1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2016 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 System.Threading;
|
---|
26 | using HeuristicLab.Analysis;
|
---|
27 | using HeuristicLab.Common;
|
---|
28 | using HeuristicLab.Core;
|
---|
29 | using HeuristicLab.Data;
|
---|
30 | using HeuristicLab.Optimization;
|
---|
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;
|
---|
36 | using HeuristicLab.Random;
|
---|
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> {
|
---|
46 | private const string RegressionSolutionResultName = "Regression solution";
|
---|
47 | private const string ModelStructureParameterName = "Model structure";
|
---|
48 | private const string IterationsParameterName = "Iterations";
|
---|
49 | private const string RestartsParameterName = "Restarts";
|
---|
50 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
---|
51 | private const string SeedParameterName = "Seed";
|
---|
52 | private const string InitParamsRandomlyParameterName = "InitializeParametersRandomly";
|
---|
53 |
|
---|
54 | public IFixedValueParameter<StringValue> ModelStructureParameter {
|
---|
55 | get { return (IFixedValueParameter<StringValue>)Parameters[ModelStructureParameterName]; }
|
---|
56 | }
|
---|
57 | public IFixedValueParameter<IntValue> IterationsParameter {
|
---|
58 | get { return (IFixedValueParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
59 | }
|
---|
60 |
|
---|
61 | public IFixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
---|
62 | get { return (IFixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
---|
63 | }
|
---|
64 |
|
---|
65 | public IFixedValueParameter<IntValue> SeedParameter {
|
---|
66 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
---|
67 | }
|
---|
68 |
|
---|
69 | public IFixedValueParameter<IntValue> RestartsParameter {
|
---|
70 | get { return (IFixedValueParameter<IntValue>)Parameters[RestartsParameterName]; }
|
---|
71 | }
|
---|
72 |
|
---|
73 | public IFixedValueParameter<BoolValue> InitParametersRandomlyParameter {
|
---|
74 | get { return (IFixedValueParameter<BoolValue>)Parameters[InitParamsRandomlyParameterName]; }
|
---|
75 | }
|
---|
76 |
|
---|
77 | public string ModelStructure {
|
---|
78 | get { return ModelStructureParameter.Value.Value; }
|
---|
79 | set { ModelStructureParameter.Value.Value = value; }
|
---|
80 | }
|
---|
81 |
|
---|
82 | public int Iterations {
|
---|
83 | get { return IterationsParameter.Value.Value; }
|
---|
84 | set { IterationsParameter.Value.Value = value; }
|
---|
85 | }
|
---|
86 |
|
---|
87 | public int Restarts {
|
---|
88 | get { return RestartsParameter.Value.Value; }
|
---|
89 | set { RestartsParameter.Value.Value = value; }
|
---|
90 | }
|
---|
91 |
|
---|
92 | public int Seed {
|
---|
93 | get { return SeedParameter.Value.Value; }
|
---|
94 | set { SeedParameter.Value.Value = value; }
|
---|
95 | }
|
---|
96 |
|
---|
97 | public bool SetSeedRandomly {
|
---|
98 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
99 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
100 | }
|
---|
101 |
|
---|
102 | public bool InitializeParametersRandomly {
|
---|
103 | get { return InitParametersRandomlyParameter.Value.Value; }
|
---|
104 | set { InitParametersRandomlyParameter.Value.Value = value; }
|
---|
105 | }
|
---|
106 |
|
---|
107 | [StorableConstructor]
|
---|
108 | private NonlinearRegression(bool deserializing) : base(deserializing) { }
|
---|
109 | private NonlinearRegression(NonlinearRegression original, Cloner cloner)
|
---|
110 | : base(original, cloner) {
|
---|
111 | }
|
---|
112 | public NonlinearRegression()
|
---|
113 | : base() {
|
---|
114 | Problem = new RegressionProblem();
|
---|
115 | 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")));
|
---|
116 | Parameters.Add(new FixedValueParameter<IntValue>(IterationsParameterName, "The maximum number of iterations for constants optimization.", new IntValue(200)));
|
---|
117 | Parameters.Add(new FixedValueParameter<IntValue>(RestartsParameterName, "The number of independent random restarts (>0)", new IntValue(10)));
|
---|
118 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The PRNG seed value.", new IntValue()));
|
---|
119 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "Switch to determine if the random number seed should be initialized randomly.", new BoolValue(true)));
|
---|
120 | 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)));
|
---|
121 |
|
---|
122 | SetParameterHiddenState();
|
---|
123 |
|
---|
124 | InitParametersRandomlyParameter.Value.ValueChanged += (sender, args) => {
|
---|
125 | SetParameterHiddenState();
|
---|
126 | };
|
---|
127 | }
|
---|
128 |
|
---|
129 | private void SetParameterHiddenState() {
|
---|
130 | var hide = !InitializeParametersRandomly;
|
---|
131 | RestartsParameter.Hidden = hide;
|
---|
132 | SeedParameter.Hidden = hide;
|
---|
133 | SetSeedRandomlyParameter.Hidden = hide;
|
---|
134 | }
|
---|
135 |
|
---|
136 | [StorableHook(HookType.AfterDeserialization)]
|
---|
137 | private void AfterDeserialization() {
|
---|
138 | // BackwardsCompatibility3.3
|
---|
139 | #region Backwards compatible code, remove with 3.4
|
---|
140 | if (!Parameters.ContainsKey(RestartsParameterName))
|
---|
141 | Parameters.Add(new FixedValueParameter<IntValue>(RestartsParameterName, "The number of independent random restarts", new IntValue(1)));
|
---|
142 | if (!Parameters.ContainsKey(SeedParameterName))
|
---|
143 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The PRNG seed value.", new IntValue()));
|
---|
144 | if (!Parameters.ContainsKey(SetSeedRandomlyParameterName))
|
---|
145 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "Switch to determine if the random number seed should be initialized randomly.", new BoolValue(true)));
|
---|
146 | if (!Parameters.ContainsKey(InitParamsRandomlyParameterName))
|
---|
147 | Parameters.Add(new FixedValueParameter<BoolValue>(InitParamsRandomlyParameterName, "Switch to determine if the numeric parameters of the model should be initialized randomly.", new BoolValue(false)));
|
---|
148 |
|
---|
149 | SetParameterHiddenState();
|
---|
150 | InitParametersRandomlyParameter.Value.ValueChanged += (sender, args) => {
|
---|
151 | SetParameterHiddenState();
|
---|
152 | };
|
---|
153 | #endregion
|
---|
154 | }
|
---|
155 |
|
---|
156 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
157 | return new NonlinearRegression(this, cloner);
|
---|
158 | }
|
---|
159 |
|
---|
160 | #region nonlinear regression
|
---|
161 | protected override void Run(CancellationToken cancellationToken) {
|
---|
162 | IRegressionSolution bestSolution = null;
|
---|
163 | if (InitializeParametersRandomly) {
|
---|
164 | var qualityTable = new DataTable("RMSE table");
|
---|
165 | qualityTable.VisualProperties.YAxisLogScale = true;
|
---|
166 | var trainRMSERow = new DataRow("RMSE (train)");
|
---|
167 | trainRMSERow.VisualProperties.ChartType = DataRowVisualProperties.DataRowChartType.Points;
|
---|
168 | var testRMSERow = new DataRow("RMSE test");
|
---|
169 | testRMSERow.VisualProperties.ChartType = DataRowVisualProperties.DataRowChartType.Points;
|
---|
170 |
|
---|
171 | qualityTable.Rows.Add(trainRMSERow);
|
---|
172 | qualityTable.Rows.Add(testRMSERow);
|
---|
173 | Results.Add(new Result(qualityTable.Name, qualityTable.Name + " for all restarts", qualityTable));
|
---|
174 | if (SetSeedRandomly) Seed = (new System.Random()).Next();
|
---|
175 | var rand = new MersenneTwister((uint)Seed);
|
---|
176 | bestSolution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations, rand);
|
---|
177 | trainRMSERow.Values.Add(bestSolution.TrainingRootMeanSquaredError);
|
---|
178 | testRMSERow.Values.Add(bestSolution.TestRootMeanSquaredError);
|
---|
179 | for (int r = 0; r < Restarts; r++) {
|
---|
180 | var solution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations, rand);
|
---|
181 | trainRMSERow.Values.Add(solution.TrainingRootMeanSquaredError);
|
---|
182 | testRMSERow.Values.Add(solution.TestRootMeanSquaredError);
|
---|
183 | if (solution.TrainingRootMeanSquaredError < bestSolution.TrainingRootMeanSquaredError) {
|
---|
184 | bestSolution = solution;
|
---|
185 | }
|
---|
186 | }
|
---|
187 | } else {
|
---|
188 | bestSolution = CreateRegressionSolution(Problem.ProblemData, ModelStructure, Iterations);
|
---|
189 | }
|
---|
190 |
|
---|
191 | Results.Add(new Result(RegressionSolutionResultName, "The nonlinear regression solution.", bestSolution));
|
---|
192 | 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)));
|
---|
193 | 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)));
|
---|
194 |
|
---|
195 | }
|
---|
196 |
|
---|
197 | /// <summary>
|
---|
198 | /// Fits a model to the data by optimizing the numeric constants.
|
---|
199 | /// Model is specified as infix expression containing variable names and numbers.
|
---|
200 | /// 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
|
---|
201 | /// used as a starting point.
|
---|
202 | /// </summary>-
|
---|
203 | /// <param name="problemData">Training and test data</param>
|
---|
204 | /// <param name="modelStructure">The function as infix expression</param>
|
---|
205 | /// <param name="maxIterations">Number of constant optimization iterations (using Levenberg-Marquardt algorithm)</param>
|
---|
206 | /// <param name="random">Optional random number generator for random initialization of numeric constants.</param>
|
---|
207 | /// <returns></returns>
|
---|
208 | public static ISymbolicRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData, string modelStructure, int maxIterations, IRandom rand = null) {
|
---|
209 | var parser = new InfixExpressionParser();
|
---|
210 | var tree = parser.Parse(modelStructure);
|
---|
211 | // parser handles double and string variables equally by creating a VariableTreeNode
|
---|
212 | // post-process to replace VariableTreeNodes by FactorVariableTreeNodes for all string variables
|
---|
213 | var factorSymbol = new FactorVariable();
|
---|
214 | factorSymbol.VariableNames =
|
---|
215 | problemData.AllowedInputVariables.Where(name => problemData.Dataset.VariableHasType<string>(name));
|
---|
216 | factorSymbol.AllVariableNames = factorSymbol.VariableNames;
|
---|
217 | factorSymbol.VariableValues =
|
---|
218 | factorSymbol.VariableNames.Select(name =>
|
---|
219 | new KeyValuePair<string, Dictionary<string, int>>(name,
|
---|
220 | problemData.Dataset.GetReadOnlyStringValues(name).Distinct()
|
---|
221 | .Select((n, i) => Tuple.Create(n, i))
|
---|
222 | .ToDictionary(tup => tup.Item1, tup => tup.Item2)));
|
---|
223 |
|
---|
224 | foreach (var parent in tree.IterateNodesPrefix().ToArray()) {
|
---|
225 | for (int i = 0; i < parent.SubtreeCount; i++) {
|
---|
226 | var varChild = parent.GetSubtree(i) as VariableTreeNode;
|
---|
227 | var factorVarChild = parent.GetSubtree(i) as FactorVariableTreeNode;
|
---|
228 | if (varChild != null && factorSymbol.VariableNames.Contains(varChild.VariableName)) {
|
---|
229 | parent.RemoveSubtree(i);
|
---|
230 | var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
|
---|
231 | factorTreeNode.VariableName = varChild.VariableName;
|
---|
232 | factorTreeNode.Weights =
|
---|
233 | factorTreeNode.Symbol.GetVariableValues(factorTreeNode.VariableName).Select(_ => 1.0).ToArray();
|
---|
234 | // weight = 1.0 for each value
|
---|
235 | parent.InsertSubtree(i, factorTreeNode);
|
---|
236 | } else if (factorVarChild != null && factorSymbol.VariableNames.Contains(factorVarChild.VariableName)) {
|
---|
237 | if (factorSymbol.GetVariableValues(factorVarChild.VariableName).Count() != factorVarChild.Weights.Length)
|
---|
238 | throw new ArgumentException(
|
---|
239 | string.Format("Factor variable {0} needs exactly {1} weights",
|
---|
240 | factorVarChild.VariableName,
|
---|
241 | factorSymbol.GetVariableValues(factorVarChild.VariableName).Count()));
|
---|
242 | parent.RemoveSubtree(i);
|
---|
243 | var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
|
---|
244 | factorTreeNode.VariableName = factorVarChild.VariableName;
|
---|
245 | factorTreeNode.Weights = factorVarChild.Weights;
|
---|
246 | parent.InsertSubtree(i, factorTreeNode);
|
---|
247 | }
|
---|
248 | }
|
---|
249 | }
|
---|
250 |
|
---|
251 | if (!SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree)) throw new ArgumentException("The optimizer does not support the specified model structure.");
|
---|
252 |
|
---|
253 | // initialize constants randomly
|
---|
254 | if (rand != null) {
|
---|
255 | foreach (var node in tree.IterateNodesPrefix().OfType<ConstantTreeNode>()) {
|
---|
256 | double f = Math.Exp(NormalDistributedRandom.NextDouble(rand, 0, 1));
|
---|
257 | double s = rand.NextDouble() < 0.5 ? -1 : 1;
|
---|
258 | node.Value = s * node.Value * f;
|
---|
259 | }
|
---|
260 | }
|
---|
261 | var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
|
---|
262 |
|
---|
263 | SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, problemData.TrainingIndices,
|
---|
264 | applyLinearScaling: false, maxIterations: maxIterations,
|
---|
265 | updateVariableWeights: false, updateConstantsInTree: true);
|
---|
266 |
|
---|
267 | var scaledModel = new SymbolicRegressionModel(problemData.TargetVariable, tree, (ISymbolicDataAnalysisExpressionTreeInterpreter)interpreter.Clone());
|
---|
268 | scaledModel.Scale(problemData);
|
---|
269 | SymbolicRegressionSolution solution = new SymbolicRegressionSolution(scaledModel, (IRegressionProblemData)problemData.Clone());
|
---|
270 | solution.Model.Name = "Regression Model";
|
---|
271 | solution.Name = "Regression Solution";
|
---|
272 | return solution;
|
---|
273 | }
|
---|
274 | #endregion
|
---|
275 | }
|
---|
276 | }
|
---|