[14024] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
[15583] | 3 | * Copyright (C) 2002-2018 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[14024] | 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;
|
---|
[14826] | 23 | using System.Collections.Generic;
|
---|
[14024] | 24 | using System.Linq;
|
---|
[14523] | 25 | using System.Threading;
|
---|
[14319] | 26 | using HeuristicLab.Analysis;
|
---|
[14024] | 27 | using HeuristicLab.Common;
|
---|
| 28 | using HeuristicLab.Core;
|
---|
| 29 | using HeuristicLab.Data;
|
---|
[14316] | 30 | using HeuristicLab.Optimization;
|
---|
[14024] | 31 | using HeuristicLab.Parameters;
|
---|
| 32 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
| 33 | using HeuristicLab.Problems.DataAnalysis;
|
---|
| 34 | using HeuristicLab.Problems.DataAnalysis.Symbolic;
|
---|
| 35 | using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
|
---|
[14258] | 36 | using HeuristicLab.Random;
|
---|
[14024] | 37 |
|
---|
| 38 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
| 39 | /// <summary>
|
---|
| 40 | /// Nonlinear regression data analysis algorithm.
|
---|
| 41 | /// </summary>
|
---|
| 42 | [Item("Nonlinear Regression (NLR)", "Nonlinear regression (curve fitting) data analysis algorithm (wrapper for ALGLIB).")]
|
---|
| 43 | [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 120)]
|
---|
| 44 | [StorableClass]
|
---|
| 45 | public sealed class NonlinearRegression : FixedDataAnalysisAlgorithm<IRegressionProblem> {
|
---|
[14109] | 46 | private const string RegressionSolutionResultName = "Regression solution";
|
---|
[14024] | 47 | private const string ModelStructureParameterName = "Model structure";
|
---|
| 48 | private const string IterationsParameterName = "Iterations";
|
---|
[14258] | 49 | private const string RestartsParameterName = "Restarts";
|
---|
| 50 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
---|
| 51 | private const string SeedParameterName = "Seed";
|
---|
[14319] | 52 | private const string InitParamsRandomlyParameterName = "InitializeParametersRandomly";
|
---|
[15502] | 53 | private const string ApplyLinearScalingParameterName = "Apply linear scaling";
|
---|
[14024] | 54 |
|
---|
| 55 | public IFixedValueParameter<StringValue> ModelStructureParameter {
|
---|
| 56 | get { return (IFixedValueParameter<StringValue>)Parameters[ModelStructureParameterName]; }
|
---|
| 57 | }
|
---|
| 58 | public IFixedValueParameter<IntValue> IterationsParameter {
|
---|
| 59 | get { return (IFixedValueParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
| 60 | }
|
---|
| 61 |
|
---|
[14258] | 62 | public IFixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
---|
| 63 | get { return (IFixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
---|
| 64 | }
|
---|
| 65 |
|
---|
| 66 | public IFixedValueParameter<IntValue> SeedParameter {
|
---|
| 67 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
---|
| 68 | }
|
---|
| 69 |
|
---|
| 70 | public IFixedValueParameter<IntValue> RestartsParameter {
|
---|
| 71 | get { return (IFixedValueParameter<IntValue>)Parameters[RestartsParameterName]; }
|
---|
| 72 | }
|
---|
| 73 |
|
---|
[14319] | 74 | public IFixedValueParameter<BoolValue> InitParametersRandomlyParameter {
|
---|
| 75 | get { return (IFixedValueParameter<BoolValue>)Parameters[InitParamsRandomlyParameterName]; }
|
---|
| 76 | }
|
---|
| 77 |
|
---|
[15502] | 78 | public IFixedValueParameter<BoolValue> ApplyLinearScalingParameter {
|
---|
| 79 | get { return (IFixedValueParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
|
---|
| 80 | }
|
---|
| 81 |
|
---|
[14024] | 82 | public string ModelStructure {
|
---|
| 83 | get { return ModelStructureParameter.Value.Value; }
|
---|
| 84 | set { ModelStructureParameter.Value.Value = value; }
|
---|
| 85 | }
|
---|
| 86 |
|
---|
| 87 | public int Iterations {
|
---|
| 88 | get { return IterationsParameter.Value.Value; }
|
---|
| 89 | set { IterationsParameter.Value.Value = value; }
|
---|
| 90 | }
|
---|
| 91 |
|
---|
[14258] | 92 | public int Restarts {
|
---|
| 93 | get { return RestartsParameter.Value.Value; }
|
---|
| 94 | set { RestartsParameter.Value.Value = value; }
|
---|
| 95 | }
|
---|
[14024] | 96 |
|
---|
[14258] | 97 | public int Seed {
|
---|
| 98 | get { return SeedParameter.Value.Value; }
|
---|
| 99 | set { SeedParameter.Value.Value = value; }
|
---|
| 100 | }
|
---|
| 101 |
|
---|
| 102 | public bool SetSeedRandomly {
|
---|
| 103 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
| 104 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
| 105 | }
|
---|
| 106 |
|
---|
[14319] | 107 | public bool InitializeParametersRandomly {
|
---|
| 108 | get { return InitParametersRandomlyParameter.Value.Value; }
|
---|
| 109 | set { InitParametersRandomlyParameter.Value.Value = value; }
|
---|
| 110 | }
|
---|
| 111 |
|
---|
[15502] | 112 | public bool ApplyLinearScaling {
|
---|
| 113 | get { return ApplyLinearScalingParameter.Value.Value; }
|
---|
| 114 | set { ApplyLinearScalingParameter.Value.Value = value; }
|
---|
| 115 | }
|
---|
| 116 |
|
---|
[14024] | 117 | [StorableConstructor]
|
---|
| 118 | private NonlinearRegression(bool deserializing) : base(deserializing) { }
|
---|
| 119 | private NonlinearRegression(NonlinearRegression original, Cloner cloner)
|
---|
| 120 | : base(original, cloner) {
|
---|
| 121 | }
|
---|
| 122 | public NonlinearRegression()
|
---|
| 123 | : base() {
|
---|
| 124 | Problem = new RegressionProblem();
|
---|
| 125 | Parameters.Add(new FixedValueParameter<StringValue>(ModelStructureParameterName, "The function for which the parameters must be fit (only numeric constants are tuned).", new StringValue("1.0 * x*x + 0.0")));
|
---|
| 126 | Parameters.Add(new FixedValueParameter<IntValue>(IterationsParameterName, "The maximum number of iterations for constants optimization.", new IntValue(200)));
|
---|
[14319] | 127 | Parameters.Add(new FixedValueParameter<IntValue>(RestartsParameterName, "The number of independent random restarts (>0)", new IntValue(10)));
|
---|
[14258] | 128 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The PRNG seed value.", new IntValue()));
|
---|
| 129 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "Switch to determine if the random number seed should be initialized randomly.", new BoolValue(true)));
|
---|
[14319] | 130 | Parameters.Add(new FixedValueParameter<BoolValue>(InitParamsRandomlyParameterName, "Switch to determine if the real-valued model parameters should be initialized randomly in each restart.", new BoolValue(false)));
|
---|
[15505] | 131 | Parameters.Add(new FixedValueParameter<BoolValue>(ApplyLinearScalingParameterName, "Switch to determine if linear scaling terms should be added to the model", new BoolValue(true)));
|
---|
[14319] | 132 |
|
---|
| 133 | SetParameterHiddenState();
|
---|
| 134 |
|
---|
| 135 | InitParametersRandomlyParameter.Value.ValueChanged += (sender, args) => {
|
---|
| 136 | SetParameterHiddenState();
|
---|
| 137 | };
|
---|
[14024] | 138 | }
|
---|
[14258] | 139 |
|
---|
[14319] | 140 | private void SetParameterHiddenState() {
|
---|
| 141 | var hide = !InitializeParametersRandomly;
|
---|
| 142 | RestartsParameter.Hidden = hide;
|
---|
| 143 | SeedParameter.Hidden = hide;
|
---|
| 144 | SetSeedRandomlyParameter.Hidden = hide;
|
---|
| 145 | }
|
---|
| 146 |
|
---|
[14024] | 147 | [StorableHook(HookType.AfterDeserialization)]
|
---|
[14258] | 148 | private void AfterDeserialization() {
|
---|
| 149 | // BackwardsCompatibility3.3
|
---|
| 150 | #region Backwards compatible code, remove with 3.4
|
---|
| 151 | if (!Parameters.ContainsKey(RestartsParameterName))
|
---|
| 152 | Parameters.Add(new FixedValueParameter<IntValue>(RestartsParameterName, "The number of independent random restarts", new IntValue(1)));
|
---|
| 153 | if (!Parameters.ContainsKey(SeedParameterName))
|
---|
| 154 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The PRNG seed value.", new IntValue()));
|
---|
| 155 | if (!Parameters.ContainsKey(SetSeedRandomlyParameterName))
|
---|
| 156 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "Switch to determine if the random number seed should be initialized randomly.", new BoolValue(true)));
|
---|
[14319] | 157 | if (!Parameters.ContainsKey(InitParamsRandomlyParameterName))
|
---|
| 158 | Parameters.Add(new FixedValueParameter<BoolValue>(InitParamsRandomlyParameterName, "Switch to determine if the numeric parameters of the model should be initialized randomly.", new BoolValue(false)));
|
---|
[15502] | 159 | if (!Parameters.ContainsKey(ApplyLinearScalingParameterName))
|
---|
[15505] | 160 | Parameters.Add(new FixedValueParameter<BoolValue>(ApplyLinearScalingParameterName, "Switch to determine if linear scaling terms should be added to the model", new BoolValue(true)));
|
---|
[14319] | 161 |
|
---|
[15502] | 162 |
|
---|
[14319] | 163 | SetParameterHiddenState();
|
---|
| 164 | InitParametersRandomlyParameter.Value.ValueChanged += (sender, args) => {
|
---|
| 165 | SetParameterHiddenState();
|
---|
| 166 | };
|
---|
[14258] | 167 | #endregion
|
---|
| 168 | }
|
---|
[14024] | 169 |
|
---|
| 170 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
| 171 | return new NonlinearRegression(this, cloner);
|
---|
| 172 | }
|
---|
| 173 |
|
---|
| 174 | #region nonlinear regression
|
---|
[14523] | 175 | protected override void Run(CancellationToken cancellationToken) {
|
---|
[14319] | 176 | IRegressionSolution bestSolution = null;
|
---|
| 177 | if (InitializeParametersRandomly) {
|
---|
| 178 | var qualityTable = new DataTable("RMSE table");
|
---|
| 179 | qualityTable.VisualProperties.YAxisLogScale = true;
|
---|
| 180 | var trainRMSERow = new DataRow("RMSE (train)");
|
---|
| 181 | trainRMSERow.VisualProperties.ChartType = DataRowVisualProperties.DataRowChartType.Points;
|
---|
| 182 | var testRMSERow = new DataRow("RMSE test");
|
---|
| 183 | testRMSERow.VisualProperties.ChartType = DataRowVisualProperties.DataRowChartType.Points;
|
---|
[14316] | 184 |
|
---|
[14319] | 185 | qualityTable.Rows.Add(trainRMSERow);
|
---|
| 186 | qualityTable.Rows.Add(testRMSERow);
|
---|
| 187 | Results.Add(new Result(qualityTable.Name, qualityTable.Name + " for all restarts", qualityTable));
|
---|
[16240] | 188 | if (SetSeedRandomly) Seed = RandomSeedGenerator.GetSeed();
|
---|
[14319] | 189 | var rand = new MersenneTwister((uint)Seed);
|
---|
[15502] | 190 | bestSolution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations, ApplyLinearScaling, rand);
|
---|
[14319] | 191 | trainRMSERow.Values.Add(bestSolution.TrainingRootMeanSquaredError);
|
---|
| 192 | testRMSERow.Values.Add(bestSolution.TestRootMeanSquaredError);
|
---|
| 193 | for (int r = 0; r < Restarts; r++) {
|
---|
[15502] | 194 | var solution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations, ApplyLinearScaling, rand);
|
---|
[14319] | 195 | trainRMSERow.Values.Add(solution.TrainingRootMeanSquaredError);
|
---|
| 196 | testRMSERow.Values.Add(solution.TestRootMeanSquaredError);
|
---|
| 197 | if (solution.TrainingRootMeanSquaredError < bestSolution.TrainingRootMeanSquaredError) {
|
---|
| 198 | bestSolution = solution;
|
---|
| 199 | }
|
---|
[14258] | 200 | }
|
---|
[14319] | 201 | } else {
|
---|
[15502] | 202 | bestSolution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations, ApplyLinearScaling);
|
---|
[14258] | 203 | }
|
---|
| 204 |
|
---|
| 205 | Results.Add(new Result(RegressionSolutionResultName, "The nonlinear regression solution.", bestSolution));
|
---|
| 206 | Results.Add(new Result("Root mean square error (train)", "The root of the mean of squared errors of the regression solution on the training set.", new DoubleValue(bestSolution.TrainingRootMeanSquaredError)));
|
---|
| 207 | Results.Add(new Result("Root mean square error (test)", "The root of the mean of squared errors of the regression solution on the test set.", new DoubleValue(bestSolution.TestRootMeanSquaredError)));
|
---|
| 208 |
|
---|
[14024] | 209 | }
|
---|
| 210 |
|
---|
[14258] | 211 | /// <summary>
|
---|
| 212 | /// Fits a model to the data by optimizing the numeric constants.
|
---|
| 213 | /// Model is specified as infix expression containing variable names and numbers.
|
---|
| 214 | /// The starting point for the numeric constants is initialized randomly if a random number generator is specified (~N(0,1)). Otherwise the user specified constants are
|
---|
| 215 | /// used as a starting point.
|
---|
[14316] | 216 | /// </summary>-
|
---|
[14258] | 217 | /// <param name="problemData">Training and test data</param>
|
---|
| 218 | /// <param name="modelStructure">The function as infix expression</param>
|
---|
| 219 | /// <param name="maxIterations">Number of constant optimization iterations (using Levenberg-Marquardt algorithm)</param>
|
---|
| 220 | /// <param name="random">Optional random number generator for random initialization of numeric constants.</param>
|
---|
| 221 | /// <returns></returns>
|
---|
[15502] | 222 | public static ISymbolicRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData, string modelStructure, int maxIterations, bool applyLinearScaling, IRandom rand = null) {
|
---|
[14024] | 223 | var parser = new InfixExpressionParser();
|
---|
| 224 | var tree = parser.Parse(modelStructure);
|
---|
[14826] | 225 | // parser handles double and string variables equally by creating a VariableTreeNode
|
---|
| 226 | // post-process to replace VariableTreeNodes by FactorVariableTreeNodes for all string variables
|
---|
| 227 | var factorSymbol = new FactorVariable();
|
---|
| 228 | factorSymbol.VariableNames =
|
---|
| 229 | problemData.AllowedInputVariables.Where(name => problemData.Dataset.VariableHasType<string>(name));
|
---|
| 230 | factorSymbol.AllVariableNames = factorSymbol.VariableNames;
|
---|
| 231 | factorSymbol.VariableValues =
|
---|
| 232 | factorSymbol.VariableNames.Select(name =>
|
---|
| 233 | new KeyValuePair<string, Dictionary<string, int>>(name,
|
---|
| 234 | problemData.Dataset.GetReadOnlyStringValues(name).Distinct()
|
---|
| 235 | .Select((n, i) => Tuple.Create(n, i))
|
---|
| 236 | .ToDictionary(tup => tup.Item1, tup => tup.Item2)));
|
---|
[14258] | 237 |
|
---|
[14826] | 238 | foreach (var parent in tree.IterateNodesPrefix().ToArray()) {
|
---|
| 239 | for (int i = 0; i < parent.SubtreeCount; i++) {
|
---|
| 240 | var varChild = parent.GetSubtree(i) as VariableTreeNode;
|
---|
| 241 | var factorVarChild = parent.GetSubtree(i) as FactorVariableTreeNode;
|
---|
| 242 | if (varChild != null && factorSymbol.VariableNames.Contains(varChild.VariableName)) {
|
---|
| 243 | parent.RemoveSubtree(i);
|
---|
| 244 | var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
|
---|
| 245 | factorTreeNode.VariableName = varChild.VariableName;
|
---|
| 246 | factorTreeNode.Weights =
|
---|
| 247 | factorTreeNode.Symbol.GetVariableValues(factorTreeNode.VariableName).Select(_ => 1.0).ToArray();
|
---|
| 248 | // weight = 1.0 for each value
|
---|
| 249 | parent.InsertSubtree(i, factorTreeNode);
|
---|
| 250 | } else if (factorVarChild != null && factorSymbol.VariableNames.Contains(factorVarChild.VariableName)) {
|
---|
| 251 | if (factorSymbol.GetVariableValues(factorVarChild.VariableName).Count() != factorVarChild.Weights.Length)
|
---|
| 252 | throw new ArgumentException(
|
---|
| 253 | string.Format("Factor variable {0} needs exactly {1} weights",
|
---|
| 254 | factorVarChild.VariableName,
|
---|
| 255 | factorSymbol.GetVariableValues(factorVarChild.VariableName).Count()));
|
---|
| 256 | parent.RemoveSubtree(i);
|
---|
| 257 | var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
|
---|
| 258 | factorTreeNode.VariableName = factorVarChild.VariableName;
|
---|
| 259 | factorTreeNode.Weights = factorVarChild.Weights;
|
---|
| 260 | parent.InsertSubtree(i, factorTreeNode);
|
---|
| 261 | }
|
---|
| 262 | }
|
---|
| 263 | }
|
---|
| 264 |
|
---|
[14024] | 265 | if (!SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree)) throw new ArgumentException("The optimizer does not support the specified model structure.");
|
---|
| 266 |
|
---|
[14258] | 267 | // initialize constants randomly
|
---|
[14319] | 268 | if (rand != null) {
|
---|
[14258] | 269 | foreach (var node in tree.IterateNodesPrefix().OfType<ConstantTreeNode>()) {
|
---|
[14319] | 270 | double f = Math.Exp(NormalDistributedRandom.NextDouble(rand, 0, 1));
|
---|
| 271 | double s = rand.NextDouble() < 0.5 ? -1 : 1;
|
---|
| 272 | node.Value = s * node.Value * f;
|
---|
[14258] | 273 | }
|
---|
| 274 | }
|
---|
[14024] | 275 | var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
|
---|
[14258] | 276 |
|
---|
| 277 | SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, problemData.TrainingIndices,
|
---|
[15502] | 278 | applyLinearScaling: applyLinearScaling, maxIterations: maxIterations,
|
---|
[14024] | 279 | updateVariableWeights: false, updateConstantsInTree: true);
|
---|
| 280 |
|
---|
[15502] | 281 | var model = new SymbolicRegressionModel(problemData.TargetVariable, tree, (ISymbolicDataAnalysisExpressionTreeInterpreter)interpreter.Clone());
|
---|
| 282 | if (applyLinearScaling)
|
---|
| 283 | model.Scale(problemData);
|
---|
| 284 |
|
---|
| 285 | SymbolicRegressionSolution solution = new SymbolicRegressionSolution(model, (IRegressionProblemData)problemData.Clone());
|
---|
[14024] | 286 | solution.Model.Name = "Regression Model";
|
---|
| 287 | solution.Name = "Regression Solution";
|
---|
| 288 | return solution;
|
---|
| 289 | }
|
---|
| 290 | #endregion
|
---|
| 291 | }
|
---|
| 292 | }
|
---|