1  #region License Information


2  /* HeuristicLab


3  * Copyright (C) 20022016 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 


23  using System;


24  using System.Collections.Generic;


25  using System.Diagnostics;


26  using System.Linq;


27  using HeuristicLab.Common;


28  using HeuristicLab.Core;


29  using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;


30 


31  namespace HeuristicLab.Algorithms.DataAnalysis {


32  // Greedy Function Approximation: A Gradient Boosting Machine (page 9)


33  [StorableClass]


34  [Item("Logistic regression loss", "")]


35  public sealed class LogisticRegressionLoss : Item, ILossFunction {


36  public LogisticRegressionLoss() { }


37 


38  public double GetLoss(IEnumerable<double> target, IEnumerable<double> pred) {


39  var targetEnum = target.GetEnumerator();


40  var predEnum = pred.GetEnumerator();


41 


42  double s = 0;


43  while (targetEnum.MoveNext() & predEnum.MoveNext()) {


44  Debug.Assert(targetEnum.Current.IsAlmost(0.0)  targetEnum.Current.IsAlmost(1.0), "labels must be 0 or 1 for logistic regression loss");


45 


46  var y = targetEnum.Current * 2  1; // y in {1,1}


47  s += Math.Log(1 + Math.Exp(2 * y * predEnum.Current));


48  }


49  if (targetEnum.MoveNext()  predEnum.MoveNext())


50  throw new ArgumentException("target and pred have different lengths");


51 


52  return s;


53  }


54 


55  public IEnumerable<double> GetLossGradient(IEnumerable<double> target, IEnumerable<double> pred) {


56  var targetEnum = target.GetEnumerator();


57  var predEnum = pred.GetEnumerator();


58 


59  while (targetEnum.MoveNext() & predEnum.MoveNext()) {


60  Debug.Assert(targetEnum.Current.IsAlmost(0.0)  targetEnum.Current.IsAlmost(1.0), "labels must be 0 or 1 for logistic regression loss");


61  var y = targetEnum.Current * 2  1; // y in {1,1}


62 


63  yield return 2 * y / (1 + Math.Exp(2 * y * predEnum.Current));


64 


65  }


66  if (targetEnum.MoveNext()  predEnum.MoveNext())


67  throw new ArgumentException("target and pred have different lengths");


68  }


69 


70  // targetArr and predArr are not changed by LineSearch


71  public double LineSearch(double[] targetArr, double[] predArr, int[] idx, int startIdx, int endIdx) {


72  if (targetArr.Length != predArr.Length)


73  throw new ArgumentException("target and pred have different lengths");


74 


75  // "Simple NewtonRaphson step" of eqn. 23


76  double sumY = 0.0;


77  double sumDiff = 0.0;


78  for (int i = startIdx; i <= endIdx; i++) {


79  var row = idx[i];


80  var y = targetArr[row] * 2  1; // y in {1,1}


81  var pseudoResponse = 2 * y / (1 + Math.Exp(2 * y * predArr[row]));


82 


83  sumY += pseudoResponse;


84  sumDiff += Math.Abs(pseudoResponse) * (2  Math.Abs(pseudoResponse));


85  }


86  // prevent divByZero


87  sumDiff = Math.Max(1E12, sumDiff);


88  return sumY / sumDiff;


89  }


90 


91  #region item implementation


92  [StorableConstructor]


93  private LogisticRegressionLoss(bool deserializing) : base(deserializing) { }


94 


95  private LogisticRegressionLoss(LogisticRegressionLoss original, Cloner cloner) : base(original, cloner) { }


96 


97  public override IDeepCloneable Clone(Cloner cloner) {


98  return new LogisticRegressionLoss(this, cloner);


99  }


100  #endregion


101 


102  }


103  }

