LabelOrder.h
Go to the documentation of this file.
1 //===========================================================================
2 /*!
3  *
4  *
5  * \brief This will relabel a given dataset to have labels 0..N-1 (and vice versa)
6  *
7  *
8  *
9  * \author Aydin Demircioglu
10  * \date 2014
11  *
12  *
13  * \par Copyright 1995-2017 Shark Development Team
14  *
15  * <BR><HR>
16  * This file is part of Shark.
17  * <http://shark-ml.org/>
18  *
19  * Shark is free software: you can redistribute it and/or modify
20  * it under the terms of the GNU Lesser General Public License as published
21  * by the Free Software Foundation, either version 3 of the License, or
22  * (at your option) any later version.
23  *
24  * Shark is distributed in the hope that it will be useful,
25  * but WITHOUT ANY WARRANTY; without even the implied warranty of
26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27  * GNU Lesser General Public License for more details.
28  *
29  * You should have received a copy of the GNU Lesser General Public License
30  * along with Shark. If not, see <http://www.gnu.org/licenses/>.
31  *
32  */
33 //===========================================================================
34 
35 
36 #ifndef SHARK_LABELORDER_H
37 #define SHARK_LABELORDER_H
38 
39 #include <shark/Core/INameable.h>
41 
42 #include <shark/Data/Dataset.h>
43 
44 
45 
46 
47 namespace shark
48 {
49 
50 
51 /// \brief This will normalize the labels of a given dataset to 0..N-1
52 ///
53 /// \par This will normalize the labels of a given dataset to 0..N-1
54 /// and store the ordering in a member variable.
55 /// After processing, the dataset will afterwards have labels ranging
56 /// from 0 to N-1, with N the number of classes, so usual Shark
57 /// trainers can work with it.
58 /// One can then revert the original labeling just by calling restoreOriginalLabels
59 class LabelOrder : public INameable
60 {
61 private:
62 
63 public:
64 
65 
66  LabelOrder() {};
67 
68 
69  virtual ~LabelOrder() {};
70 
71 
72  /// \brief From INameable: return the class name.
73  std::string name() const
74  { return "LabelOrder"; }
75 
76 
77  /// \brief This will normalize the labels and store the ordering in the
78  /// member variables. The dataset will afterwards have labels ranging
79  /// from 0 to N-1, with N the number of classes.
80  /// This will overwrite any previously stored label ordering in the object.
81  ///
82  /// \param[in,out] dataset dataset that will be relabeled
83 
85  {
86  // determine the min and max labels of the given dataset
87  unsigned int minLabel = std::numeric_limits<unsigned int>::max();
88  unsigned int maxLabel = 0;
89  for(std::size_t i = 0; i < dataset.numberOfElements(); ++i)
90  {
91  unsigned int label = dataset.labels().element(i);
92 
93  if(label < minLabel)
94  minLabel = label;
95  if(label > maxLabel)
96  maxLabel = label;
97  }
98 
99  // now we create an vector that can hold the label ordering
100  m_labelOrder.clear();
101 
102  // and one array that tracks what we already encountered
103  unsigned int maxval = std::numeric_limits<unsigned int>::max();
104  std::vector<unsigned int> foundLabels(maxLabel - minLabel + 1, maxval);
105 
106  // and insert all labels we encounter
107  unsigned int currentPosition = 0;
108  for(std::size_t i = 0; i < dataset.numberOfElements(); i++)
109  {
110  // is it a new label?
111  unsigned int label = dataset.labels().element(i);
112  if(foundLabels[label - minLabel] == maxval)
113  {
114  foundLabels[label - minLabel] = currentPosition;
115  m_labelOrder.push_back(label);
116  currentPosition++;
117  }
118  }
119 
120  // now map every label
121  for(std::size_t i = 0; i < dataset.numberOfElements(); i++)
122  {
123  unsigned int label = dataset.labels().element(i);
124  dataset.labels().element(i) = foundLabels[label - minLabel];
125  }
126  }
127 
128 
129 
130  /// \brief This will restore the original labels of the dataset. This
131  /// must be called with data compatible the original dataset, so that the labels will
132  /// fit. The label ordering will not be destroyed after calling this function, so
133  /// it can be called multiple times, e.g. to testsets or similar data.
134  ///
135  /// \param[in,out] dataset dataset to relabel (restore labels)
136 
138  {
139  // now map every label
140  for(std::size_t i = 0; i < dataset.numberOfElements(); ++i)
141  {
142  unsigned int label = dataset.labels().element(i);
143 
144  // check if the reordering fit the data
145  SHARK_RUNTIME_CHECK(label < m_labelOrder.size(),"Dataset labels does not fit to the stored ordering!");
146 
147  // relabel
148  label = m_labelOrder[label];
149  dataset.labels().element(i) = label;
150  }
151  }
152 
153 
154 
155  /// \brief Get label ordering directly
156  ///
157  /// \param[out] labelOrder vector to store the current label order.
158 
159  void getLabelOrder(std::vector<unsigned int>& labelOrder)
160  {
161  labelOrder = m_labelOrder;
162  }
163 
164 
165  /// \brief Set label ordering directly
166  ///
167  /// \param[in] labelOrder vector with the new label order
168 
169  void setLabelOrder(std::vector<unsigned int> const& labelOrder)
170  {
171  m_labelOrder = labelOrder;
172  }
173 
174 
175 protected:
176 
177  std::vector<unsigned int> m_labelOrder;
178 };
179 
180 }
181 
182 #endif
183