Free cookie consent management tool by TermsFeed Policy Generator

source: branches/GBT-trunkintegration/HeuristicLab.Algorithms.DataAnalysis/3.4/GradientBoostedTrees/GradientBoostedTreesAlgorithmStatic.cs @ 12632

Last change on this file since 12632 was 12632, checked in by gkronber, 9 years ago

#2261 implemented node expansion using a priority queue (and changed parameter MaxDepth to MaxSize). Moved unit tests to a separate project.

File size: 7.7 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2015 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
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
23using System;
24using System.Collections.Generic;
25using System.Diagnostics.Contracts;
26using System.Linq;
27using HeuristicLab.Problems.DataAnalysis;
28using HeuristicLab.Random;
29
30namespace HeuristicLab.Algorithms.DataAnalysis {
31  public static class GradientBoostedTreesAlgorithmStatic {
32    #region static API
33
34    public interface IGbmState {
35      IRegressionModel GetModel();
36      double GetTrainLoss();
37      double GetTestLoss();
38      IEnumerable<KeyValuePair<string, double>> GetVariableRelevance();
39    }
40
41    // created through factory method
42    // GbmState details are private API users can only use methods from IGbmState
43    private class GbmState : IGbmState {
44      internal IRegressionProblemData problemData { get; set; }
45      internal MersenneTwister random { get; private set; }
46      internal ILossFunction lossFunction { get; set; }
47      internal int maxSize { get; set; }
48      internal double nu { get; set; }
49      internal double r { get; set; }
50      internal double m { get; set; }
51      internal readonly RegressionTreeBuilder treeBuilder;
52
53
54      // array members (allocate only once)
55      internal double[] pred;
56      internal double[] predTest;
57      internal double[] w;
58      internal double[] y;
59      internal int[] activeIdx;
60      internal double[] pseudoRes;
61
62      internal IList<IRegressionModel> models;
63      internal IList<double> weights;
64
65      public GbmState(IRegressionProblemData problemData, ILossFunction lossFunction, uint randSeed, int maxSize, double r, double m, double nu) {
66        // default settings for MaxSize, Nu and R
67        this.maxSize = maxSize;
68        this.nu = nu;
69        this.r = r;
70        this.m = m;
71
72        random = new MersenneTwister(randSeed);
73        this.problemData = problemData;
74        this.lossFunction = lossFunction;
75
76        int nRows = problemData.TrainingIndices.Count();
77
78        y = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices).ToArray();
79        // weights are all 1 for now (HL doesn't support weights yet)
80        w = Enumerable.Repeat(1.0, nRows).ToArray();
81
82        treeBuilder = new RegressionTreeBuilder(problemData, random);
83
84        activeIdx = Enumerable.Range(0, nRows).ToArray();
85
86        var zeros = Enumerable.Repeat(0.0, nRows);
87        var ones = Enumerable.Repeat(1.0, nRows);
88        double f0 = lossFunction.GetLineSearchFunc(y, zeros, ones)(activeIdx, 0, nRows - 1); // initial constant value (mean for squared errors)
89        pred = Enumerable.Repeat(f0, nRows).ToArray();
90        predTest = Enumerable.Repeat(f0, problemData.TestIndices.Count()).ToArray();
91        pseudoRes = new double[nRows];
92
93        models = new List<IRegressionModel>();
94        weights = new List<double>();
95        // add constant model
96        models.Add(new ConstantRegressionModel(f0));
97        weights.Add(1.0);
98      }
99
100      public IRegressionModel GetModel() {
101        return new GradientBoostedTreesModel(models, weights);
102      }
103      public IEnumerable<KeyValuePair<string, double>> GetVariableRelevance() {
104        return treeBuilder.GetVariableRelevance();
105      }
106
107      public double GetTrainLoss() {
108        int nRows = y.Length;
109        return lossFunction.GetLoss(y, pred, w) / nRows;
110      }
111      public double GetTestLoss() {
112        var yTest = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TestIndices);
113        var wTest = problemData.TestIndices.Select(_ => 1.0); // ones
114        var nRows = problemData.TestIndices.Count();
115        return lossFunction.GetLoss(yTest, predTest, wTest) / nRows;
116      }
117    }
118
119    // simple interface
120    public static IRegressionSolution TrainGbm(IRegressionProblemData problemData, int maxSize, double nu, double r, int maxIterations) {
121      return TrainGbm(problemData, new SquaredErrorLoss(), maxSize, nu, r, maxIterations);
122    }
123
124    // simple interface
125    public static IRegressionSolution TrainGbm(IRegressionProblemData problemData, ILossFunction lossFunction,
126      int maxSize, double nu, double r, int maxIterations, uint randSeed = 31415) {
127      Contract.Assert(r > 0);
128      Contract.Assert(r <= 1.0);
129      Contract.Assert(nu > 0);
130      Contract.Assert(nu <= 1.0);
131
132      var state = (GbmState)CreateGbmState(problemData, lossFunction, randSeed);
133      state.maxSize = maxSize;
134      state.r = r;
135      state.nu = nu;
136
137      for (int iter = 0; iter < maxIterations; iter++) {
138        MakeStep(state);
139      }
140
141      var model = new GradientBoostedTreesModel(state.models, state.weights);
142      return new RegressionSolution(model, (IRegressionProblemData)problemData.Clone());
143    }
144
145    // for custom stepping & termination
146    public static IGbmState CreateGbmState(IRegressionProblemData problemData, ILossFunction lossFunction, uint randSeed, int maxSize = 3, double r = 0.66, double m = 0.5, double nu = 0.01) {
147      return new GbmState(problemData, lossFunction, randSeed, maxSize, r, m, nu);
148    }
149
150    // use default settings for maxDepth, nu, r from state
151    public static void MakeStep(IGbmState state) {
152      var gbmState = state as GbmState;
153      if (gbmState == null) throw new ArgumentException("state");
154
155      MakeStep(gbmState, gbmState.maxSize, gbmState.nu, gbmState.r, gbmState.m);
156    }
157
158    // allow dynamic adaptation of maxDepth, nu and r (even though this is not used)
159    public static void MakeStep(IGbmState state, int maxSize, double nu, double r, double m) {
160      var gbmState = state as GbmState;
161      if (gbmState == null) throw new ArgumentException("state");
162
163      var problemData = gbmState.problemData;
164      var lossFunction = gbmState.lossFunction;
165      var yPred = gbmState.pred;
166      var yPredTest = gbmState.predTest;
167      var w = gbmState.w;
168      var treeBuilder = gbmState.treeBuilder;
169      var y = gbmState.y;
170      var activeIdx = gbmState.activeIdx;
171      var pseudoRes = gbmState.pseudoRes;
172
173      // copy output of gradient function to pre-allocated rim array (pseudo-residuals)
174      int rimIdx = 0;
175      foreach (var g in lossFunction.GetLossGradient(y, yPred, w)) {
176        pseudoRes[rimIdx++] = g;
177      }
178
179      var tree = treeBuilder.CreateRegressionTreeForGradientBoosting(pseudoRes, maxSize, activeIdx, lossFunction.GetLineSearchFunc(y, yPred, w), r, m);
180
181      int i = 0;
182      foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, problemData.TrainingIndices)) {
183        yPred[i] = yPred[i] + nu * pred;
184        i++;
185      }
186      // update predictions for validation set
187      i = 0;
188      foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, problemData.TestIndices)) {
189        yPredTest[i] = yPredTest[i] + nu * pred;
190        i++;
191      }
192
193      gbmState.weights.Add(nu);
194      gbmState.models.Add(tree);
195    }
196    #endregion
197  }
198}
Note: See TracBrowser for help on using the repository browser.