TrainingError.h
Go to the documentation of this file.
1 /*!
2  *
3  *
4  * \brief Stopping Criterion which stops, when the trainign error seems to converge
5  *
6  *
7  *
8  * \author O. Krause
9  * \date 2010
10  *
11  *
12  * \par Copyright 1995-2017 Shark Development Team
13  *
14  * <BR><HR>
15  * This file is part of Shark.
16  * <http://shark-ml.org/>
17  *
18  * Shark is free software: you can redistribute it and/or modify
19  * it under the terms of the GNU Lesser General Public License as published
20  * by the Free Software Foundation, either version 3 of the License, or
21  * (at your option) any later version.
22  *
23  * Shark is distributed in the hope that it will be useful,
24  * but WITHOUT ANY WARRANTY; without even the implied warranty of
25  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26  * GNU Lesser General Public License for more details.
27  *
28  * You should have received a copy of the GNU Lesser General Public License
29  * along with Shark. If not, see <http://www.gnu.org/licenses/>.
30  *
31  */
32 
33 #ifndef SHARK_TRAINERS_STOPPINGCRITERA_TRAININGERROR_H
34 #define SHARK_TRAINERS_STOPPINGCRITERA_TRAININGERROR_H
35 
37 #include <shark/Core/ResultSets.h>
38 #include <queue>
39 #include <numeric>
40 #include <shark/LinAlg/Base.h>
41 namespace shark{
42 
43 /// \brief This stopping criterion tracks the improvement of the error function of the training error over an interval of iterations.
44 ///
45 /// If at one point, the difference between the error values of the beginning and the end of the interval are smaller
46 /// than a certain value, this stopping criterion assumes convergence and stops.
47 /// Of course, this may be misleading, when the algorithm temporarily gets stuck at a saddle point of the error surface.
48 /// The functions assumes that the algorithm is minimizing. For details, see:
49 ///
50 /// Lutz Prechelt. Early Stopping - but when? In Genevieve B. Orr and
51 /// Klaus-Robert Müller: Neural Networks: Tricks of the Trade, volume
52 /// 1524 of LNCS, Springer, 1997.
53 ///
54 template<class PointType = RealVector>
55 class TrainingError: public AbstractStoppingCriterion< SingleObjectiveResultSet<PointType> >{
56 public:
57  /// constructs the TrainingError generalization loss
58  /// @param intervalSize size of the interval over which the progress is monitored
59  /// @param minDifference minimum difference between start and end of the interval allowed before training stops
60  TrainingError(size_t intervalSize, double minDifference){
61  m_minDifference = minDifference;
62  m_intervalSize = intervalSize;
63  reset();
64  }
65  /// returns true if training should stop
67 
68  m_interval.pop();
69  m_interval.push(set.value);
70  return (m_interval.front()-set.value) >= 0
71  && (m_interval.front()-set.value) < m_minDifference;
72  }
73  /// resets the internal state
74  void reset(){
75  m_interval = std::queue<double>();
76  for(size_t i = 0; i != m_intervalSize;++i) {
77  m_interval.push(std::numeric_limits<double>::max());
78  }
79  }
80 protected:
81  /// monitored training interval
82  std::queue<double> m_interval;
83  /// minmum difference allowed
85  /// size of the interval
87 };
88 }
89 
90 
91 #endif