[11315] | 1 | #region License Information
|
---|
| 2 |
|
---|
| 3 | /* HeuristicLab
|
---|
[12009] | 4 | * Copyright (C) 2002-2015 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[11315] | 5 | *
|
---|
| 6 | * This file is part of HeuristicLab.
|
---|
| 7 | *
|
---|
| 8 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
| 9 | * it under the terms of the GNU General Public License as published by
|
---|
| 10 | * the Free Software Foundation, either version 3 of the License, or
|
---|
| 11 | * (at your option) any later version.
|
---|
| 12 | *
|
---|
| 13 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 16 | * GNU General Public License for more details.
|
---|
| 17 | *
|
---|
| 18 | * You should have received a copy of the GNU General Public License
|
---|
| 19 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 20 | */
|
---|
| 21 |
|
---|
| 22 | #endregion
|
---|
| 23 |
|
---|
| 24 | using System;
|
---|
| 25 | using System.Collections.Generic;
|
---|
| 26 | using System.Linq;
|
---|
| 27 | using System.Linq.Expressions;
|
---|
| 28 | using System.Threading.Tasks;
|
---|
| 29 | using HeuristicLab.Common;
|
---|
[11901] | 30 | using HeuristicLab.Core;
|
---|
[11315] | 31 | using HeuristicLab.Data;
|
---|
[11901] | 32 | using HeuristicLab.Parameters;
|
---|
| 33 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
[11315] | 34 | using HeuristicLab.Problems.DataAnalysis;
|
---|
[11901] | 35 | using HeuristicLab.Random;
|
---|
[11315] | 36 |
|
---|
| 37 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
[11901] | 38 | [Item("RFParameter", "A random forest parameter collection")]
|
---|
| 39 | [StorableClass]
|
---|
| 40 | public class RFParameter : ParameterCollection {
|
---|
| 41 | public RFParameter() {
|
---|
| 42 | base.Add(new FixedValueParameter<IntValue>("N", "The number of random forest trees", new IntValue(50)));
|
---|
| 43 | base.Add(new FixedValueParameter<DoubleValue>("M", "The ratio of features that will be used in the construction of individual trees (0<m<=1)", new DoubleValue(0.1)));
|
---|
| 44 | base.Add(new FixedValueParameter<DoubleValue>("R", "The ratio of the training set that will be used in the construction of individual trees (0<r<=1)", new DoubleValue(0.1)));
|
---|
| 45 | }
|
---|
[11315] | 46 |
|
---|
[11901] | 47 | [StorableConstructor]
|
---|
| 48 | protected RFParameter(bool deserializing)
|
---|
| 49 | : base(deserializing) {
|
---|
| 50 | }
|
---|
[11315] | 51 |
|
---|
[11901] | 52 | protected RFParameter(RFParameter original, Cloner cloner)
|
---|
| 53 | : base(original, cloner) {
|
---|
| 54 | this.N = original.N;
|
---|
| 55 | this.R = original.R;
|
---|
| 56 | this.M = original.M;
|
---|
| 57 | }
|
---|
[11315] | 58 |
|
---|
[11901] | 59 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
| 60 | return new RFParameter(this, cloner);
|
---|
[11315] | 61 | }
|
---|
| 62 |
|
---|
[11901] | 63 | private IFixedValueParameter<IntValue> NParameter {
|
---|
| 64 | get { return (IFixedValueParameter<IntValue>)base["N"]; }
|
---|
| 65 | }
|
---|
[11315] | 66 |
|
---|
[11901] | 67 | private IFixedValueParameter<DoubleValue> RParameter {
|
---|
| 68 | get { return (IFixedValueParameter<DoubleValue>)base["R"]; }
|
---|
| 69 | }
|
---|
[11315] | 70 |
|
---|
[11901] | 71 | private IFixedValueParameter<DoubleValue> MParameter {
|
---|
| 72 | get { return (IFixedValueParameter<DoubleValue>)base["M"]; }
|
---|
[11315] | 73 | }
|
---|
| 74 |
|
---|
[11901] | 75 | public int N {
|
---|
| 76 | get { return NParameter.Value.Value; }
|
---|
| 77 | set { NParameter.Value.Value = value; }
|
---|
[11315] | 78 | }
|
---|
| 79 |
|
---|
[11901] | 80 | public double R {
|
---|
| 81 | get { return RParameter.Value.Value; }
|
---|
| 82 | set { RParameter.Value.Value = value; }
|
---|
| 83 | }
|
---|
[11315] | 84 |
|
---|
[11901] | 85 | public double M {
|
---|
| 86 | get { return MParameter.Value.Value; }
|
---|
| 87 | set { MParameter.Value.Value = value; }
|
---|
| 88 | }
|
---|
| 89 | }
|
---|
[11315] | 90 |
|
---|
[11901] | 91 | public static class RandomForestUtil {
|
---|
| 92 | private static readonly object locker = new object();
|
---|
[11315] | 93 |
|
---|
[11901] | 94 | private static void CrossValidate(IRegressionProblemData problemData, Tuple<IEnumerable<int>, IEnumerable<int>>[] partitions, int nTrees, double r, double m, int seed, out double avgTestMse) {
|
---|
| 95 | avgTestMse = 0;
|
---|
| 96 | var ds = problemData.Dataset;
|
---|
| 97 | var targetVariable = GetTargetVariableName(problemData);
|
---|
| 98 | foreach (var tuple in partitions) {
|
---|
[11315] | 99 | double rmsError, avgRelError, outOfBagAvgRelError, outOfBagRmsError;
|
---|
[11901] | 100 | var trainingRandomForestPartition = tuple.Item1;
|
---|
| 101 | var testRandomForestPartition = tuple.Item2;
|
---|
| 102 | var model = RandomForestModel.CreateRegressionModel(problemData, trainingRandomForestPartition, nTrees, r, m, seed, out rmsError, out avgRelError, out outOfBagRmsError, out outOfBagAvgRelError);
|
---|
| 103 | var estimatedValues = model.GetEstimatedValues(ds, testRandomForestPartition);
|
---|
| 104 | var targetValues = ds.GetDoubleValues(targetVariable, testRandomForestPartition);
|
---|
[11315] | 105 | OnlineCalculatorError calculatorError;
|
---|
[11901] | 106 | double mse = OnlineMeanSquaredErrorCalculator.Calculate(estimatedValues, targetValues, out calculatorError);
|
---|
[11315] | 107 | if (calculatorError != OnlineCalculatorError.None)
|
---|
| 108 | mse = double.NaN;
|
---|
[11901] | 109 | avgTestMse += mse;
|
---|
[11315] | 110 | }
|
---|
[11901] | 111 | avgTestMse /= partitions.Length;
|
---|
| 112 | }
|
---|
[11315] | 113 |
|
---|
[11901] | 114 | private static void CrossValidate(IClassificationProblemData problemData, Tuple<IEnumerable<int>, IEnumerable<int>>[] partitions, int nTrees, double r, double m, int seed, out double avgTestAccuracy) {
|
---|
| 115 | avgTestAccuracy = 0;
|
---|
| 116 | var ds = problemData.Dataset;
|
---|
| 117 | var targetVariable = GetTargetVariableName(problemData);
|
---|
| 118 | foreach (var tuple in partitions) {
|
---|
| 119 | double rmsError, avgRelError, outOfBagAvgRelError, outOfBagRmsError;
|
---|
| 120 | var trainingRandomForestPartition = tuple.Item1;
|
---|
| 121 | var testRandomForestPartition = tuple.Item2;
|
---|
| 122 | var model = RandomForestModel.CreateClassificationModel(problemData, trainingRandomForestPartition, nTrees, r, m, seed, out rmsError, out avgRelError, out outOfBagRmsError, out outOfBagAvgRelError);
|
---|
| 123 | var estimatedValues = model.GetEstimatedClassValues(ds, testRandomForestPartition);
|
---|
| 124 | var targetValues = ds.GetDoubleValues(targetVariable, testRandomForestPartition);
|
---|
| 125 | OnlineCalculatorError calculatorError;
|
---|
| 126 | double accuracy = OnlineAccuracyCalculator.Calculate(estimatedValues, targetValues, out calculatorError);
|
---|
| 127 | if (calculatorError != OnlineCalculatorError.None)
|
---|
| 128 | accuracy = double.NaN;
|
---|
| 129 | avgTestAccuracy += accuracy;
|
---|
| 130 | }
|
---|
| 131 | avgTestAccuracy /= partitions.Length;
|
---|
[11315] | 132 | }
|
---|
| 133 |
|
---|
[11901] | 134 | // grid search without cross-validation since in the case of random forests, the out-of-bag estimate is unbiased
|
---|
| 135 | public static RFParameter GridSearch(IRegressionProblemData problemData, Dictionary<string, IEnumerable<double>> parameterRanges, int seed = 12345, int maxDegreeOfParallelism = 1) {
|
---|
| 136 | var setters = parameterRanges.Keys.Select(GenerateSetter).ToList();
|
---|
| 137 | var crossProduct = parameterRanges.Values.CartesianProduct();
|
---|
| 138 | double bestOutOfBagRmsError = double.MaxValue;
|
---|
| 139 | RFParameter bestParameters = new RFParameter();
|
---|
| 140 |
|
---|
| 141 | Parallel.ForEach(crossProduct, new ParallelOptions { MaxDegreeOfParallelism = maxDegreeOfParallelism }, parameterCombination => {
|
---|
| 142 | var parameterValues = parameterCombination.ToList();
|
---|
| 143 | var parameters = new RFParameter();
|
---|
| 144 | for (int i = 0; i < setters.Count; ++i) { setters[i](parameters, parameterValues[i]); }
|
---|
| 145 | double rmsError, outOfBagRmsError, avgRelError, outOfBagAvgRelError;
|
---|
| 146 | RandomForestModel.CreateRegressionModel(problemData, problemData.TrainingIndices, parameters.N, parameters.R, parameters.M, seed, out rmsError, out outOfBagRmsError, out avgRelError, out outOfBagAvgRelError);
|
---|
| 147 |
|
---|
| 148 | lock (locker) {
|
---|
| 149 | if (bestOutOfBagRmsError > outOfBagRmsError) {
|
---|
| 150 | bestOutOfBagRmsError = outOfBagRmsError;
|
---|
| 151 | bestParameters = (RFParameter)parameters.Clone();
|
---|
| 152 | }
|
---|
| 153 | }
|
---|
| 154 | });
|
---|
| 155 | return bestParameters;
|
---|
| 156 | }
|
---|
| 157 |
|
---|
| 158 | public static RFParameter GridSearch(IClassificationProblemData problemData, Dictionary<string, IEnumerable<double>> parameterRanges, int seed = 12345, int maxDegreeOfParallelism = 1) {
|
---|
| 159 | var setters = parameterRanges.Keys.Select(GenerateSetter).ToList();
|
---|
| 160 | var crossProduct = parameterRanges.Values.CartesianProduct();
|
---|
| 161 |
|
---|
| 162 | double bestOutOfBagRmsError = double.MaxValue;
|
---|
| 163 | RFParameter bestParameters = new RFParameter();
|
---|
| 164 |
|
---|
| 165 | Parallel.ForEach(crossProduct, new ParallelOptions { MaxDegreeOfParallelism = maxDegreeOfParallelism }, parameterCombination => {
|
---|
| 166 | var parameterValues = parameterCombination.ToList();
|
---|
| 167 | var parameters = new RFParameter();
|
---|
| 168 | for (int i = 0; i < setters.Count; ++i) { setters[i](parameters, parameterValues[i]); }
|
---|
| 169 | double rmsError, outOfBagRmsError, avgRelError, outOfBagAvgRelError;
|
---|
| 170 | RandomForestModel.CreateClassificationModel(problemData, problemData.TrainingIndices, parameters.N, parameters.R, parameters.M, seed,
|
---|
| 171 | out rmsError, out outOfBagRmsError, out avgRelError, out outOfBagAvgRelError);
|
---|
| 172 |
|
---|
| 173 | lock (locker) {
|
---|
| 174 | if (bestOutOfBagRmsError > outOfBagRmsError) {
|
---|
| 175 | bestOutOfBagRmsError = outOfBagRmsError;
|
---|
| 176 | bestParameters = (RFParameter)parameters.Clone();
|
---|
| 177 | }
|
---|
| 178 | }
|
---|
| 179 | });
|
---|
| 180 | return bestParameters;
|
---|
| 181 | }
|
---|
| 182 |
|
---|
| 183 | public static RFParameter GridSearch(IRegressionProblemData problemData, int numberOfFolds, bool shuffleFolds, Dictionary<string, IEnumerable<double>> parameterRanges, int seed = 12345, int maxDegreeOfParallelism = 1) {
|
---|
[11315] | 184 | DoubleValue mse = new DoubleValue(Double.MaxValue);
|
---|
[11901] | 185 | RFParameter bestParameter = new RFParameter();
|
---|
[11315] | 186 |
|
---|
[11901] | 187 | var setters = parameterRanges.Keys.Select(GenerateSetter).ToList();
|
---|
| 188 | var partitions = GenerateRandomForestPartitions(problemData, numberOfFolds);
|
---|
| 189 | var crossProduct = parameterRanges.Values.CartesianProduct();
|
---|
[11315] | 190 |
|
---|
[11901] | 191 | Parallel.ForEach(crossProduct, new ParallelOptions { MaxDegreeOfParallelism = maxDegreeOfParallelism }, parameterCombination => {
|
---|
| 192 | var parameterValues = parameterCombination.ToList();
|
---|
[11315] | 193 | double testMSE;
|
---|
| 194 | var parameters = new RFParameter();
|
---|
[11901] | 195 | for (int i = 0; i < setters.Count; ++i) {
|
---|
| 196 | setters[i](parameters, parameterValues[i]);
|
---|
[11315] | 197 | }
|
---|
[11901] | 198 | CrossValidate(problemData, partitions, parameters.N, parameters.R, parameters.M, seed, out testMSE);
|
---|
| 199 |
|
---|
| 200 | lock (locker) {
|
---|
| 201 | if (testMSE < mse.Value) {
|
---|
[11315] | 202 | mse.Value = testMSE;
|
---|
[11901] | 203 | bestParameter = (RFParameter)parameters.Clone();
|
---|
[11315] | 204 | }
|
---|
[11901] | 205 | }
|
---|
| 206 | });
|
---|
| 207 | return bestParameter;
|
---|
| 208 | }
|
---|
| 209 |
|
---|
| 210 | public static RFParameter GridSearch(IClassificationProblemData problemData, int numberOfFolds, bool shuffleFolds, Dictionary<string, IEnumerable<double>> parameterRanges, int seed = 12345, int maxDegreeOfParallelism = 1) {
|
---|
| 211 | DoubleValue accuracy = new DoubleValue(0);
|
---|
| 212 | RFParameter bestParameter = new RFParameter();
|
---|
| 213 |
|
---|
| 214 | var setters = parameterRanges.Keys.Select(GenerateSetter).ToList();
|
---|
| 215 | var crossProduct = parameterRanges.Values.CartesianProduct();
|
---|
| 216 | var partitions = GenerateRandomForestPartitions(problemData, numberOfFolds, shuffleFolds);
|
---|
| 217 |
|
---|
| 218 | Parallel.ForEach(crossProduct, new ParallelOptions { MaxDegreeOfParallelism = maxDegreeOfParallelism }, parameterCombination => {
|
---|
| 219 | var parameterValues = parameterCombination.ToList();
|
---|
| 220 | double testAccuracy;
|
---|
| 221 | var parameters = new RFParameter();
|
---|
| 222 | for (int i = 0; i < setters.Count; ++i) {
|
---|
| 223 | setters[i](parameters, parameterValues[i]);
|
---|
| 224 | }
|
---|
| 225 | CrossValidate(problemData, partitions, parameters.N, parameters.R, parameters.M, seed, out testAccuracy);
|
---|
| 226 |
|
---|
| 227 | lock (locker) {
|
---|
| 228 | if (testAccuracy > accuracy.Value) {
|
---|
| 229 | accuracy.Value = testAccuracy;
|
---|
[11315] | 230 | bestParameter = (RFParameter)parameters.Clone();
|
---|
| 231 | }
|
---|
| 232 | }
|
---|
| 233 | });
|
---|
| 234 | return bestParameter;
|
---|
| 235 | }
|
---|
[11901] | 236 |
|
---|
| 237 | private static Tuple<IEnumerable<int>, IEnumerable<int>>[] GenerateRandomForestPartitions(IDataAnalysisProblemData problemData, int numberOfFolds, bool shuffleFolds = false) {
|
---|
| 238 | var folds = GenerateFolds(problemData, numberOfFolds, shuffleFolds).ToList();
|
---|
| 239 | var partitions = new Tuple<IEnumerable<int>, IEnumerable<int>>[numberOfFolds];
|
---|
| 240 |
|
---|
| 241 | for (int i = 0; i < numberOfFolds; ++i) {
|
---|
| 242 | int p = i; // avoid "access to modified closure" warning
|
---|
| 243 | var trainingRows = folds.SelectMany((par, j) => j != p ? par : Enumerable.Empty<int>());
|
---|
| 244 | var testRows = folds[i];
|
---|
| 245 | partitions[i] = new Tuple<IEnumerable<int>, IEnumerable<int>>(trainingRows, testRows);
|
---|
| 246 | }
|
---|
| 247 | return partitions;
|
---|
| 248 | }
|
---|
| 249 |
|
---|
| 250 | public static IEnumerable<IEnumerable<int>> GenerateFolds(IDataAnalysisProblemData problemData, int numberOfFolds, bool shuffleFolds = false) {
|
---|
| 251 | var random = new MersenneTwister((uint)Environment.TickCount);
|
---|
| 252 | if (problemData is IRegressionProblemData) {
|
---|
| 253 | var trainingIndices = shuffleFolds ? problemData.TrainingIndices.OrderBy(x => random.Next()) : problemData.TrainingIndices;
|
---|
| 254 | return GenerateFolds(trainingIndices, problemData.TrainingPartition.Size, numberOfFolds);
|
---|
| 255 | }
|
---|
| 256 | if (problemData is IClassificationProblemData) {
|
---|
| 257 | // when shuffle is enabled do stratified folds generation, some folds may have zero elements
|
---|
| 258 | // otherwise, generate folds normally
|
---|
| 259 | return shuffleFolds ? GenerateFoldsStratified(problemData as IClassificationProblemData, numberOfFolds, random) : GenerateFolds(problemData.TrainingIndices, problemData.TrainingPartition.Size, numberOfFolds);
|
---|
| 260 | }
|
---|
| 261 | throw new ArgumentException("Problem data is neither regression or classification problem data.");
|
---|
| 262 | }
|
---|
| 263 |
|
---|
| 264 | /// <summary>
|
---|
| 265 | /// Stratified fold generation from classification data. Stratification means that we ensure the same distribution of class labels for each fold.
|
---|
| 266 | /// The samples are grouped by class label and each group is split into @numberOfFolds parts. The final folds are formed from the joining of
|
---|
| 267 | /// the corresponding parts from each class label.
|
---|
| 268 | /// </summary>
|
---|
| 269 | /// <param name="problemData">The classification problem data.</param>
|
---|
| 270 | /// <param name="numberOfFolds">The number of folds in which to split the data.</param>
|
---|
| 271 | /// <param name="random">The random generator used to shuffle the folds.</param>
|
---|
| 272 | /// <returns>An enumerable sequece of folds, where a fold is represented by a sequence of row indices.</returns>
|
---|
| 273 | private static IEnumerable<IEnumerable<int>> GenerateFoldsStratified(IClassificationProblemData problemData, int numberOfFolds, IRandom random) {
|
---|
| 274 | var values = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices);
|
---|
| 275 | var valuesIndices = problemData.TrainingIndices.Zip(values, (i, v) => new { Index = i, Value = v }).ToList();
|
---|
| 276 | IEnumerable<IEnumerable<IEnumerable<int>>> foldsByClass = valuesIndices.GroupBy(x => x.Value, x => x.Index).Select(g => GenerateFolds(g, g.Count(), numberOfFolds));
|
---|
| 277 | var enumerators = foldsByClass.Select(f => f.GetEnumerator()).ToList();
|
---|
| 278 | while (enumerators.All(e => e.MoveNext())) {
|
---|
| 279 | yield return enumerators.SelectMany(e => e.Current).OrderBy(x => random.Next()).ToList();
|
---|
| 280 | }
|
---|
| 281 | }
|
---|
| 282 |
|
---|
| 283 | private static IEnumerable<IEnumerable<T>> GenerateFolds<T>(IEnumerable<T> values, int valuesCount, int numberOfFolds) {
|
---|
| 284 | // if number of folds is greater than the number of values, some empty folds will be returned
|
---|
| 285 | if (valuesCount < numberOfFolds) {
|
---|
| 286 | for (int i = 0; i < numberOfFolds; ++i)
|
---|
| 287 | yield return i < valuesCount ? values.Skip(i).Take(1) : Enumerable.Empty<T>();
|
---|
| 288 | } else {
|
---|
| 289 | int f = valuesCount / numberOfFolds, r = valuesCount % numberOfFolds; // number of folds rounded to integer and remainder
|
---|
| 290 | int start = 0, end = f;
|
---|
| 291 | for (int i = 0; i < numberOfFolds; ++i) {
|
---|
| 292 | if (r > 0) {
|
---|
| 293 | ++end;
|
---|
| 294 | --r;
|
---|
| 295 | }
|
---|
| 296 | yield return values.Skip(start).Take(end - start);
|
---|
| 297 | start = end;
|
---|
| 298 | end += f;
|
---|
| 299 | }
|
---|
| 300 | }
|
---|
| 301 | }
|
---|
| 302 |
|
---|
| 303 | private static Action<RFParameter, double> GenerateSetter(string field) {
|
---|
| 304 | var targetExp = Expression.Parameter(typeof(RFParameter));
|
---|
| 305 | var valueExp = Expression.Parameter(typeof(double));
|
---|
| 306 | var fieldExp = Expression.Property(targetExp, field);
|
---|
| 307 | var assignExp = Expression.Assign(fieldExp, Expression.Convert(valueExp, fieldExp.Type));
|
---|
| 308 | var setter = Expression.Lambda<Action<RFParameter, double>>(assignExp, targetExp, valueExp).Compile();
|
---|
| 309 | return setter;
|
---|
| 310 | }
|
---|
| 311 |
|
---|
| 312 | private static string GetTargetVariableName(IDataAnalysisProblemData problemData) {
|
---|
| 313 | var regressionProblemData = problemData as IRegressionProblemData;
|
---|
| 314 | var classificationProblemData = problemData as IClassificationProblemData;
|
---|
| 315 |
|
---|
| 316 | if (regressionProblemData != null)
|
---|
| 317 | return regressionProblemData.TargetVariable;
|
---|
| 318 | if (classificationProblemData != null)
|
---|
| 319 | return classificationProblemData.TargetVariable;
|
---|
| 320 |
|
---|
| 321 | throw new ArgumentException("Problem data is neither regression or classification problem data.");
|
---|
| 322 | }
|
---|
[11315] | 323 | }
|
---|
| 324 | }
|
---|