1 | #ifndef _theplu_yat_classifier_subset_generator_ |
---|
2 | #define _theplu_yat_classifier_subset_generator_ |
---|
3 | |
---|
4 | // $Id: SubsetGenerator.h 1086 2008-02-14 05:43:10Z peter $ |
---|
5 | |
---|
6 | /* |
---|
7 | Copyright (C) 2006 Jari Häkkinen, Markus Ringnér, Peter Johansson |
---|
8 | Copyright (C) 2007 Peter Johansson |
---|
9 | |
---|
10 | This file is part of the yat library, http://trac.thep.lu.se/yat |
---|
11 | |
---|
12 | The yat library is free software; you can redistribute it and/or |
---|
13 | modify it under the terms of the GNU General Public License as |
---|
14 | published by the Free Software Foundation; either version 2 of the |
---|
15 | License, or (at your option) any later version. |
---|
16 | |
---|
17 | The yat library is distributed in the hope that it will be useful, |
---|
18 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
19 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
---|
20 | General Public License for more details. |
---|
21 | |
---|
22 | You should have received a copy of the GNU General Public License |
---|
23 | along with this program; if not, write to the Free Software |
---|
24 | Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA |
---|
25 | 02111-1307, USA. |
---|
26 | */ |
---|
27 | |
---|
28 | #include "DataLookup2D.h" |
---|
29 | #include "FeatureSelector.h" |
---|
30 | #include "KernelLookup.h" |
---|
31 | #include "MatrixLookup.h" |
---|
32 | #include "MatrixLookupWeighted.h" |
---|
33 | #include "Target.h" |
---|
34 | #include "Sampler.h" |
---|
35 | #include "yat/utility/yat_assert.h" |
---|
36 | |
---|
37 | #include <algorithm> |
---|
38 | #include <cassert> |
---|
39 | #include <utility> |
---|
40 | #include <typeinfo> |
---|
41 | #include <vector> |
---|
42 | |
---|
43 | namespace theplu { |
---|
44 | namespace yat { |
---|
45 | namespace classifier { |
---|
46 | |
---|
47 | /// |
---|
48 | /// @brief Class splitting a set into training set and validation set. |
---|
49 | /// |
---|
50 | template <typename T> |
---|
51 | class SubsetGenerator |
---|
52 | { |
---|
53 | public: |
---|
54 | typedef T value_type; |
---|
55 | |
---|
56 | /// |
---|
57 | /// @brief Constructor |
---|
58 | /// |
---|
59 | /// @param sampler sampler |
---|
60 | /// @param data data to split up in validation and training. |
---|
61 | /// |
---|
62 | SubsetGenerator(const Sampler& sampler, const T& data); |
---|
63 | |
---|
64 | /// |
---|
65 | /// @brief Constructor |
---|
66 | /// |
---|
67 | /// @param sampler taking care of partioning dataset |
---|
68 | /// @param data data to be split up in validation and training. |
---|
69 | /// @param fs Object selecting features for each subset |
---|
70 | /// |
---|
71 | SubsetGenerator(const Sampler& sampler, const T& data, |
---|
72 | FeatureSelector& fs); |
---|
73 | |
---|
74 | /// |
---|
75 | /// Destructor |
---|
76 | /// |
---|
77 | ~SubsetGenerator(); |
---|
78 | |
---|
79 | /// |
---|
80 | /// @return number of subsets |
---|
81 | /// |
---|
82 | u_long size(void) const; |
---|
83 | |
---|
84 | /// |
---|
85 | /// @return the target for the total set |
---|
86 | /// |
---|
87 | const Target& target(void) const; |
---|
88 | |
---|
89 | /// |
---|
90 | /// @return the sampler for the total set |
---|
91 | /// |
---|
92 | // const Sampler& sampler(void) const; |
---|
93 | |
---|
94 | /// |
---|
95 | /// @return training data |
---|
96 | /// |
---|
97 | const T& training_data(size_t i) const; |
---|
98 | |
---|
99 | /// |
---|
100 | /// @return training features |
---|
101 | /// |
---|
102 | const std::vector<size_t>& |
---|
103 | training_features(std::vector<size_t>::size_type i) const; |
---|
104 | |
---|
105 | /// |
---|
106 | /// @return training index |
---|
107 | /// |
---|
108 | const std::vector<size_t>& |
---|
109 | training_index(std::vector<size_t>::size_type i) const; |
---|
110 | |
---|
111 | /// |
---|
112 | /// @return training target |
---|
113 | /// |
---|
114 | const Target& training_target(std::vector<Target>::size_type i) const; |
---|
115 | |
---|
116 | /// |
---|
117 | /// @return validation data |
---|
118 | /// |
---|
119 | const T& validation_data(size_t i) const; |
---|
120 | |
---|
121 | /// |
---|
122 | /// @return validation index |
---|
123 | /// |
---|
124 | const std::vector<size_t>& |
---|
125 | validation_index(std::vector<size_t>::size_type i) const; |
---|
126 | |
---|
127 | /// |
---|
128 | /// @return validation target |
---|
129 | /// |
---|
130 | const Target& validation_target(std::vector<Target>::size_type i) const; |
---|
131 | |
---|
132 | /// |
---|
133 | /// @return true if weighted |
---|
134 | /// @todo remove this function |
---|
135 | //bool weighted(void) const; |
---|
136 | |
---|
137 | private: |
---|
138 | void build(const MatrixLookup&); |
---|
139 | void build(const MatrixLookupWeighted&); |
---|
140 | void build(const KernelLookup&); |
---|
141 | |
---|
142 | SubsetGenerator(const SubsetGenerator&); |
---|
143 | const SubsetGenerator& operator=(const SubsetGenerator&) const; |
---|
144 | |
---|
145 | FeatureSelector* f_selector_; |
---|
146 | std::vector<std::vector<size_t> > features_; |
---|
147 | const Sampler& sampler_; |
---|
148 | std::vector<const T*> training_data_; |
---|
149 | std::vector<Target> training_target_; |
---|
150 | std::vector<const T*> validation_data_; |
---|
151 | std::vector<Target> validation_target_; |
---|
152 | |
---|
153 | }; |
---|
154 | |
---|
155 | |
---|
156 | // templates |
---|
157 | |
---|
158 | template<typename T> |
---|
159 | SubsetGenerator<T>::SubsetGenerator(const Sampler& sampler, |
---|
160 | const T& data) |
---|
161 | : f_selector_(NULL), sampler_(sampler) |
---|
162 | { |
---|
163 | utility::yat_assert<std::runtime_error>(target().size()==data.columns()); |
---|
164 | |
---|
165 | training_data_.reserve(sampler_.size()); |
---|
166 | validation_data_.reserve(sampler_.size()); |
---|
167 | for (size_t i=0; i<sampler_.size(); ++i){ |
---|
168 | // Dynamically allocated. Must be deleted in destructor. |
---|
169 | training_data_.push_back(data.training_data(sampler.training_index(i))); |
---|
170 | validation_data_.push_back(data.validation_data(sampler.training_index(i), |
---|
171 | sampler.validation_index(i))); |
---|
172 | |
---|
173 | training_target_.push_back(Target(target(),sampler.training_index(i))); |
---|
174 | validation_target_.push_back(Target(target(), |
---|
175 | sampler.validation_index(i))); |
---|
176 | utility::yat_assert<std::runtime_error>(training_data_.size()==i+1); |
---|
177 | utility::yat_assert<std::runtime_error>(training_target_.size()==i+1); |
---|
178 | utility::yat_assert<std::runtime_error>(validation_data_.size()==i+1); |
---|
179 | utility::yat_assert<std::runtime_error>(validation_target_.size()==i+1); |
---|
180 | } |
---|
181 | |
---|
182 | // No feature selection, hence features same for all partitions |
---|
183 | // and can be stored in features_[0] |
---|
184 | features_.resize(1); |
---|
185 | features_[0].reserve(data.rows()); |
---|
186 | for (size_t i=0; i<data.rows(); ++i) |
---|
187 | features_[0].push_back(i); |
---|
188 | |
---|
189 | utility::yat_assert<std::runtime_error>(training_data_.size()==size()); |
---|
190 | utility::yat_assert<std::runtime_error>(training_target_.size()==size()); |
---|
191 | utility::yat_assert<std::runtime_error>(validation_data_.size()==size()); |
---|
192 | utility::yat_assert<std::runtime_error>(validation_target_.size()==size()); |
---|
193 | } |
---|
194 | |
---|
195 | |
---|
196 | template<typename T> |
---|
197 | SubsetGenerator<T>::SubsetGenerator(const Sampler& sampler, |
---|
198 | const T& data, |
---|
199 | FeatureSelector& fs) |
---|
200 | : f_selector_(&fs), sampler_(sampler) |
---|
201 | { |
---|
202 | utility::yat_assert<std::runtime_error>(target().size()==data.columns()); |
---|
203 | features_.reserve(size()); |
---|
204 | training_data_.reserve(size()); |
---|
205 | validation_data_.reserve(size()); |
---|
206 | build(data); |
---|
207 | utility::yat_assert<std::runtime_error>(training_data_.size()==size()); |
---|
208 | utility::yat_assert<std::runtime_error>(training_target_.size()==size()); |
---|
209 | utility::yat_assert<std::runtime_error>(validation_data_.size()==size()); |
---|
210 | utility::yat_assert<std::runtime_error>(validation_target_.size()==size()); |
---|
211 | } |
---|
212 | |
---|
213 | |
---|
214 | template<typename T> |
---|
215 | SubsetGenerator<T>::~SubsetGenerator() |
---|
216 | { |
---|
217 | utility::yat_assert<std::runtime_error>(training_data_.size()==validation_data_.size()); |
---|
218 | for (size_t i=0; i<training_data_.size(); i++) |
---|
219 | delete training_data_[i]; |
---|
220 | for (size_t i=0; i<validation_data_.size(); i++) |
---|
221 | delete validation_data_[i]; |
---|
222 | } |
---|
223 | |
---|
224 | |
---|
225 | template<typename T> |
---|
226 | void SubsetGenerator<T>::build(const MatrixLookup& ml) |
---|
227 | { |
---|
228 | for (size_t k=0; k<size(); k++){ |
---|
229 | training_target_.push_back(Target(target(),training_index(k))); |
---|
230 | validation_target_.push_back(Target(target(),validation_index(k))); |
---|
231 | // training data with no feature selection |
---|
232 | const MatrixLookup* train_data_all_feat = |
---|
233 | ml.training_data(training_index(k)); |
---|
234 | // use these data to create feature selection |
---|
235 | utility::yat_assert<std::runtime_error>(train_data_all_feat); |
---|
236 | f_selector_->update(*train_data_all_feat, training_target(k)); |
---|
237 | // get features |
---|
238 | features_.push_back(f_selector_->features()); |
---|
239 | utility::yat_assert<std::runtime_error>(train_data_all_feat); |
---|
240 | delete train_data_all_feat; |
---|
241 | |
---|
242 | // Dynamically allocated. Must be deleted in destructor. |
---|
243 | training_data_.push_back(new MatrixLookup(ml,features_.back(), |
---|
244 | training_index(k))); |
---|
245 | validation_data_.push_back(new MatrixLookup(ml,features_.back(), |
---|
246 | validation_index(k))); |
---|
247 | } |
---|
248 | |
---|
249 | } |
---|
250 | |
---|
251 | |
---|
252 | template<typename T> |
---|
253 | void SubsetGenerator<T>::build(const MatrixLookupWeighted& ml) |
---|
254 | { |
---|
255 | for (u_long k=0; k<size(); k++){ |
---|
256 | training_target_.push_back(Target(target(),training_index(k))); |
---|
257 | validation_target_.push_back(Target(target(),validation_index(k))); |
---|
258 | // training data with no feature selection |
---|
259 | const MatrixLookupWeighted* train_data_all_feat = |
---|
260 | ml.training_data(training_index(k)); |
---|
261 | // use these data to create feature selection |
---|
262 | f_selector_->update(*train_data_all_feat, training_target(k)); |
---|
263 | // get features |
---|
264 | features_.push_back(f_selector_->features()); |
---|
265 | delete train_data_all_feat; |
---|
266 | |
---|
267 | // Dynamically allocated. Must be deleted in destructor. |
---|
268 | training_data_.push_back(new MatrixLookupWeighted(ml, features_.back(), |
---|
269 | training_index(k))); |
---|
270 | validation_data_.push_back(new MatrixLookupWeighted(ml, features_.back(), |
---|
271 | validation_index(k))); |
---|
272 | } |
---|
273 | } |
---|
274 | |
---|
275 | template<typename T> |
---|
276 | void SubsetGenerator<T>::build(const KernelLookup& kernel) |
---|
277 | { |
---|
278 | for (u_long k=0; k<size(); k++){ |
---|
279 | training_target_.push_back(Target(target(),training_index(k))); |
---|
280 | validation_target_.push_back(Target(target(),validation_index(k))); |
---|
281 | const DataLookup2D* matrix = kernel.data(); |
---|
282 | // dynamically allocated must be deleted |
---|
283 | const DataLookup2D* training_matrix = |
---|
284 | matrix->training_data(training_index(k)); |
---|
285 | if (matrix->weighted()){ |
---|
286 | const MatrixLookupWeighted& ml = |
---|
287 | dynamic_cast<const MatrixLookupWeighted&>(*matrix); |
---|
288 | f_selector_->update(MatrixLookupWeighted(ml,training_index(k),false), |
---|
289 | training_target(k)); |
---|
290 | } |
---|
291 | else { |
---|
292 | const MatrixLookup& ml = |
---|
293 | dynamic_cast<const MatrixLookup&>(*matrix); |
---|
294 | f_selector_->update(MatrixLookup(ml,training_index(k), false), |
---|
295 | training_target(k)); |
---|
296 | } |
---|
297 | std::vector<size_t> dummie=f_selector_->features(); |
---|
298 | features_.push_back(dummie); |
---|
299 | //features_.push_back(f_selector_->features()); |
---|
300 | const KernelLookup* kl = kernel.selected(features_.back()); |
---|
301 | utility::yat_assert<std::runtime_error>(training_matrix); |
---|
302 | delete training_matrix; |
---|
303 | |
---|
304 | // Dynamically allocated. Must be deleted in destructor. |
---|
305 | training_data_.push_back(kl->training_data(training_index(k))); |
---|
306 | validation_data_.push_back(kl->validation_data(training_index(k), |
---|
307 | validation_index(k))); |
---|
308 | utility::yat_assert<std::runtime_error>(kl); |
---|
309 | delete kl; |
---|
310 | } |
---|
311 | } |
---|
312 | |
---|
313 | |
---|
314 | template<typename T> |
---|
315 | u_long SubsetGenerator<T>::size(void) const |
---|
316 | { |
---|
317 | return sampler_.size(); |
---|
318 | } |
---|
319 | |
---|
320 | |
---|
321 | template<typename T> |
---|
322 | const Target& SubsetGenerator<T>::target(void) const |
---|
323 | { |
---|
324 | return sampler_.target(); |
---|
325 | } |
---|
326 | |
---|
327 | |
---|
328 | template<typename T> |
---|
329 | const T& |
---|
330 | SubsetGenerator<T>::training_data(size_t i) const |
---|
331 | { |
---|
332 | return *(training_data_[i]); |
---|
333 | } |
---|
334 | |
---|
335 | |
---|
336 | template<typename T> |
---|
337 | const std::vector<size_t>& |
---|
338 | SubsetGenerator<T>::training_features(typename std::vector<size_t>::size_type i) const |
---|
339 | { |
---|
340 | return f_selector_ ? features_[i] : features_[0]; |
---|
341 | } |
---|
342 | |
---|
343 | |
---|
344 | template<typename T> |
---|
345 | const std::vector<size_t>& |
---|
346 | SubsetGenerator<T>::training_index(std::vector<size_t>::size_type i) const |
---|
347 | { |
---|
348 | return sampler_.training_index(i); |
---|
349 | } |
---|
350 | |
---|
351 | |
---|
352 | template<typename T> |
---|
353 | const Target& |
---|
354 | SubsetGenerator<T>::training_target(std::vector<Target>::size_type i) const |
---|
355 | { |
---|
356 | return training_target_[i]; |
---|
357 | } |
---|
358 | |
---|
359 | |
---|
360 | template<typename T> |
---|
361 | const T& |
---|
362 | SubsetGenerator<T>::validation_data(size_t i) const |
---|
363 | { |
---|
364 | return *(validation_data_[i]); |
---|
365 | } |
---|
366 | |
---|
367 | |
---|
368 | template<typename T> |
---|
369 | const std::vector<size_t>& |
---|
370 | SubsetGenerator<T>::validation_index(std::vector<size_t>::size_type i) const |
---|
371 | { |
---|
372 | return sampler_.validation_index(i); |
---|
373 | } |
---|
374 | |
---|
375 | |
---|
376 | template<typename T> |
---|
377 | const Target& |
---|
378 | SubsetGenerator<T>::validation_target(std::vector<Target>::size_type i) const |
---|
379 | { |
---|
380 | return validation_target_[i]; |
---|
381 | } |
---|
382 | |
---|
383 | }}} // of namespace classifier, yat, and theplu |
---|
384 | |
---|
385 | #endif |
---|
386 | |
---|