1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2015 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.Optimization;
|
---|
29 | using HeuristicLab.Parameters;
|
---|
30 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
31 | using HeuristicLab.Problems.DataAnalysis;
|
---|
32 | using LibSVM;
|
---|
33 |
|
---|
34 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
35 | /// <summary>
|
---|
36 | /// Support vector machine regression data analysis algorithm.
|
---|
37 | /// </summary>
|
---|
38 | [Item("Support Vector Regression (SVM)", "Support vector machine regression data analysis algorithm (wrapper for libSVM).")]
|
---|
39 | [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 110)]
|
---|
40 | [StorableType("11DB1DEB-3B8D-44F0-B522-28328E57FF4B")]
|
---|
41 | public sealed class SupportVectorRegression : FixedDataAnalysisAlgorithm<IRegressionProblem> {
|
---|
42 | private const string SvmTypeParameterName = "SvmType";
|
---|
43 | private const string KernelTypeParameterName = "KernelType";
|
---|
44 | private const string CostParameterName = "Cost";
|
---|
45 | private const string NuParameterName = "Nu";
|
---|
46 | private const string GammaParameterName = "Gamma";
|
---|
47 | private const string EpsilonParameterName = "Epsilon";
|
---|
48 | private const string DegreeParameterName = "Degree";
|
---|
49 | private const string CreateSolutionParameterName = "CreateSolution";
|
---|
50 |
|
---|
51 | #region parameter properties
|
---|
52 | public IConstrainedValueParameter<StringValue> SvmTypeParameter {
|
---|
53 | get { return (IConstrainedValueParameter<StringValue>)Parameters[SvmTypeParameterName]; }
|
---|
54 | }
|
---|
55 | public IConstrainedValueParameter<StringValue> KernelTypeParameter {
|
---|
56 | get { return (IConstrainedValueParameter<StringValue>)Parameters[KernelTypeParameterName]; }
|
---|
57 | }
|
---|
58 | public IValueParameter<DoubleValue> NuParameter {
|
---|
59 | get { return (IValueParameter<DoubleValue>)Parameters[NuParameterName]; }
|
---|
60 | }
|
---|
61 | public IValueParameter<DoubleValue> CostParameter {
|
---|
62 | get { return (IValueParameter<DoubleValue>)Parameters[CostParameterName]; }
|
---|
63 | }
|
---|
64 | public IValueParameter<DoubleValue> GammaParameter {
|
---|
65 | get { return (IValueParameter<DoubleValue>)Parameters[GammaParameterName]; }
|
---|
66 | }
|
---|
67 | public IValueParameter<DoubleValue> EpsilonParameter {
|
---|
68 | get { return (IValueParameter<DoubleValue>)Parameters[EpsilonParameterName]; }
|
---|
69 | }
|
---|
70 | public IValueParameter<IntValue> DegreeParameter {
|
---|
71 | get { return (IValueParameter<IntValue>)Parameters[DegreeParameterName]; }
|
---|
72 | }
|
---|
73 | public IFixedValueParameter<BoolValue> CreateSolutionParameter {
|
---|
74 | get { return (IFixedValueParameter<BoolValue>)Parameters[CreateSolutionParameterName]; }
|
---|
75 | }
|
---|
76 | #endregion
|
---|
77 | #region properties
|
---|
78 | public StringValue SvmType {
|
---|
79 | get { return SvmTypeParameter.Value; }
|
---|
80 | set { SvmTypeParameter.Value = value; }
|
---|
81 | }
|
---|
82 | public StringValue KernelType {
|
---|
83 | get { return KernelTypeParameter.Value; }
|
---|
84 | set { KernelTypeParameter.Value = value; }
|
---|
85 | }
|
---|
86 | public DoubleValue Nu {
|
---|
87 | get { return NuParameter.Value; }
|
---|
88 | }
|
---|
89 | public DoubleValue Cost {
|
---|
90 | get { return CostParameter.Value; }
|
---|
91 | }
|
---|
92 | public DoubleValue Gamma {
|
---|
93 | get { return GammaParameter.Value; }
|
---|
94 | }
|
---|
95 | public DoubleValue Epsilon {
|
---|
96 | get { return EpsilonParameter.Value; }
|
---|
97 | }
|
---|
98 | public IntValue Degree {
|
---|
99 | get { return DegreeParameter.Value; }
|
---|
100 | }
|
---|
101 | public bool CreateSolution {
|
---|
102 | get { return CreateSolutionParameter.Value.Value; }
|
---|
103 | set { CreateSolutionParameter.Value.Value = value; }
|
---|
104 | }
|
---|
105 | #endregion
|
---|
106 | [StorableConstructor]
|
---|
107 | private SupportVectorRegression(bool deserializing) : base(deserializing) { }
|
---|
108 | private SupportVectorRegression(SupportVectorRegression original, Cloner cloner)
|
---|
109 | : base(original, cloner) {
|
---|
110 | }
|
---|
111 | public SupportVectorRegression()
|
---|
112 | : base() {
|
---|
113 | Problem = new RegressionProblem();
|
---|
114 |
|
---|
115 | List<StringValue> svrTypes = (from type in new List<string> { "NU_SVR", "EPSILON_SVR" }
|
---|
116 | select new StringValue(type).AsReadOnly())
|
---|
117 | .ToList();
|
---|
118 | ItemSet<StringValue> svrTypeSet = new ItemSet<StringValue>(svrTypes);
|
---|
119 | List<StringValue> kernelTypes = (from type in new List<string> { "LINEAR", "POLY", "SIGMOID", "RBF" }
|
---|
120 | select new StringValue(type).AsReadOnly())
|
---|
121 | .ToList();
|
---|
122 | ItemSet<StringValue> kernelTypeSet = new ItemSet<StringValue>(kernelTypes);
|
---|
123 | Parameters.Add(new ConstrainedValueParameter<StringValue>(SvmTypeParameterName, "The type of SVM to use.", svrTypeSet, svrTypes[0]));
|
---|
124 | Parameters.Add(new ConstrainedValueParameter<StringValue>(KernelTypeParameterName, "The kernel type to use for the SVM.", kernelTypeSet, kernelTypes[3]));
|
---|
125 | Parameters.Add(new ValueParameter<DoubleValue>(NuParameterName, "The value of the nu parameter of the nu-SVR.", new DoubleValue(0.5)));
|
---|
126 | Parameters.Add(new ValueParameter<DoubleValue>(CostParameterName, "The value of the C (cost) parameter of epsilon-SVR and nu-SVR.", new DoubleValue(1.0)));
|
---|
127 | Parameters.Add(new ValueParameter<DoubleValue>(GammaParameterName, "The value of the gamma parameter in the kernel function.", new DoubleValue(1.0)));
|
---|
128 | Parameters.Add(new ValueParameter<DoubleValue>(EpsilonParameterName, "The value of the epsilon parameter for epsilon-SVR.", new DoubleValue(0.1)));
|
---|
129 | Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName, "The degree parameter for the polynomial kernel function.", new IntValue(3)));
|
---|
130 | Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName, "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
|
---|
131 | Parameters[CreateSolutionParameterName].Hidden = true;
|
---|
132 | }
|
---|
133 | [StorableHook(HookType.AfterDeserialization)]
|
---|
134 | private void AfterDeserialization() {
|
---|
135 | #region backwards compatibility (change with 3.4)
|
---|
136 |
|
---|
137 | if (!Parameters.ContainsKey(DegreeParameterName)) {
|
---|
138 | Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName,
|
---|
139 | "The degree parameter for the polynomial kernel function.", new IntValue(3)));
|
---|
140 | }
|
---|
141 | if (!Parameters.ContainsKey(CreateSolutionParameterName)) {
|
---|
142 | Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName, "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
|
---|
143 | Parameters[CreateSolutionParameterName].Hidden = true;
|
---|
144 | }
|
---|
145 | #endregion
|
---|
146 | }
|
---|
147 |
|
---|
148 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
149 | return new SupportVectorRegression(this, cloner);
|
---|
150 | }
|
---|
151 |
|
---|
152 | #region support vector regression
|
---|
153 | protected override void Run() {
|
---|
154 | IRegressionProblemData problemData = Problem.ProblemData;
|
---|
155 | IEnumerable<string> selectedInputVariables = problemData.AllowedInputVariables;
|
---|
156 | int nSv;
|
---|
157 | ISupportVectorMachineModel model;
|
---|
158 | Run(problemData, selectedInputVariables, SvmType.Value, KernelType.Value, Cost.Value, Nu.Value, Gamma.Value, Epsilon.Value, Degree.Value, out model, out nSv);
|
---|
159 |
|
---|
160 | if (CreateSolution) {
|
---|
161 | var solution = new SupportVectorRegressionSolution((SupportVectorMachineModel)model, (IRegressionProblemData)problemData.Clone());
|
---|
162 | Results.Add(new Result("Support vector regression solution", "The support vector regression solution.", solution));
|
---|
163 | }
|
---|
164 |
|
---|
165 | Results.Add(new Result("Number of support vectors", "The number of support vectors of the SVR solution.", new IntValue(nSv)));
|
---|
166 |
|
---|
167 |
|
---|
168 | {
|
---|
169 | // calculate regression model metrics
|
---|
170 | var ds = problemData.Dataset;
|
---|
171 | var trainRows = problemData.TrainingIndices;
|
---|
172 | var testRows = problemData.TestIndices;
|
---|
173 | var yTrain = ds.GetDoubleValues(problemData.TargetVariable, trainRows);
|
---|
174 | var yTest = ds.GetDoubleValues(problemData.TargetVariable, testRows);
|
---|
175 | var yPredTrain = model.GetEstimatedValues(ds, trainRows).ToArray();
|
---|
176 | var yPredTest = model.GetEstimatedValues(ds, testRows).ToArray();
|
---|
177 |
|
---|
178 | OnlineCalculatorError error;
|
---|
179 | var trainMse = OnlineMeanSquaredErrorCalculator.Calculate(yPredTrain, yTrain, out error);
|
---|
180 | if (error != OnlineCalculatorError.None) trainMse = double.MaxValue;
|
---|
181 | var testMse = OnlineMeanSquaredErrorCalculator.Calculate(yPredTest, yTest, out error);
|
---|
182 | if (error != OnlineCalculatorError.None) testMse = double.MaxValue;
|
---|
183 |
|
---|
184 | Results.Add(new Result("Mean squared error (training)", "The mean of squared errors of the SVR solution on the training partition.", new DoubleValue(trainMse)));
|
---|
185 | Results.Add(new Result("Mean squared error (test)", "The mean of squared errors of the SVR solution on the test partition.", new DoubleValue(testMse)));
|
---|
186 |
|
---|
187 |
|
---|
188 | var trainMae = OnlineMeanAbsoluteErrorCalculator.Calculate(yPredTrain, yTrain, out error);
|
---|
189 | if (error != OnlineCalculatorError.None) trainMae = double.MaxValue;
|
---|
190 | var testMae = OnlineMeanAbsoluteErrorCalculator.Calculate(yPredTest, yTest, out error);
|
---|
191 | if (error != OnlineCalculatorError.None) testMae = double.MaxValue;
|
---|
192 |
|
---|
193 | Results.Add(new Result("Mean absolute error (training)", "The mean of absolute errors of the SVR solution on the training partition.", new DoubleValue(trainMae)));
|
---|
194 | Results.Add(new Result("Mean absolute error (test)", "The mean of absolute errors of the SVR solution on the test partition.", new DoubleValue(testMae)));
|
---|
195 |
|
---|
196 |
|
---|
197 | var trainRelErr = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(yPredTrain, yTrain, out error);
|
---|
198 | if (error != OnlineCalculatorError.None) trainRelErr = double.MaxValue;
|
---|
199 | var testRelErr = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(yPredTest, yTest, out error);
|
---|
200 | if (error != OnlineCalculatorError.None) testRelErr = double.MaxValue;
|
---|
201 |
|
---|
202 | Results.Add(new Result("Average relative error (training)", "The mean of relative errors of the SVR solution on the training partition.", new DoubleValue(trainRelErr)));
|
---|
203 | Results.Add(new Result("Average relative error (test)", "The mean of relative errors of the SVR solution on the test partition.", new DoubleValue(testRelErr)));
|
---|
204 | }
|
---|
205 | }
|
---|
206 |
|
---|
207 | // BackwardsCompatibility3.4
|
---|
208 | #region Backwards compatible code, remove with 3.5
|
---|
209 | // for compatibility with old API
|
---|
210 | public static SupportVectorRegressionSolution CreateSupportVectorRegressionSolution(
|
---|
211 | IRegressionProblemData problemData, IEnumerable<string> allowedInputVariables,
|
---|
212 | string svmType, string kernelType, double cost, double nu, double gamma, double epsilon, int degree,
|
---|
213 | out double trainingR2, out double testR2, out int nSv) {
|
---|
214 | ISupportVectorMachineModel model;
|
---|
215 | Run(problemData, allowedInputVariables, svmType, kernelType, cost, nu, gamma, epsilon, degree, out model, out nSv);
|
---|
216 |
|
---|
217 | var solution = new SupportVectorRegressionSolution((SupportVectorMachineModel)model, (IRegressionProblemData)problemData.Clone());
|
---|
218 | trainingR2 = solution.TrainingRSquared;
|
---|
219 | testR2 = solution.TestRSquared;
|
---|
220 | return solution;
|
---|
221 | }
|
---|
222 | #endregion
|
---|
223 |
|
---|
224 | public static void Run(IRegressionProblemData problemData, IEnumerable<string> allowedInputVariables,
|
---|
225 | string svmType, string kernelType, double cost, double nu, double gamma, double epsilon, int degree,
|
---|
226 | out ISupportVectorMachineModel model, out int nSv) {
|
---|
227 | var dataset = problemData.Dataset;
|
---|
228 | string targetVariable = problemData.TargetVariable;
|
---|
229 | IEnumerable<int> rows = problemData.TrainingIndices;
|
---|
230 |
|
---|
231 | svm_parameter parameter = new svm_parameter {
|
---|
232 | svm_type = GetSvmType(svmType),
|
---|
233 | kernel_type = GetKernelType(kernelType),
|
---|
234 | C = cost,
|
---|
235 | nu = nu,
|
---|
236 | gamma = gamma,
|
---|
237 | p = epsilon,
|
---|
238 | cache_size = 500,
|
---|
239 | probability = 0,
|
---|
240 | eps = 0.001,
|
---|
241 | degree = degree,
|
---|
242 | shrinking = 1,
|
---|
243 | coef0 = 0
|
---|
244 | };
|
---|
245 |
|
---|
246 | svm_problem problem = SupportVectorMachineUtil.CreateSvmProblem(dataset, targetVariable, allowedInputVariables, rows);
|
---|
247 | RangeTransform rangeTransform = RangeTransform.Compute(problem);
|
---|
248 | svm_problem scaledProblem = rangeTransform.Scale(problem);
|
---|
249 | var svmModel = svm.svm_train(scaledProblem, parameter);
|
---|
250 | nSv = svmModel.SV.Length;
|
---|
251 |
|
---|
252 | model = new SupportVectorMachineModel(svmModel, rangeTransform, targetVariable, allowedInputVariables);
|
---|
253 | }
|
---|
254 |
|
---|
255 | private static int GetSvmType(string svmType) {
|
---|
256 | if (svmType == "NU_SVR") return svm_parameter.NU_SVR;
|
---|
257 | if (svmType == "EPSILON_SVR") return svm_parameter.EPSILON_SVR;
|
---|
258 | throw new ArgumentException("Unknown SVM type");
|
---|
259 | }
|
---|
260 |
|
---|
261 | private static int GetKernelType(string kernelType) {
|
---|
262 | if (kernelType == "LINEAR") return svm_parameter.LINEAR;
|
---|
263 | if (kernelType == "POLY") return svm_parameter.POLY;
|
---|
264 | if (kernelType == "SIGMOID") return svm_parameter.SIGMOID;
|
---|
265 | if (kernelType == "RBF") return svm_parameter.RBF;
|
---|
266 | throw new ArgumentException("Unknown kernel type");
|
---|
267 | }
|
---|
268 | #endregion
|
---|
269 | }
|
---|
270 | }
|
---|