[12332] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
[17180] | 3 | * Copyright (C) Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[12332] | 4 | * and the BEACON Center for the Study of Evolution in Action.
|
---|
| 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 | #endregion
|
---|
| 22 |
|
---|
[17030] | 23 | using System;
|
---|
[12332] | 24 | using System.Linq;
|
---|
| 25 | using System.Threading;
|
---|
[17030] | 26 | using HeuristicLab.Algorithms.DataAnalysis.GradientBoostedTrees;
|
---|
[12332] | 27 | using HeuristicLab.Analysis;
|
---|
| 28 | using HeuristicLab.Common;
|
---|
| 29 | using HeuristicLab.Core;
|
---|
| 30 | using HeuristicLab.Data;
|
---|
| 31 | using HeuristicLab.Optimization;
|
---|
| 32 | using HeuristicLab.Parameters;
|
---|
[16565] | 33 | using HEAL.Attic;
|
---|
[12332] | 34 | using HeuristicLab.PluginInfrastructure;
|
---|
| 35 | using HeuristicLab.Problems.DataAnalysis;
|
---|
| 36 |
|
---|
| 37 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
[13646] | 38 | [Item("Gradient Boosted Trees (GBT)", "Gradient boosted trees algorithm. Specific implementation of gradient boosting for regression trees. Friedman, J. \"Greedy Function Approximation: A Gradient Boosting Machine\", IMS 1999 Reitz Lecture.")]
|
---|
[16565] | 39 | [StorableType("8CCB55BD-4935-4868-855F-D3E5D55127AA")]
|
---|
[12590] | 40 | [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 125)]
|
---|
[14523] | 41 | public class GradientBoostedTreesAlgorithm : FixedDataAnalysisAlgorithm<IRegressionProblem> {
|
---|
[12332] | 42 | #region ParameterNames
|
---|
| 43 | private const string IterationsParameterName = "Iterations";
|
---|
[12632] | 44 | private const string MaxSizeParameterName = "Maximum Tree Size";
|
---|
[12332] | 45 | private const string NuParameterName = "Nu";
|
---|
| 46 | private const string RParameterName = "R";
|
---|
| 47 | private const string MParameterName = "M";
|
---|
| 48 | private const string SeedParameterName = "Seed";
|
---|
| 49 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
---|
| 50 | private const string LossFunctionParameterName = "LossFunction";
|
---|
| 51 | private const string UpdateIntervalParameterName = "UpdateInterval";
|
---|
[17030] | 52 | private const string ModelCreationParameterName = "ModelCreation";
|
---|
[12332] | 53 | #endregion
|
---|
| 54 |
|
---|
| 55 | #region ParameterProperties
|
---|
| 56 | public IFixedValueParameter<IntValue> IterationsParameter {
|
---|
| 57 | get { return (IFixedValueParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
| 58 | }
|
---|
[12632] | 59 | public IFixedValueParameter<IntValue> MaxSizeParameter {
|
---|
| 60 | get { return (IFixedValueParameter<IntValue>)Parameters[MaxSizeParameterName]; }
|
---|
[12332] | 61 | }
|
---|
| 62 | public IFixedValueParameter<DoubleValue> NuParameter {
|
---|
| 63 | get { return (IFixedValueParameter<DoubleValue>)Parameters[NuParameterName]; }
|
---|
| 64 | }
|
---|
| 65 | public IFixedValueParameter<DoubleValue> RParameter {
|
---|
| 66 | get { return (IFixedValueParameter<DoubleValue>)Parameters[RParameterName]; }
|
---|
| 67 | }
|
---|
| 68 | public IFixedValueParameter<DoubleValue> MParameter {
|
---|
| 69 | get { return (IFixedValueParameter<DoubleValue>)Parameters[MParameterName]; }
|
---|
| 70 | }
|
---|
| 71 | public IFixedValueParameter<IntValue> SeedParameter {
|
---|
| 72 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
---|
| 73 | }
|
---|
| 74 | public FixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
---|
| 75 | get { return (FixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
---|
| 76 | }
|
---|
[12873] | 77 | public IConstrainedValueParameter<ILossFunction> LossFunctionParameter {
|
---|
| 78 | get { return (IConstrainedValueParameter<ILossFunction>)Parameters[LossFunctionParameterName]; }
|
---|
[12332] | 79 | }
|
---|
| 80 | public IFixedValueParameter<IntValue> UpdateIntervalParameter {
|
---|
| 81 | get { return (IFixedValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
|
---|
| 82 | }
|
---|
[17030] | 83 | private IFixedValueParameter<EnumValue<ModelCreation>> ModelCreationParameter {
|
---|
| 84 | get { return (IFixedValueParameter<EnumValue<ModelCreation>>)Parameters[ModelCreationParameterName]; }
|
---|
[12373] | 85 | }
|
---|
[12332] | 86 | #endregion
|
---|
| 87 |
|
---|
| 88 | #region Properties
|
---|
| 89 | public int Iterations {
|
---|
| 90 | get { return IterationsParameter.Value.Value; }
|
---|
| 91 | set { IterationsParameter.Value.Value = value; }
|
---|
| 92 | }
|
---|
| 93 | public int Seed {
|
---|
| 94 | get { return SeedParameter.Value.Value; }
|
---|
| 95 | set { SeedParameter.Value.Value = value; }
|
---|
| 96 | }
|
---|
| 97 | public bool SetSeedRandomly {
|
---|
| 98 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
| 99 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
| 100 | }
|
---|
[12632] | 101 | public int MaxSize {
|
---|
| 102 | get { return MaxSizeParameter.Value.Value; }
|
---|
| 103 | set { MaxSizeParameter.Value.Value = value; }
|
---|
[12332] | 104 | }
|
---|
| 105 | public double Nu {
|
---|
| 106 | get { return NuParameter.Value.Value; }
|
---|
| 107 | set { NuParameter.Value.Value = value; }
|
---|
| 108 | }
|
---|
| 109 | public double R {
|
---|
| 110 | get { return RParameter.Value.Value; }
|
---|
| 111 | set { RParameter.Value.Value = value; }
|
---|
| 112 | }
|
---|
| 113 | public double M {
|
---|
| 114 | get { return MParameter.Value.Value; }
|
---|
| 115 | set { MParameter.Value.Value = value; }
|
---|
| 116 | }
|
---|
[17030] | 117 | public ModelCreation ModelCreation {
|
---|
| 118 | get { return ModelCreationParameter.Value.Value; }
|
---|
| 119 | set { ModelCreationParameter.Value.Value = value; }
|
---|
[12373] | 120 | }
|
---|
[12332] | 121 | #endregion
|
---|
| 122 |
|
---|
| 123 | #region ResultsProperties
|
---|
| 124 | private double ResultsBestQuality {
|
---|
| 125 | get { return ((DoubleValue)Results["Best Quality"].Value).Value; }
|
---|
| 126 | set { ((DoubleValue)Results["Best Quality"].Value).Value = value; }
|
---|
| 127 | }
|
---|
| 128 | private DataTable ResultsQualities {
|
---|
| 129 | get { return ((DataTable)Results["Qualities"].Value); }
|
---|
| 130 | }
|
---|
| 131 | #endregion
|
---|
| 132 |
|
---|
| 133 | [StorableConstructor]
|
---|
[16565] | 134 | protected GradientBoostedTreesAlgorithm(StorableConstructorFlag _) : base(_) { }
|
---|
[12332] | 135 |
|
---|
| 136 | protected GradientBoostedTreesAlgorithm(GradientBoostedTreesAlgorithm original, Cloner cloner)
|
---|
| 137 | : base(original, cloner) {
|
---|
| 138 | }
|
---|
| 139 |
|
---|
| 140 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
| 141 | return new GradientBoostedTreesAlgorithm(this, cloner);
|
---|
| 142 | }
|
---|
| 143 |
|
---|
| 144 | public GradientBoostedTreesAlgorithm() {
|
---|
| 145 | Problem = new RegressionProblem(); // default problem
|
---|
| 146 |
|
---|
| 147 | Parameters.Add(new FixedValueParameter<IntValue>(IterationsParameterName, "Number of iterations (set as high as possible, adjust in combination with nu, when increasing iterations also decrease nu)", new IntValue(1000)));
|
---|
| 148 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName, "The random seed used to initialize the new pseudo random number generator.", new IntValue(0)));
|
---|
| 149 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName, "True if the random seed should be set to a random value, otherwise false.", new BoolValue(true)));
|
---|
[17030] | 150 | Parameters.Add(new FixedValueParameter<IntValue>(MaxSizeParameterName, "Maximal size of the tree learned in each step (prefer smaller sizes (3 to 10) if possible)", new IntValue(10)));
|
---|
[12332] | 151 | Parameters.Add(new FixedValueParameter<DoubleValue>(RParameterName, "Ratio of training rows selected randomly in each step (0 < R <= 1)", new DoubleValue(0.5)));
|
---|
| 152 | Parameters.Add(new FixedValueParameter<DoubleValue>(MParameterName, "Ratio of variables selected randomly in each step (0 < M <= 1)", new DoubleValue(0.5)));
|
---|
| 153 | Parameters.Add(new FixedValueParameter<DoubleValue>(NuParameterName, "Learning rate nu (step size for the gradient update, should be small 0 < nu < 0.1)", new DoubleValue(0.002)));
|
---|
[12373] | 154 | Parameters.Add(new FixedValueParameter<IntValue>(UpdateIntervalParameterName, "", new IntValue(100)));
|
---|
| 155 | Parameters[UpdateIntervalParameterName].Hidden = true;
|
---|
[17030] | 156 | Parameters.Add(new FixedValueParameter<EnumValue<ModelCreation>>(ModelCreationParameterName, "Defines the results produced at the end of the run (Surrogate => Less disk space, lazy recalculation of model)", new EnumValue<ModelCreation>(ModelCreation.Model)));
|
---|
| 157 | Parameters[ModelCreationParameterName].Hidden = true;
|
---|
[12332] | 158 |
|
---|
[12873] | 159 | var lossFunctions = ApplicationManager.Manager.GetInstances<ILossFunction>();
|
---|
| 160 | Parameters.Add(new ConstrainedValueParameter<ILossFunction>(LossFunctionParameterName, "The loss function", new ItemSet<ILossFunction>(lossFunctions)));
|
---|
| 161 | LossFunctionParameter.Value = LossFunctionParameter.ValidValues.First(f => f.ToString().Contains("Squared")); // squared error loss is the default
|
---|
[12332] | 162 | }
|
---|
| 163 |
|
---|
[12873] | 164 | [StorableHook(HookType.AfterDeserialization)]
|
---|
| 165 | private void AfterDeserialization() {
|
---|
| 166 | // BackwardsCompatibility3.4
|
---|
| 167 | #region Backwards compatible code, remove with 3.5
|
---|
[17030] | 168 |
|
---|
| 169 | #region LossFunction
|
---|
[12873] | 170 | // parameter type has been changed
|
---|
| 171 | var lossFunctionParam = Parameters[LossFunctionParameterName] as ConstrainedValueParameter<StringValue>;
|
---|
| 172 | if (lossFunctionParam != null) {
|
---|
| 173 | Parameters.Remove(LossFunctionParameterName);
|
---|
| 174 | var selectedValue = lossFunctionParam.Value; // to be restored below
|
---|
[12332] | 175 |
|
---|
[12873] | 176 | var lossFunctions = ApplicationManager.Manager.GetInstances<ILossFunction>();
|
---|
| 177 | Parameters.Add(new ConstrainedValueParameter<ILossFunction>(LossFunctionParameterName, "The loss function", new ItemSet<ILossFunction>(lossFunctions)));
|
---|
| 178 | // try to restore selected value
|
---|
| 179 | var selectedLossFunction =
|
---|
| 180 | LossFunctionParameter.ValidValues.FirstOrDefault(f => f.ToString() == selectedValue.Value);
|
---|
| 181 | if (selectedLossFunction != null) {
|
---|
| 182 | LossFunctionParameter.Value = selectedLossFunction;
|
---|
| 183 | } else {
|
---|
| 184 | LossFunctionParameter.Value = LossFunctionParameter.ValidValues.First(f => f.ToString().Contains("Squared")); // default: SE
|
---|
| 185 | }
|
---|
| 186 | }
|
---|
| 187 | #endregion
|
---|
[17030] | 188 |
|
---|
| 189 | #region CreateSolution
|
---|
| 190 | // parameter type has been changed
|
---|
| 191 | if (Parameters.ContainsKey("CreateSolution")) {
|
---|
| 192 | var createSolutionParam = Parameters["CreateSolution"] as FixedValueParameter<BoolValue>;
|
---|
| 193 | Parameters.Remove(createSolutionParam);
|
---|
| 194 |
|
---|
| 195 | ModelCreation value = createSolutionParam.Value.Value ? ModelCreation.Model : ModelCreation.QualityOnly;
|
---|
| 196 | Parameters.Add(new FixedValueParameter<EnumValue<ModelCreation>>(ModelCreationParameterName, "Defines the results produced at the end of the run (Surrogate => Less disk space, lazy recalculation of model)", new EnumValue<ModelCreation>(value)));
|
---|
| 197 | Parameters[ModelCreationParameterName].Hidden = true;
|
---|
[17154] | 198 | } else if (!Parameters.ContainsKey(ModelCreationParameterName)) {
|
---|
| 199 | // very old version contains neither ModelCreationParameter nor CreateSolutionParameter
|
---|
| 200 | Parameters.Add(new FixedValueParameter<EnumValue<ModelCreation>>(ModelCreationParameterName, "Defines the results produced at the end of the run (Surrogate => Less disk space, lazy recalculation of model)", new EnumValue<ModelCreation>(ModelCreation.Model)));
|
---|
| 201 | Parameters[ModelCreationParameterName].Hidden = true;
|
---|
[17030] | 202 | }
|
---|
| 203 | #endregion
|
---|
| 204 | #endregion
|
---|
[12873] | 205 | }
|
---|
| 206 |
|
---|
[12332] | 207 | protected override void Run(CancellationToken cancellationToken) {
|
---|
| 208 | // Set up the algorithm
|
---|
[16071] | 209 | if (SetSeedRandomly) Seed = Random.RandomSeedGenerator.GetSeed();
|
---|
[12332] | 210 |
|
---|
| 211 | // Set up the results display
|
---|
| 212 | var iterations = new IntValue(0);
|
---|
| 213 | Results.Add(new Result("Iterations", iterations));
|
---|
| 214 |
|
---|
| 215 | var table = new DataTable("Qualities");
|
---|
| 216 | table.Rows.Add(new DataRow("Loss (train)"));
|
---|
| 217 | table.Rows.Add(new DataRow("Loss (test)"));
|
---|
[14841] | 218 | table.Rows["Loss (train)"].VisualProperties.StartIndexZero = true;
|
---|
| 219 | table.Rows["Loss (test)"].VisualProperties.StartIndexZero = true;
|
---|
| 220 |
|
---|
[12332] | 221 | Results.Add(new Result("Qualities", table));
|
---|
| 222 | var curLoss = new DoubleValue();
|
---|
[12373] | 223 | Results.Add(new Result("Loss (train)", curLoss));
|
---|
[12332] | 224 |
|
---|
| 225 | // init
|
---|
[12620] | 226 | var problemData = (IRegressionProblemData)Problem.ProblemData.Clone();
|
---|
[12873] | 227 | var lossFunction = LossFunctionParameter.Value;
|
---|
[12632] | 228 | var state = GradientBoostedTreesAlgorithmStatic.CreateGbmState(problemData, lossFunction, (uint)Seed, MaxSize, R, M, Nu);
|
---|
[12332] | 229 |
|
---|
| 230 | var updateInterval = UpdateIntervalParameter.Value.Value;
|
---|
| 231 | // Loop until iteration limit reached or canceled.
|
---|
| 232 | for (int i = 0; i < Iterations; i++) {
|
---|
| 233 | cancellationToken.ThrowIfCancellationRequested();
|
---|
| 234 |
|
---|
| 235 | GradientBoostedTreesAlgorithmStatic.MakeStep(state);
|
---|
| 236 |
|
---|
| 237 | // iteration results
|
---|
| 238 | if (i % updateInterval == 0) {
|
---|
| 239 | curLoss.Value = state.GetTrainLoss();
|
---|
| 240 | table.Rows["Loss (train)"].Values.Add(curLoss.Value);
|
---|
| 241 | table.Rows["Loss (test)"].Values.Add(state.GetTestLoss());
|
---|
| 242 | iterations.Value = i;
|
---|
| 243 | }
|
---|
| 244 | }
|
---|
| 245 |
|
---|
| 246 | // final results
|
---|
| 247 | iterations.Value = Iterations;
|
---|
| 248 | curLoss.Value = state.GetTrainLoss();
|
---|
| 249 | table.Rows["Loss (train)"].Values.Add(curLoss.Value);
|
---|
| 250 | table.Rows["Loss (test)"].Values.Add(state.GetTestLoss());
|
---|
| 251 |
|
---|
| 252 | // produce variable relevance
|
---|
| 253 | var orderedImpacts = state.GetVariableRelevance().Select(t => new { name = t.Key, impact = t.Value }).ToList();
|
---|
| 254 |
|
---|
| 255 | var impacts = new DoubleMatrix();
|
---|
| 256 | var matrix = impacts as IStringConvertibleMatrix;
|
---|
| 257 | matrix.Rows = orderedImpacts.Count;
|
---|
| 258 | matrix.RowNames = orderedImpacts.Select(x => x.name);
|
---|
| 259 | matrix.Columns = 1;
|
---|
| 260 | matrix.ColumnNames = new string[] { "Relative variable relevance" };
|
---|
| 261 |
|
---|
| 262 | int rowIdx = 0;
|
---|
| 263 | foreach (var p in orderedImpacts) {
|
---|
| 264 | matrix.SetValue(string.Format("{0:N2}", p.impact), rowIdx++, 0);
|
---|
| 265 | }
|
---|
| 266 |
|
---|
| 267 | Results.Add(new Result("Variable relevance", impacts));
|
---|
[12373] | 268 | Results.Add(new Result("Loss (test)", new DoubleValue(state.GetTestLoss())));
|
---|
[12332] | 269 |
|
---|
| 270 | // produce solution
|
---|
[17030] | 271 | if (ModelCreation == ModelCreation.SurrogateModel || ModelCreation == ModelCreation.Model) {
|
---|
| 272 | IRegressionModel model = state.GetModel();
|
---|
[12868] | 273 |
|
---|
[17030] | 274 | if (ModelCreation == ModelCreation.SurrogateModel) {
|
---|
[17044] | 275 | model = new GradientBoostedTreesModelSurrogate((GradientBoostedTreesModel)model, problemData, (uint)Seed, lossFunction, Iterations, MaxSize, R, M, Nu);
|
---|
[17030] | 276 | }
|
---|
| 277 |
|
---|
[12611] | 278 | // for logistic regression we produce a classification solution
|
---|
| 279 | if (lossFunction is LogisticRegressionLoss) {
|
---|
[13065] | 280 | var classificationModel = new DiscriminantFunctionClassificationModel(model,
|
---|
[12611] | 281 | new AccuracyMaximizationThresholdCalculator());
|
---|
| 282 | var classificationProblemData = new ClassificationProblemData(problemData.Dataset,
|
---|
[17845] | 283 | problemData.AllowedInputVariables, problemData.TargetVariable, transformations: problemData.Transformations);
|
---|
[14780] | 284 | classificationProblemData.TrainingPartition.Start = Problem.ProblemData.TrainingPartition.Start;
|
---|
| 285 | classificationProblemData.TrainingPartition.End = Problem.ProblemData.TrainingPartition.End;
|
---|
| 286 | classificationProblemData.TestPartition.Start = Problem.ProblemData.TestPartition.Start;
|
---|
| 287 | classificationProblemData.TestPartition.End = Problem.ProblemData.TestPartition.End;
|
---|
| 288 |
|
---|
[14841] | 289 | classificationModel.SetThresholdsAndClassValues(new double[] { double.NegativeInfinity, 0.0 }, new[] { 0.0, 1.0 });
|
---|
[12611] | 290 |
|
---|
[14780] | 291 |
|
---|
[13065] | 292 | var classificationSolution = new DiscriminantFunctionClassificationSolution(classificationModel, classificationProblemData);
|
---|
[12619] | 293 | Results.Add(new Result("Solution", classificationSolution));
|
---|
[12611] | 294 | } else {
|
---|
| 295 | // otherwise we produce a regression solution
|
---|
[14345] | 296 | Results.Add(new Result("Solution", new GradientBoostedTreesSolution(model, problemData)));
|
---|
[12611] | 297 | }
|
---|
[17030] | 298 | } else if (ModelCreation == ModelCreation.QualityOnly) {
|
---|
| 299 | //Do nothing
|
---|
| 300 | } else {
|
---|
| 301 | throw new NotImplementedException("Selected parameter for CreateSolution isn't implemented yet");
|
---|
[12611] | 302 | }
|
---|
[12332] | 303 | }
|
---|
| 304 | }
|
---|
| 305 | }
|
---|