExecutionContext.Scope;
|
|---|
| 133 | while (scope.Parent != null)
|
|---|
| 134 | scope = scope.Parent;
|
|---|
| 135 | scope.Variables.Add(new Core.Variable(EvaluatedTreesResultName, new IntValue()));
|
|---|
| 136 | }
|
|---|
| 137 | if (EvaluatedTreeNodesParameter.ActualValue == null) {
|
|---|
| 138 | var scope = ExecutionContext.Scope;
|
|---|
| 139 | while (scope.Parent != null)
|
|---|
| 140 | scope = scope.Parent;
|
|---|
| 141 | scope.Variables.Add(new Core.Variable(EvaluatedTreeNodesResultName, new IntValue()));
|
|---|
| 142 | }
|
|---|
| 143 | }
|
|---|
| 144 |
|
|---|
| 145 | public override double Evaluate(IExecutionContext context, ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows) {
|
|---|
| 146 | SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
|
|---|
| 147 | EstimationLimitsParameter.ExecutionContext = context;
|
|---|
| 148 |
|
|---|
| 149 | double r2 = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, problemData, rows);
|
|---|
| 150 |
|
|---|
| 151 | SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
|
|---|
| 152 | EstimationLimitsParameter.ExecutionContext = null;
|
|---|
| 153 |
|
|---|
| 154 | return r2;
|
|---|
| 155 | }
|
|---|
| 156 |
|
|---|
| 157 | public static double OptimizeConstants(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree tree, IRegressionProblemData problemData,
|
|---|
| 158 | IEnumerable<int> rows, double improvement, int iterations, double differentialStep, double upperEstimationLimit = double.MaxValue, double lowerEstimationLimit = double.MinValue, IntValue evaluatedTrees = null, IntValue evaluatedTreeNodes = null) {
|
|---|
| 159 |
|
|---|
| 160 | List<AutoDiff.Variable> variables = new List<AutoDiff.Variable>();
|
|---|
| 161 | List<AutoDiff.Variable> parameters = new List<AutoDiff.Variable>();
|
|---|
| 162 | List<string> variableNames = new List<string>();
|
|---|
| 163 |
|
|---|
| 164 | var func = TransformToAutoDiff(tree.Root.GetSubtree(0), variables, parameters, variableNames);
|
|---|
| 165 | if (variableNames.Count == 0) return 0.0;
|
|---|
| 166 |
|
|---|
| 167 |
|
|---|
| 168 | AutoDiff.IParametricCompiledTerm compiledFunc = AutoDiff.TermUtils.Compile(func, variables.ToArray(), parameters.ToArray());
|
|---|
| 169 |
|
|---|
| 170 | List<SymbolicExpressionTreeTerminalNode> terminalNodes = tree.Root.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>().ToList();
|
|---|
| 171 | double[] c = new double[variables.Count];
|
|---|
| 172 |
|
|---|
| 173 | c[0] = 0.0;
|
|---|
| 174 | c[1] = 1.0;
|
|---|
| 175 | //extract inital constants
|
|---|
| 176 | for (int i = 0; i < terminalNodes.Count; i++) {
|
|---|
| 177 | ConstantTreeNode constantTreeNode = terminalNodes[i] as ConstantTreeNode;
|
|---|
| 178 | if (constantTreeNode != null) c[i + 2] = constantTreeNode.Value;
|
|---|
| 179 | VariableTreeNode variableTreeNode = terminalNodes[i] as VariableTreeNode;
|
|---|
| 180 | if (variableTreeNode != null) c[i + 2] = variableTreeNode.Weight;
|
|---|
| 181 | }
|
|---|
| 182 |
|
|---|
| 183 | double epsg = 0;
|
|---|
| 184 | double epsf = improvement;
|
|---|
| 185 | double epsx = 0;
|
|---|
| 186 | int maxits = iterations;
|
|---|
| 187 | double diffstep = differentialStep;
|
|---|
| 188 |
|
|---|
| 189 | alglib.lsfitstate state;
|
|---|
| 190 | alglib.lsfitreport rep;
|
|---|
| 191 | int info;
|
|---|
| 192 |
|
|---|
| 193 | Dataset ds = problemData.Dataset;
|
|---|
| 194 | double[,] x = new double[rows.Count(), variableNames.Count];
|
|---|
| 195 | int row = 0;
|
|---|
| 196 | foreach (var r in rows) {
|
|---|
| 197 | for (int col = 0; col < variableNames.Count; col++) {
|
|---|
| 198 | x[row, col] = ds.GetDoubleValue(variableNames[col], r);
|
|---|
| 199 | }
|
|---|
| 200 | row++;
|
|---|
| 201 | }
|
|---|
| 202 | double[] y = ds.GetDoubleValues(problemData.TargetVariable, rows).ToArray();
|
|---|
| 203 | int n = x.GetLength(0);
|
|---|
| 204 | int m = x.GetLength(1);
|
|---|
| 205 | int k = c.Length;
|
|---|
| 206 |
|
|---|
| 207 | alglib.ndimensional_pfunc function_cx_1_func = CreatePFunc(compiledFunc, variables, parameters);
|
|---|
| 208 | alglib.ndimensional_pgrad function_cx_1_grad = CreatePGrad(compiledFunc, variables, parameters);
|
|---|
| 209 |
|
|---|
| 210 | alglib.lsfitcreatefg(x, y, c, n, m, k, false, out state);
|
|---|
| 211 | alglib.lsfitsetcond(state, epsf, epsx, maxits);
|
|---|
| 212 | alglib.lsfitfit(state, function_cx_1_func, function_cx_1_grad, null, null);
|
|---|
| 213 | alglib.lsfitresults(state, out info, out c, out rep);
|
|---|
| 214 | //alglib.minlmstate state;
|
|---|
| 215 | //alglib.minlmreport report;
|
|---|
| 216 |
|
|---|
| 217 | //alglib.minlmcreatev(1, c, diffstep, out state);
|
|---|
| 218 | //alglib.minlmsetcond(state, epsg, epsf, epsx, maxits);
|
|---|
| 219 | //alglib.minlmoptimize(state, CreateCallBack(interpreter, tree, problemData, rows, upperEstimationLimit, lowerEstimationLimit, treeLength, evaluatedTrees, evaluatedTreeNodes), null, terminalNodes);
|
|---|
| 220 | //alglib.minlmresults(state, out c, out report);
|
|---|
| 221 |
|
|---|
| 222 | for (int i = 2; i < c.Length; i++) {
|
|---|
| 223 | ConstantTreeNode constantTreeNode = terminalNodes[i - 2] as ConstantTreeNode;
|
|---|
| 224 | if (constantTreeNode != null) constantTreeNode.Value = c[i];
|
|---|
| 225 | VariableTreeNode variableTreeNode = terminalNodes[i - 2] as VariableTreeNode;
|
|---|
| 226 | if (variableTreeNode != null) variableTreeNode.Weight = c[i];
|
|---|
| 227 | }
|
|---|
| 228 |
|
|---|
| 229 | return SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows);
|
|---|
| 230 |
|
|---|
| 231 | }
|
|---|
| 232 |
|
|---|
| 233 | private static alglib.ndimensional_pfunc CreatePFunc(AutoDiff.IParametricCompiledTerm compiledFunc, List<AutoDiff.Variable> variables, List<AutoDiff.Variable> parameters) {
|
|---|
| 234 | return (double[] c, double[] x, ref double func, object o) => {
|
|---|
| 235 | func = compiledFunc.Evaluate(c, x);
|
|---|
| 236 | };
|
|---|
| 237 | }
|
|---|
| 238 |
|
|---|
| 239 | private static alglib.ndimensional_pgrad CreatePGrad(AutoDiff.IParametricCompiledTerm compiledFunc, List<AutoDiff.Variable> variables, List<AutoDiff.Variable> parameters) {
|
|---|
| 240 | return (double[] c, double[] x, ref double func, double[] grad, object o) => {
|
|---|
| 241 | var tupel = compiledFunc.Differentiate(c, x);
|
|---|
| 242 | func = tupel.Item2;
|
|---|
| 243 | Array.Copy(tupel.Item1, grad, grad.Length);
|
|---|
| 244 | };
|
|---|
| 245 | }
|
|---|
| 246 |
|
|---|
| 247 | private static AutoDiff.Term TransformToAutoDiff(ISymbolicExpressionTreeNode node, List<AutoDiff.Variable> variables, List<AutoDiff.Variable> parameters, List<string> variableNames) {
|
|---|
| 248 | if (node.Symbol is Constant) {
|
|---|
| 249 | var var = new AutoDiff.Variable();
|
|---|
| 250 | variables.Add(var);
|
|---|
| 251 | return var;
|
|---|
| 252 | }
|
|---|
| 253 | if (node.Symbol is Variable) {
|
|---|
| 254 | var w = new AutoDiff.Variable();
|
|---|
| 255 | variables.Add(w);
|
|---|
| 256 | var par = new AutoDiff.Variable();
|
|---|
| 257 | parameters.Add(par);
|
|---|
| 258 | variableNames.Add((node as VariableTreeNode).VariableName);
|
|---|
| 259 | return AutoDiff.TermBuilder.Product(w, par);
|
|---|
| 260 | }
|
|---|
| 261 | if (node.Symbol is Addition)
|
|---|
| 262 | return AutoDiff.TermBuilder.Sum(
|
|---|
| 263 | from subTree in node.Subtrees
|
|---|
| 264 | select TransformToAutoDiff(subTree, variables, parameters, variableNames)
|
|---|
| 265 | );
|
|---|
| 266 | if (node.Symbol is Multiplication)
|
|---|
| 267 | return AutoDiff.TermBuilder.Product(
|
|---|
| 268 | TransformToAutoDiff(node.GetSubtree(0), variables, parameters, variableNames),
|
|---|
| 269 | TransformToAutoDiff(node.GetSubtree(1), variables, parameters, variableNames),
|
|---|
| 270 | (from subTree in node.Subtrees.Skip(2)
|
|---|
| 271 | select TransformToAutoDiff(subTree, variables, parameters, variableNames)).ToArray()
|
|---|
| 272 | );
|
|---|
| 273 | if (node.Symbol is Logarithm)
|
|---|
| 274 | return AutoDiff.TermBuilder.Log(
|
|---|
| 275 | TransformToAutoDiff(node.GetSubtree(0), variables, parameters, variableNames));
|
|---|
| 276 | if (node.Symbol is Exponential)
|
|---|
| 277 | return AutoDiff.TermBuilder.Exp(
|
|---|
| 278 | TransformToAutoDiff(node.GetSubtree(0), variables, parameters, variableNames));
|
|---|
| 279 | if (node.Symbol is StartSymbol) {
|
|---|
| 280 | var alpha = new AutoDiff.Variable();
|
|---|
| 281 | var beta = new AutoDiff.Variable();
|
|---|
| 282 | variables.Add(beta);
|
|---|
| 283 | variables.Add(alpha);
|
|---|
| 284 | return TransformToAutoDiff(node.GetSubtree(0), variables, parameters, variableNames) * alpha + beta;
|
|---|
| 285 | }
|
|---|
| 286 | throw new NotImplementedException();
|
|---|
| 287 | }
|
|---|
| 288 |
|
|---|
| 289 | private static alglib.ndimensional_fvec CreateCallBack(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows, double upperEstimationLimit, double lowerEstimationLimit, int treeLength, IntValue evaluatedTrees = null, IntValue evaluatedTreeNodes = null) {
|
|---|
| 290 | return (double[] arg, double[] fi, object obj) => {
|
|---|
| 291 | // update constants of tree
|
|---|
| 292 | List<SymbolicExpressionTreeTerminalNode> terminalNodes = (List<SymbolicExpressionTreeTerminalNode>)obj;
|
|---|
| 293 | for (int i = 0; i < terminalNodes.Count; i++) {
|
|---|
| 294 | ConstantTreeNode constantTreeNode = terminalNodes[i] as ConstantTreeNode;
|
|---|
| 295 | if (constantTreeNode != null) constantTreeNode.Value = arg[i];
|
|---|
| 296 | VariableTreeNode variableTreeNode = terminalNodes[i] as VariableTreeNode;
|
|---|
| 297 | if (variableTreeNode != null) variableTreeNode.Weight = arg[i];
|
|---|
| 298 | }
|
|---|
| 299 |
|
|---|
| 300 | double quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows);
|
|---|
| 301 |
|
|---|
| 302 | fi[0] = 1 - quality;
|
|---|
| 303 | if (evaluatedTrees != null) evaluatedTrees.Value++;
|
|---|
| 304 | if (evaluatedTreeNodes != null) evaluatedTreeNodes.Value += treeLength;
|
|---|
| 305 | };
|
|---|
| 306 | }
|
|---|
| 307 |
|
|---|
| 308 | }
|
|---|
| 309 | }
|
|---|
