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 HeuristicLab.Common;
|
---|
26 | using HeuristicLab.Core;
|
---|
27 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
28 | using HeuristicLab.Problems.DataAnalysis;
|
---|
29 |
|
---|
30 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
31 | /// <summary>
|
---|
32 | /// Represents a random forest model for regression and classification
|
---|
33 | /// </summary>
|
---|
34 | [StorableClass]
|
---|
35 | [Item("RandomForestModel", "Represents a random forest for regression and classification.")]
|
---|
36 | public sealed class RandomForestModel : ClassificationModel, IRandomForestModel {
|
---|
37 | // not persisted
|
---|
38 | private alglib.decisionforest randomForest;
|
---|
39 | private alglib.decisionforest RandomForest {
|
---|
40 | get {
|
---|
41 | // recalculate lazily
|
---|
42 | if (randomForest.innerobj.trees == null || randomForest.innerobj.trees.Length == 0) RecalculateModel();
|
---|
43 | return randomForest;
|
---|
44 | }
|
---|
45 | }
|
---|
46 |
|
---|
47 | public override IEnumerable<string> VariablesUsedForPrediction {
|
---|
48 | get { return originalTrainingData.AllowedInputVariables; }
|
---|
49 | }
|
---|
50 |
|
---|
51 |
|
---|
52 | // instead of storing the data of the model itself
|
---|
53 | // we instead only store data necessary to recalculate the same model lazily on demand
|
---|
54 | [Storable]
|
---|
55 | private int seed;
|
---|
56 | [Storable]
|
---|
57 | private IDataAnalysisProblemData originalTrainingData;
|
---|
58 | [Storable]
|
---|
59 | private double[] classValues;
|
---|
60 | [Storable]
|
---|
61 | private int nTrees;
|
---|
62 | [Storable]
|
---|
63 | private double r;
|
---|
64 | [Storable]
|
---|
65 | private double m;
|
---|
66 |
|
---|
67 |
|
---|
68 | [StorableConstructor]
|
---|
69 | private RandomForestModel(bool deserializing)
|
---|
70 | : base(deserializing) {
|
---|
71 | // for backwards compatibility (loading old solutions)
|
---|
72 | randomForest = new alglib.decisionforest();
|
---|
73 | }
|
---|
74 | private RandomForestModel(RandomForestModel original, Cloner cloner)
|
---|
75 | : base(original, cloner) {
|
---|
76 | randomForest = new alglib.decisionforest();
|
---|
77 | randomForest.innerobj.bufsize = original.randomForest.innerobj.bufsize;
|
---|
78 | randomForest.innerobj.nclasses = original.randomForest.innerobj.nclasses;
|
---|
79 | randomForest.innerobj.ntrees = original.randomForest.innerobj.ntrees;
|
---|
80 | randomForest.innerobj.nvars = original.randomForest.innerobj.nvars;
|
---|
81 | // we assume that the trees array (double[]) is immutable in alglib
|
---|
82 | randomForest.innerobj.trees = original.randomForest.innerobj.trees;
|
---|
83 |
|
---|
84 | // allowedInputVariables is immutable so we don't need to clone
|
---|
85 | allowedInputVariables = original.allowedInputVariables;
|
---|
86 |
|
---|
87 | // clone data which is necessary to rebuild the model
|
---|
88 | this.seed = original.seed;
|
---|
89 | this.originalTrainingData = cloner.Clone(original.originalTrainingData);
|
---|
90 | // classvalues is immutable so we don't need to clone
|
---|
91 | this.classValues = original.classValues;
|
---|
92 | this.nTrees = original.nTrees;
|
---|
93 | this.r = original.r;
|
---|
94 | this.m = original.m;
|
---|
95 | }
|
---|
96 |
|
---|
97 | // random forest models can only be created through the static factory methods CreateRegressionModel and CreateClassificationModel
|
---|
98 | private RandomForestModel(string targetVariable, alglib.decisionforest randomForest,
|
---|
99 | int seed, IDataAnalysisProblemData originalTrainingData,
|
---|
100 | int nTrees, double r, double m, double[] classValues = null)
|
---|
101 | : base(targetVariable) {
|
---|
102 | this.name = ItemName;
|
---|
103 | this.description = ItemDescription;
|
---|
104 | // the model itself
|
---|
105 | this.randomForest = randomForest;
|
---|
106 | // data which is necessary for recalculation of the model
|
---|
107 | this.seed = seed;
|
---|
108 | this.originalTrainingData = (IDataAnalysisProblemData)originalTrainingData.Clone();
|
---|
109 | this.classValues = classValues;
|
---|
110 | this.nTrees = nTrees;
|
---|
111 | this.r = r;
|
---|
112 | this.m = m;
|
---|
113 | }
|
---|
114 |
|
---|
115 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
116 | return new RandomForestModel(this, cloner);
|
---|
117 | }
|
---|
118 |
|
---|
119 | private void RecalculateModel() {
|
---|
120 | double rmsError, oobRmsError, relClassError, oobRelClassError;
|
---|
121 | var regressionProblemData = originalTrainingData as IRegressionProblemData;
|
---|
122 | var classificationProblemData = originalTrainingData as IClassificationProblemData;
|
---|
123 | if (regressionProblemData != null) {
|
---|
124 | var model = CreateRegressionModel(regressionProblemData,
|
---|
125 | nTrees, r, m, seed, out rmsError, out oobRmsError,
|
---|
126 | out relClassError, out oobRelClassError);
|
---|
127 | randomForest = model.randomForest;
|
---|
128 | } else if (classificationProblemData != null) {
|
---|
129 | var model = CreateClassificationModel(classificationProblemData,
|
---|
130 | nTrees, r, m, seed, out rmsError, out oobRmsError,
|
---|
131 | out relClassError, out oobRelClassError);
|
---|
132 | randomForest = model.randomForest;
|
---|
133 | }
|
---|
134 | }
|
---|
135 |
|
---|
136 | public IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {
|
---|
137 | double[,] inputData = AlglibUtil.PrepareInputMatrix(dataset, AllowedInputVariables, rows);
|
---|
138 | AssertInputMatrix(inputData);
|
---|
139 |
|
---|
140 | int n = inputData.GetLength(0);
|
---|
141 | int columns = inputData.GetLength(1);
|
---|
142 | double[] x = new double[columns];
|
---|
143 | double[] y = new double[1];
|
---|
144 |
|
---|
145 | for (int row = 0; row < n; row++) {
|
---|
146 | for (int column = 0; column < columns; column++) {
|
---|
147 | x[column] = inputData[row, column];
|
---|
148 | }
|
---|
149 | alglib.dfprocess(RandomForest, x, ref y);
|
---|
150 | yield return y[0];
|
---|
151 | }
|
---|
152 | }
|
---|
153 |
|
---|
154 | public IEnumerable<double> GetEstimatedVariances(IDataset dataset, IEnumerable<int> rows) {
|
---|
155 | double[,] inputData = AlglibUtil.PrepareInputMatrix(dataset, AllowedInputVariables, rows);
|
---|
156 | AssertInputMatrix(inputData);
|
---|
157 |
|
---|
158 | int n = inputData.GetLength(0);
|
---|
159 | int columns = inputData.GetLength(1);
|
---|
160 | double[] x = new double[columns];
|
---|
161 | double[] ys = new double[this.RandomForest.innerobj.ntrees];
|
---|
162 |
|
---|
163 | for (int row = 0; row < n; row++) {
|
---|
164 | for (int column = 0; column < columns; column++) {
|
---|
165 | x[column] = inputData[row, column];
|
---|
166 | }
|
---|
167 | alglib.dforest.dfprocessraw(RandomForest.innerobj, x, ref ys);
|
---|
168 | yield return ys.VariancePop();
|
---|
169 | }
|
---|
170 | }
|
---|
171 |
|
---|
172 | public override IEnumerable<double> GetEstimatedClassValues(IDataset dataset, IEnumerable<int> rows) {
|
---|
173 | double[,] inputData = AlglibUtil.PrepareInputMatrix(dataset, AllowedInputVariables, rows);
|
---|
174 | AssertInputMatrix(inputData);
|
---|
175 |
|
---|
176 | int n = inputData.GetLength(0);
|
---|
177 | int columns = inputData.GetLength(1);
|
---|
178 | double[] x = new double[columns];
|
---|
179 | double[] y = new double[RandomForest.innerobj.nclasses];
|
---|
180 |
|
---|
181 | for (int row = 0; row < n; row++) {
|
---|
182 | for (int column = 0; column < columns; column++) {
|
---|
183 | x[column] = inputData[row, column];
|
---|
184 | }
|
---|
185 | alglib.dfprocess(randomForest, x, ref y);
|
---|
186 | // find class for with the largest probability value
|
---|
187 | int maxProbClassIndex = 0;
|
---|
188 | double maxProb = y[0];
|
---|
189 | for (int i = 1; i < y.Length; i++) {
|
---|
190 | if (maxProb < y[i]) {
|
---|
191 | maxProb = y[i];
|
---|
192 | maxProbClassIndex = i;
|
---|
193 | }
|
---|
194 | }
|
---|
195 | yield return classValues[maxProbClassIndex];
|
---|
196 | }
|
---|
197 | }
|
---|
198 |
|
---|
199 |
|
---|
200 | public IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) {
|
---|
201 | return new RandomForestRegressionSolution(this, new RegressionProblemData(problemData));
|
---|
202 | }
|
---|
203 | public override IClassificationSolution CreateClassificationSolution(IClassificationProblemData problemData) {
|
---|
204 | return new RandomForestClassificationSolution(this, new ClassificationProblemData(problemData));
|
---|
205 | }
|
---|
206 |
|
---|
207 | public static RandomForestModel CreateRegressionModel(IRegressionProblemData problemData, int nTrees, double r, double m, int seed,
|
---|
208 | out double rmsError, out double outOfBagRmsError, out double avgRelError, out double outOfBagAvgRelError) {
|
---|
209 | return CreateRegressionModel(problemData, problemData.TrainingIndices, nTrees, r, m, seed, out rmsError, out avgRelError, out outOfBagAvgRelError, out outOfBagRmsError);
|
---|
210 | }
|
---|
211 |
|
---|
212 | public static RandomForestModel CreateRegressionModel(IRegressionProblemData problemData, IEnumerable<int> trainingIndices, int nTrees, double r, double m, int seed,
|
---|
213 | out double rmsError, out double outOfBagRmsError, out double avgRelError, out double outOfBagAvgRelError) {
|
---|
214 | var variables = problemData.AllowedInputVariables.Concat(new string[] { problemData.TargetVariable });
|
---|
215 | double[,] inputMatrix = AlglibUtil.PrepareInputMatrix(problemData.Dataset, variables, trainingIndices);
|
---|
216 |
|
---|
217 | alglib.dfreport rep;
|
---|
218 | var dForest = CreateRandomForestModel(seed, inputMatrix, nTrees, r, m, 1, out rep);
|
---|
219 |
|
---|
220 | rmsError = rep.rmserror;
|
---|
221 | avgRelError = rep.avgrelerror;
|
---|
222 | outOfBagAvgRelError = rep.oobavgrelerror;
|
---|
223 | outOfBagRmsError = rep.oobrmserror;
|
---|
224 |
|
---|
225 | return new RandomForestModel(problemData.TargetVariable, dForest, seed, problemData, nTrees, r, m);
|
---|
226 | }
|
---|
227 |
|
---|
228 | public static RandomForestModel CreateClassificationModel(IClassificationProblemData problemData, int nTrees, double r, double m, int seed,
|
---|
229 | out double rmsError, out double outOfBagRmsError, out double relClassificationError, out double outOfBagRelClassificationError) {
|
---|
230 | return CreateClassificationModel(problemData, problemData.TrainingIndices, nTrees, r, m, seed, out rmsError, out outOfBagRmsError, out relClassificationError, out outOfBagRelClassificationError);
|
---|
231 | }
|
---|
232 |
|
---|
233 | public static RandomForestModel CreateClassificationModel(IClassificationProblemData problemData, IEnumerable<int> trainingIndices, int nTrees, double r, double m, int seed,
|
---|
234 | out double rmsError, out double outOfBagRmsError, out double relClassificationError, out double outOfBagRelClassificationError) {
|
---|
235 |
|
---|
236 | var variables = problemData.AllowedInputVariables.Concat(new string[] { problemData.TargetVariable });
|
---|
237 | double[,] inputMatrix = AlglibUtil.PrepareInputMatrix(problemData.Dataset, variables, trainingIndices);
|
---|
238 |
|
---|
239 | var classValues = problemData.ClassValues.ToArray();
|
---|
240 | int nClasses = classValues.Length;
|
---|
241 |
|
---|
242 | // map original class values to values [0..nClasses-1]
|
---|
243 | var classIndices = new Dictionary<double, double>();
|
---|
244 | for (int i = 0; i < nClasses; i++) {
|
---|
245 | classIndices[classValues[i]] = i;
|
---|
246 | }
|
---|
247 |
|
---|
248 | int nRows = inputMatrix.GetLength(0);
|
---|
249 | int nColumns = inputMatrix.GetLength(1);
|
---|
250 | for (int row = 0; row < nRows; row++) {
|
---|
251 | inputMatrix[row, nColumns - 1] = classIndices[inputMatrix[row, nColumns - 1]];
|
---|
252 | }
|
---|
253 |
|
---|
254 | alglib.dfreport rep;
|
---|
255 | var dForest = CreateRandomForestModel(seed, inputMatrix, nTrees, r, m, nClasses, out rep);
|
---|
256 |
|
---|
257 | rmsError = rep.rmserror;
|
---|
258 | outOfBagRmsError = rep.oobrmserror;
|
---|
259 | relClassificationError = rep.relclserror;
|
---|
260 | outOfBagRelClassificationError = rep.oobrelclserror;
|
---|
261 |
|
---|
262 | return new RandomForestModel(problemData.TargetVariable, dForest, seed, problemData, nTrees, r, m, classValues);
|
---|
263 | }
|
---|
264 |
|
---|
265 | private static alglib.decisionforest CreateRandomForestModel(int seed, double[,] inputMatrix, int nTrees, double r, double m, int nClasses, out alglib.dfreport rep) {
|
---|
266 | AssertParameters(r, m);
|
---|
267 | AssertInputMatrix(inputMatrix);
|
---|
268 |
|
---|
269 | int info = 0;
|
---|
270 | alglib.math.rndobject = new System.Random(seed);
|
---|
271 | var dForest = new alglib.decisionforest();
|
---|
272 | rep = new alglib.dfreport();
|
---|
273 | int nRows = inputMatrix.GetLength(0);
|
---|
274 | int nColumns = inputMatrix.GetLength(1);
|
---|
275 | int sampleSize = Math.Max((int)Math.Round(r * nRows), 1);
|
---|
276 | int nFeatures = Math.Max((int)Math.Round(m * (nColumns - 1)), 1);
|
---|
277 |
|
---|
278 | alglib.dforest.dfbuildinternal(inputMatrix, nRows, nColumns - 1, nClasses, nTrees, sampleSize, nFeatures, alglib.dforest.dfusestrongsplits + alglib.dforest.dfuseevs, ref info, dForest.innerobj, rep.innerobj);
|
---|
279 | if (info != 1) throw new ArgumentException("Error in calculation of random forest model");
|
---|
280 | return dForest;
|
---|
281 | }
|
---|
282 |
|
---|
283 | private static void AssertParameters(double r, double m) {
|
---|
284 | if (r <= 0 || r > 1) throw new ArgumentException("The R parameter for random forest modeling must be between 0 and 1.");
|
---|
285 | if (m <= 0 || m > 1) throw new ArgumentException("The M parameter for random forest modeling must be between 0 and 1.");
|
---|
286 | }
|
---|
287 |
|
---|
288 | private static void AssertInputMatrix(double[,] inputMatrix) {
|
---|
289 | if (inputMatrix.Cast<double>().Any(x => Double.IsNaN(x) || Double.IsInfinity(x)))
|
---|
290 | throw new NotSupportedException("Random forest modeling does not support NaN or infinity values in the input dataset.");
|
---|
291 | }
|
---|
292 |
|
---|
293 | #region persistence for backwards compatibility
|
---|
294 | // when the originalTrainingData is null this means the model was loaded from an old file
|
---|
295 | // therefore, we cannot use the new persistence mechanism because the original data is not available anymore
|
---|
296 | // in such cases we still store the compete model
|
---|
297 | private bool IsCompatibilityLoaded { get { return originalTrainingData == null; } }
|
---|
298 |
|
---|
299 | private string[] allowedInputVariables;
|
---|
300 | [Storable(Name = "allowedInputVariables")]
|
---|
301 | private string[] AllowedInputVariables {
|
---|
302 | get {
|
---|
303 | if (IsCompatibilityLoaded) return allowedInputVariables;
|
---|
304 | else return originalTrainingData.AllowedInputVariables.ToArray();
|
---|
305 | }
|
---|
306 | set { allowedInputVariables = value; }
|
---|
307 | }
|
---|
308 | [Storable]
|
---|
309 | private int RandomForestBufSize {
|
---|
310 | get {
|
---|
311 | if (IsCompatibilityLoaded) return randomForest.innerobj.bufsize;
|
---|
312 | else return 0;
|
---|
313 | }
|
---|
314 | set {
|
---|
315 | randomForest.innerobj.bufsize = value;
|
---|
316 | }
|
---|
317 | }
|
---|
318 | [Storable]
|
---|
319 | private int RandomForestNClasses {
|
---|
320 | get {
|
---|
321 | if (IsCompatibilityLoaded) return randomForest.innerobj.nclasses;
|
---|
322 | else return 0;
|
---|
323 | }
|
---|
324 | set {
|
---|
325 | randomForest.innerobj.nclasses = value;
|
---|
326 | }
|
---|
327 | }
|
---|
328 | [Storable]
|
---|
329 | private int RandomForestNTrees {
|
---|
330 | get {
|
---|
331 | if (IsCompatibilityLoaded) return randomForest.innerobj.ntrees;
|
---|
332 | else return 0;
|
---|
333 | }
|
---|
334 | set {
|
---|
335 | randomForest.innerobj.ntrees = value;
|
---|
336 | }
|
---|
337 | }
|
---|
338 | [Storable]
|
---|
339 | private int RandomForestNVars {
|
---|
340 | get {
|
---|
341 | if (IsCompatibilityLoaded) return randomForest.innerobj.nvars;
|
---|
342 | else return 0;
|
---|
343 | }
|
---|
344 | set {
|
---|
345 | randomForest.innerobj.nvars = value;
|
---|
346 | }
|
---|
347 | }
|
---|
348 | [Storable]
|
---|
349 | private double[] RandomForestTrees {
|
---|
350 | get {
|
---|
351 | if (IsCompatibilityLoaded) return randomForest.innerobj.trees;
|
---|
352 | else return new double[] { };
|
---|
353 | }
|
---|
354 | set {
|
---|
355 | randomForest.innerobj.trees = value;
|
---|
356 | }
|
---|
357 | }
|
---|
358 | #endregion
|
---|
359 | }
|
---|
360 | }
|
---|