1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 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 HEAL.Attic;
|
---|
27 | using HeuristicLab.Analysis;
|
---|
28 | using HeuristicLab.Common;
|
---|
29 | using HeuristicLab.Core;
|
---|
30 | using HeuristicLab.Data;
|
---|
31 | using HeuristicLab.Optimization;
|
---|
32 | using HeuristicLab.Parameters;
|
---|
33 | using HeuristicLab.Problems.DataAnalysis;
|
---|
34 | using HeuristicLab.Random;
|
---|
35 |
|
---|
36 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
37 | [Item("Generalized Additive Model (GAM)", "Generalized additive model using uni-variate penalized regression splines as base learner.")]
|
---|
38 | [StorableType("98A887E7-73DD-4602-BD6C-2F6B9E6FBBC5")]
|
---|
39 | [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 600)]
|
---|
40 | public sealed class GeneralizedAdditiveModelAlgorithm : FixedDataAnalysisAlgorithm<IRegressionProblem> {
|
---|
41 | #region ParameterNames
|
---|
42 |
|
---|
43 | private const string IterationsParameterName = "Iterations";
|
---|
44 | private const string LambdaParameterName = "Lambda";
|
---|
45 | private const string SeedParameterName = "Seed";
|
---|
46 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
---|
47 | private const string CreateSolutionParameterName = "CreateSolution";
|
---|
48 | #endregion
|
---|
49 |
|
---|
50 | #region ParameterProperties
|
---|
51 |
|
---|
52 | public IFixedValueParameter<IntValue> IterationsParameter {
|
---|
53 | get { return (IFixedValueParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
54 | }
|
---|
55 |
|
---|
56 | public IFixedValueParameter<DoubleValue> LambdaParameter {
|
---|
57 | get { return (IFixedValueParameter<DoubleValue>)Parameters[LambdaParameterName]; }
|
---|
58 | }
|
---|
59 |
|
---|
60 | public IFixedValueParameter<IntValue> SeedParameter {
|
---|
61 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
---|
62 | }
|
---|
63 |
|
---|
64 | public FixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
---|
65 | get { return (FixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
---|
66 | }
|
---|
67 |
|
---|
68 | public IFixedValueParameter<BoolValue> CreateSolutionParameter {
|
---|
69 | get { return (IFixedValueParameter<BoolValue>)Parameters[CreateSolutionParameterName]; }
|
---|
70 | }
|
---|
71 |
|
---|
72 | #endregion
|
---|
73 |
|
---|
74 | #region Properties
|
---|
75 |
|
---|
76 | public int Iterations {
|
---|
77 | get { return IterationsParameter.Value.Value; }
|
---|
78 | set { IterationsParameter.Value.Value = value; }
|
---|
79 | }
|
---|
80 |
|
---|
81 | public double Lambda {
|
---|
82 | get { return LambdaParameter.Value.Value; }
|
---|
83 | set { LambdaParameter.Value.Value = value; }
|
---|
84 | }
|
---|
85 |
|
---|
86 | public int Seed {
|
---|
87 | get { return SeedParameter.Value.Value; }
|
---|
88 | set { SeedParameter.Value.Value = value; }
|
---|
89 | }
|
---|
90 |
|
---|
91 | public bool SetSeedRandomly {
|
---|
92 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
93 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
94 | }
|
---|
95 |
|
---|
96 | public bool CreateSolution {
|
---|
97 | get { return CreateSolutionParameter.Value.Value; }
|
---|
98 | set { CreateSolutionParameter.Value.Value = value; }
|
---|
99 | }
|
---|
100 |
|
---|
101 | #endregion
|
---|
102 |
|
---|
103 | [StorableConstructor]
|
---|
104 | private GeneralizedAdditiveModelAlgorithm(StorableConstructorFlag deserializing)
|
---|
105 | : base(deserializing) {
|
---|
106 | }
|
---|
107 |
|
---|
108 | private GeneralizedAdditiveModelAlgorithm(GeneralizedAdditiveModelAlgorithm original, Cloner cloner)
|
---|
109 | : base(original, cloner) {
|
---|
110 | }
|
---|
111 |
|
---|
112 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
113 | return new GeneralizedAdditiveModelAlgorithm(this, cloner);
|
---|
114 | }
|
---|
115 |
|
---|
116 | public GeneralizedAdditiveModelAlgorithm() {
|
---|
117 | Problem = new RegressionProblem(); // default problem
|
---|
118 |
|
---|
119 | Parameters.Add(new FixedValueParameter<IntValue>(IterationsParameterName,
|
---|
120 | "Number of iterations. Try a large value and check convergence of the error over iterations. Usually, only a few iterations (e.g. 10) are needed for convergence.", new IntValue(10)));
|
---|
121 | Parameters.Add(new FixedValueParameter<DoubleValue>(LambdaParameterName,
|
---|
122 | "The penalty parameter for the penalized regression splines. Set to a value between -8 (weak smoothing) and 8 (strong smooting). Usually, a value between -4 and 4 should be fine", new DoubleValue(3)));
|
---|
123 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName,
|
---|
124 | "The random seed used to initialize the new pseudo random number generator.", new IntValue(0)));
|
---|
125 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName,
|
---|
126 | "True if the random seed should be set to a random value, otherwise false.", new BoolValue(true)));
|
---|
127 | Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName,
|
---|
128 | "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
|
---|
129 | Parameters[CreateSolutionParameterName].Hidden = true;
|
---|
130 | }
|
---|
131 |
|
---|
132 | protected override void Run(CancellationToken cancellationToken) {
|
---|
133 | // Set up the algorithm
|
---|
134 | if (SetSeedRandomly) Seed = new System.Random().Next();
|
---|
135 | var rand = new MersenneTwister((uint)Seed);
|
---|
136 |
|
---|
137 | // calculates a GAM model using univariate non-linear functions
|
---|
138 | // using backfitting algorithm (see The Elements of Statistical Learning page 298)
|
---|
139 |
|
---|
140 | // init
|
---|
141 | var problemData = Problem.ProblemData;
|
---|
142 | var ds = problemData.Dataset;
|
---|
143 | var trainRows = problemData.TrainingIndices;
|
---|
144 | var testRows = problemData.TestIndices;
|
---|
145 | var avgY = problemData.TargetVariableTrainingValues.Average();
|
---|
146 | var inputVars = problemData.AllowedInputVariables.ToArray();
|
---|
147 |
|
---|
148 | int nTerms = inputVars.Length;
|
---|
149 |
|
---|
150 | #region init results
|
---|
151 | // Set up the results display
|
---|
152 | var iterations = new IntValue(0);
|
---|
153 | Results.Add(new Result("Iterations", iterations));
|
---|
154 |
|
---|
155 | var table = new DataTable("Qualities");
|
---|
156 | var rmseRow = new DataRow("RMSE (train)");
|
---|
157 | var rmseRowTest = new DataRow("RMSE (test)");
|
---|
158 | table.Rows.Add(rmseRow);
|
---|
159 | table.Rows.Add(rmseRowTest);
|
---|
160 | Results.Add(new Result("Qualities", table));
|
---|
161 | var curRMSE = new DoubleValue();
|
---|
162 | var curRMSETest = new DoubleValue();
|
---|
163 | Results.Add(new Result("RMSE (train)", curRMSE));
|
---|
164 | Results.Add(new Result("RMSE (test)", curRMSETest));
|
---|
165 |
|
---|
166 | // calculate table with residual contributions of each term
|
---|
167 | var rssTable = new DoubleMatrix(nTerms, 1, new string[] { "RSS" }, inputVars);
|
---|
168 | Results.Add(new Result("RSS Values", rssTable));
|
---|
169 | #endregion
|
---|
170 |
|
---|
171 | // start with a set of constant models = 0
|
---|
172 | IRegressionModel[] f = new IRegressionModel[nTerms];
|
---|
173 | for (int i = 0; i < f.Length; i++) {
|
---|
174 | f[i] = new ConstantModel(0.0, problemData.TargetVariable);
|
---|
175 | }
|
---|
176 | // init res which contains the current residual vector
|
---|
177 | double[] res = problemData.TargetVariableTrainingValues.Select(yi => yi - avgY).ToArray();
|
---|
178 | double[] resTest = problemData.TargetVariableTestValues.Select(yi => yi - avgY).ToArray();
|
---|
179 |
|
---|
180 | curRMSE.Value = res.StandardDeviation();
|
---|
181 | curRMSETest.Value = resTest.StandardDeviation();
|
---|
182 | rmseRow.Values.Add(res.StandardDeviation());
|
---|
183 | rmseRowTest.Values.Add(resTest.StandardDeviation());
|
---|
184 |
|
---|
185 |
|
---|
186 | double lambda = Lambda;
|
---|
187 | var idx = Enumerable.Range(0, nTerms).ToArray();
|
---|
188 |
|
---|
189 | // Loop until iteration limit reached or canceled.
|
---|
190 | for (int i = 0; i < Iterations && !cancellationToken.IsCancellationRequested; i++) {
|
---|
191 | // shuffle order of terms in each iteration to remove bias on earlier terms
|
---|
192 | idx.ShuffleInPlace(rand);
|
---|
193 | foreach (var inputIdx in idx) {
|
---|
194 | var inputVar = inputVars[inputIdx];
|
---|
195 | // first remove the effect of the previous model for the inputIdx (by adding the output of the current model to the residual)
|
---|
196 | AddInPlace(res, f[inputIdx].GetEstimatedValues(ds, trainRows));
|
---|
197 | AddInPlace(resTest, f[inputIdx].GetEstimatedValues(ds, testRows));
|
---|
198 |
|
---|
199 | rssTable[inputIdx, 0] = res.Variance();
|
---|
200 | f[inputIdx] = RegressSpline(problemData, inputVar, res, lambda);
|
---|
201 |
|
---|
202 | SubtractInPlace(res, f[inputIdx].GetEstimatedValues(ds, trainRows));
|
---|
203 | SubtractInPlace(resTest, f[inputIdx].GetEstimatedValues(ds, testRows));
|
---|
204 | }
|
---|
205 |
|
---|
206 | curRMSE.Value = res.StandardDeviation();
|
---|
207 | curRMSETest.Value = resTest.StandardDeviation();
|
---|
208 | rmseRow.Values.Add(curRMSE.Value);
|
---|
209 | rmseRowTest.Values.Add(curRMSETest.Value);
|
---|
210 | iterations.Value = i;
|
---|
211 | }
|
---|
212 |
|
---|
213 | // produce solution
|
---|
214 | if (CreateSolution) {
|
---|
215 | var model = new RegressionEnsembleModel(f.Concat(new[] { new ConstantModel(avgY, problemData.TargetVariable) }));
|
---|
216 | model.AverageModelEstimates = false;
|
---|
217 | var solution = model.CreateRegressionSolution((IRegressionProblemData)problemData.Clone());
|
---|
218 | Results.Add(new Result("Ensemble solution", solution));
|
---|
219 | }
|
---|
220 | }
|
---|
221 |
|
---|
222 | private IRegressionModel RegressSpline(IRegressionProblemData problemData, string inputVar, double[] target, double lambda) {
|
---|
223 | var x = problemData.Dataset.GetDoubleValues(inputVar, problemData.TrainingIndices).ToArray();
|
---|
224 | var y = (double[])target.Clone();
|
---|
225 | int info;
|
---|
226 | alglib.spline1dinterpolant s;
|
---|
227 | alglib.spline1dfitreport rep;
|
---|
228 | int numKnots = (int)Math.Min(50, 3 * Math.Sqrt(x.Length)); // heuristic for number of knots (Elements of Statistical Learning)
|
---|
229 |
|
---|
230 | alglib.spline1dfitpenalized(x, y, numKnots, lambda, out info, out s, out rep);
|
---|
231 |
|
---|
232 | return new Spline1dModel(s.innerobj, problemData.TargetVariable, inputVar);
|
---|
233 | }
|
---|
234 |
|
---|
235 |
|
---|
236 | private static void AddInPlace(double[] a, IEnumerable<double> enumerable) {
|
---|
237 | int i = 0;
|
---|
238 | foreach (var elem in enumerable) {
|
---|
239 | a[i] += elem;
|
---|
240 | i++;
|
---|
241 | }
|
---|
242 | }
|
---|
243 |
|
---|
244 | private static void SubtractInPlace(double[] a, IEnumerable<double> enumerable) {
|
---|
245 | int i = 0;
|
---|
246 | foreach (var elem in enumerable) {
|
---|
247 | a[i] -= elem;
|
---|
248 | i++;
|
---|
249 | }
|
---|
250 | }
|
---|
251 | }
|
---|
252 | }
|
---|