1 | // Copyright 2015 Georg-August-Universität Göttingen, Germany
|
---|
2 | //
|
---|
3 | // Licensed under the Apache License, Version 2.0 (the "License");
|
---|
4 | // you may not use this file except in compliance with the License.
|
---|
5 | // You may obtain a copy of the License at
|
---|
6 | //
|
---|
7 | // http://www.apache.org/licenses/LICENSE-2.0
|
---|
8 | //
|
---|
9 | // Unless required by applicable law or agreed to in writing, software
|
---|
10 | // distributed under the License is distributed on an "AS IS" BASIS,
|
---|
11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
---|
12 | // See the License for the specific language governing permissions and
|
---|
13 | // limitations under the License.
|
---|
14 |
|
---|
15 | package de.ugoe.cs.cpdp.execution;
|
---|
16 |
|
---|
17 | import java.io.File;
|
---|
18 | import java.util.Collections;
|
---|
19 | import java.util.LinkedList;
|
---|
20 | import java.util.List;
|
---|
21 | import java.util.logging.Level;
|
---|
22 |
|
---|
23 | import org.apache.commons.collections4.list.SetUniqueList;
|
---|
24 |
|
---|
25 | import de.ugoe.cs.cpdp.ExperimentConfiguration;
|
---|
26 | import de.ugoe.cs.cpdp.dataprocessing.IProcessesingStrategy;
|
---|
27 | import de.ugoe.cs.cpdp.dataprocessing.ISetWiseProcessingStrategy;
|
---|
28 | import de.ugoe.cs.cpdp.dataselection.IPointWiseDataselectionStrategy;
|
---|
29 | import de.ugoe.cs.cpdp.dataselection.ISetWiseDataselectionStrategy;
|
---|
30 | import de.ugoe.cs.cpdp.eval.IEvaluationStrategy;
|
---|
31 | import de.ugoe.cs.cpdp.eval.IResultStorage;
|
---|
32 | import de.ugoe.cs.cpdp.loader.IVersionLoader;
|
---|
33 | import de.ugoe.cs.cpdp.training.ISetWiseTestdataAwareTrainingStrategy;
|
---|
34 | import de.ugoe.cs.cpdp.training.ISetWiseTrainingStrategy;
|
---|
35 | import de.ugoe.cs.cpdp.training.ITestAwareTrainingStrategy;
|
---|
36 | import de.ugoe.cs.cpdp.training.ITrainer;
|
---|
37 | import de.ugoe.cs.cpdp.training.ITrainingStrategy;
|
---|
38 | import de.ugoe.cs.cpdp.training.IWekaCompatibleTrainer;
|
---|
39 | import de.ugoe.cs.cpdp.versions.IVersionFilter;
|
---|
40 | import de.ugoe.cs.cpdp.versions.SoftwareVersion;
|
---|
41 | import de.ugoe.cs.util.console.Console;
|
---|
42 | import weka.core.Instances;
|
---|
43 |
|
---|
44 | /**
|
---|
45 | * Class responsible for executing an experiment according to an {@link ExperimentConfiguration}.
|
---|
46 | * The steps of an experiment are as follows:
|
---|
47 | * <ul>
|
---|
48 | * <li>load the data from the provided data path</li>
|
---|
49 | * <li>filter the data sets according to the provided version filters</li>
|
---|
50 | * <li>execute the following steps for each data sets as test data that is not ignored through the
|
---|
51 | * test version filter:
|
---|
52 | * <ul>
|
---|
53 | * <li>filter the data sets to setup the candidate training data:
|
---|
54 | * <ul>
|
---|
55 | * <li>remove all data sets from the same project</li>
|
---|
56 | * <li>filter all data sets according to the training data filter
|
---|
57 | * </ul>
|
---|
58 | * </li>
|
---|
59 | * <li>apply the setwise preprocessors</li>
|
---|
60 | * <li>apply the setwise data selection algorithms</li>
|
---|
61 | * <li>apply the setwise postprocessors</li>
|
---|
62 | * <li>train the setwise training classifiers</li>
|
---|
63 | * <li>unify all remaining training data into one data set</li>
|
---|
64 | * <li>apply the preprocessors</li>
|
---|
65 | * <li>apply the pointwise data selection algorithms</li>
|
---|
66 | * <li>apply the postprocessors</li>
|
---|
67 | * <li>train the normal classifiers</li>
|
---|
68 | * <li>evaluate the results for all trained classifiers on the training data</li>
|
---|
69 | * </ul>
|
---|
70 | * </li>
|
---|
71 | * </ul>
|
---|
72 | *
|
---|
73 | * Note that this class implements {@link Runnable}, i.e., each experiment can be started in its own
|
---|
74 | * thread.
|
---|
75 | *
|
---|
76 | * @author Steffen Herbold
|
---|
77 | */
|
---|
78 | public abstract class AbstractCrossProjectExperiment implements IExecutionStrategy {
|
---|
79 |
|
---|
80 | /**
|
---|
81 | * configuration of the experiment
|
---|
82 | */
|
---|
83 | protected final ExperimentConfiguration config;
|
---|
84 |
|
---|
85 | /**
|
---|
86 | * Constructor. Creates a new experiment based on a configuration.
|
---|
87 | *
|
---|
88 | * @param config
|
---|
89 | * configuration of the experiment
|
---|
90 | */
|
---|
91 | public AbstractCrossProjectExperiment(ExperimentConfiguration config) {
|
---|
92 | this.config = config;
|
---|
93 | }
|
---|
94 |
|
---|
95 | /**
|
---|
96 | * <p>
|
---|
97 | * Defines which products are allowed for training.
|
---|
98 | * </p>
|
---|
99 | *
|
---|
100 | * @param trainingVersion
|
---|
101 | * training version
|
---|
102 | * @param testVersion
|
---|
103 | * test candidate
|
---|
104 | * @return true if test candidate can be used for training
|
---|
105 | */
|
---|
106 | protected abstract boolean isTrainingVersion(SoftwareVersion trainingVersion,
|
---|
107 | SoftwareVersion testVersion);
|
---|
108 |
|
---|
109 | /**
|
---|
110 | * Helper method that combines a set of Weka {@link Instances} sets into a single
|
---|
111 | * {@link Instances} set.
|
---|
112 | *
|
---|
113 | * @param traindataSet
|
---|
114 | * set of {@link Instances} to be combines
|
---|
115 | * @return single {@link Instances} set
|
---|
116 | */
|
---|
117 | public static Instances makeSingleTrainingSet(SetUniqueList<Instances> traindataSet) {
|
---|
118 | Instances traindataFull = null;
|
---|
119 | for (Instances traindata : traindataSet) {
|
---|
120 | if (traindataFull == null) {
|
---|
121 | traindataFull = new Instances(traindata);
|
---|
122 | }
|
---|
123 | else {
|
---|
124 | for (int i = 0; i < traindata.numInstances(); i++) {
|
---|
125 | traindataFull.add(traindata.instance(i));
|
---|
126 | }
|
---|
127 | }
|
---|
128 | }
|
---|
129 | return traindataFull;
|
---|
130 | }
|
---|
131 |
|
---|
132 | /**
|
---|
133 | * Executes the experiment with the steps as described in the class comment.
|
---|
134 | *
|
---|
135 | * @see Runnable#run()
|
---|
136 | */
|
---|
137 | @Override
|
---|
138 | public void run() {
|
---|
139 | final List<SoftwareVersion> versions = new LinkedList<>();
|
---|
140 |
|
---|
141 | for (IVersionLoader loader : config.getLoaders()) {
|
---|
142 | versions.addAll(loader.load());
|
---|
143 | }
|
---|
144 |
|
---|
145 | for (IVersionFilter filter : config.getVersionFilters()) {
|
---|
146 | filter.apply(versions);
|
---|
147 | }
|
---|
148 | boolean writeHeader = true;
|
---|
149 | int versionCount = 1;
|
---|
150 | int testVersionCount = 0;
|
---|
151 |
|
---|
152 | for (SoftwareVersion testVersion : versions) {
|
---|
153 | if (isVersion(testVersion, config.getTestVersionFilters())) {
|
---|
154 | testVersionCount++;
|
---|
155 | }
|
---|
156 | }
|
---|
157 |
|
---|
158 | // sort versions
|
---|
159 | Collections.sort(versions);
|
---|
160 |
|
---|
161 | for (SoftwareVersion testVersion : versions) {
|
---|
162 | if (isVersion(testVersion, config.getTestVersionFilters())) {
|
---|
163 | Console.traceln(Level.INFO,
|
---|
164 | String.format("[%s] [%02d/%02d] %s: starting",
|
---|
165 | config.getExperimentName(), versionCount,
|
---|
166 | testVersionCount, testVersion.getVersion()));
|
---|
167 | int numResultsAvailable = resultsAvailable(testVersion);
|
---|
168 | if (numResultsAvailable >= config.getRepetitions()) {
|
---|
169 | Console.traceln(Level.INFO,
|
---|
170 | String.format(
|
---|
171 | "[%s] [%02d/%02d] %s: results already available; skipped",
|
---|
172 | config.getExperimentName(), versionCount,
|
---|
173 | testVersionCount, testVersion.getVersion()));
|
---|
174 | versionCount++;
|
---|
175 | continue;
|
---|
176 | }
|
---|
177 |
|
---|
178 | // Setup testdata and training data
|
---|
179 | Instances testdata = testVersion.getInstances();
|
---|
180 | List<Double> efforts = testVersion.getEfforts();
|
---|
181 | SetUniqueList<Instances> traindataSet =
|
---|
182 | SetUniqueList.setUniqueList(new LinkedList<Instances>());
|
---|
183 | for (SoftwareVersion trainingVersion : versions) {
|
---|
184 | if (isVersion(trainingVersion, config.getTrainingVersionFilters())) {
|
---|
185 | if (trainingVersion != testVersion) {
|
---|
186 | if (isTrainingVersion(trainingVersion, testVersion)) {
|
---|
187 | traindataSet.add(trainingVersion.getInstances());
|
---|
188 | }
|
---|
189 | }
|
---|
190 | }
|
---|
191 | }
|
---|
192 |
|
---|
193 | for (ISetWiseProcessingStrategy processor : config.getSetWisePreprocessors()) {
|
---|
194 | Console.traceln(Level.FINE,
|
---|
195 | String.format(
|
---|
196 | "[%s] [%02d/%02d] %s: applying setwise preprocessor %s",
|
---|
197 | config.getExperimentName(), versionCount,
|
---|
198 | testVersionCount, testVersion.getVersion(),
|
---|
199 | processor.getClass().getName()));
|
---|
200 | processor.apply(testdata, traindataSet);
|
---|
201 | }
|
---|
202 | for (ISetWiseDataselectionStrategy dataselector : config.getSetWiseSelectors()) {
|
---|
203 | Console
|
---|
204 | .traceln(Level.FINE,
|
---|
205 | String.format("[%s] [%02d/%02d] %s: applying setwise selection %s",
|
---|
206 | config.getExperimentName(), versionCount,
|
---|
207 | testVersionCount, testVersion.getVersion(),
|
---|
208 | dataselector.getClass().getName()));
|
---|
209 | dataselector.apply(testdata, traindataSet);
|
---|
210 | }
|
---|
211 | for (ISetWiseProcessingStrategy processor : config.getSetWisePostprocessors()) {
|
---|
212 | Console.traceln(Level.FINE,
|
---|
213 | String.format(
|
---|
214 | "[%s] [%02d/%02d] %s: applying setwise postprocessor %s",
|
---|
215 | config.getExperimentName(), versionCount,
|
---|
216 | testVersionCount, testVersion.getVersion(),
|
---|
217 | processor.getClass().getName()));
|
---|
218 | processor.apply(testdata, traindataSet);
|
---|
219 | }
|
---|
220 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) {
|
---|
221 | Console
|
---|
222 | .traceln(Level.FINE,
|
---|
223 | String.format("[%s] [%02d/%02d] %s: applying setwise trainer %s",
|
---|
224 | config.getExperimentName(), versionCount,
|
---|
225 | testVersionCount, testVersion.getVersion(),
|
---|
226 | setwiseTrainer.getName()));
|
---|
227 | setwiseTrainer.apply(traindataSet);
|
---|
228 | }
|
---|
229 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config
|
---|
230 | .getSetWiseTestdataAwareTrainers())
|
---|
231 | {
|
---|
232 | Console.traceln(Level.FINE,
|
---|
233 | String.format(
|
---|
234 | "[%s] [%02d/%02d] %s: applying testdata aware setwise trainer %s",
|
---|
235 | config.getExperimentName(), versionCount,
|
---|
236 | testVersionCount, testVersion.getVersion(),
|
---|
237 | setwiseTestdataAwareTrainer.getName()));
|
---|
238 | setwiseTestdataAwareTrainer.apply(traindataSet, testdata);
|
---|
239 | }
|
---|
240 | Instances traindata = makeSingleTrainingSet(traindataSet);
|
---|
241 | for (IProcessesingStrategy processor : config.getPreProcessors()) {
|
---|
242 | Console.traceln(Level.FINE,
|
---|
243 | String.format("[%s] [%02d/%02d] %s: applying preprocessor %s",
|
---|
244 | config.getExperimentName(), versionCount,
|
---|
245 | testVersionCount, testVersion.getVersion(),
|
---|
246 | processor.getClass().getName()));
|
---|
247 | processor.apply(testdata, traindata);
|
---|
248 | }
|
---|
249 | for (IPointWiseDataselectionStrategy dataselector : config
|
---|
250 | .getPointWiseSelectors())
|
---|
251 | {
|
---|
252 | Console.traceln(Level.FINE,
|
---|
253 | String.format(
|
---|
254 | "[%s] [%02d/%02d] %s: applying pointwise selection %s",
|
---|
255 | config.getExperimentName(), versionCount,
|
---|
256 | testVersionCount, testVersion.getVersion(),
|
---|
257 | dataselector.getClass().getName()));
|
---|
258 | traindata = dataselector.apply(testdata, traindata);
|
---|
259 | }
|
---|
260 | for (IProcessesingStrategy processor : config.getPostProcessors()) {
|
---|
261 | Console.traceln(Level.FINE,
|
---|
262 | String.format(
|
---|
263 | "[%s] [%02d/%02d] %s: applying setwise postprocessor %s",
|
---|
264 | config.getExperimentName(), versionCount,
|
---|
265 | testVersionCount, testVersion.getVersion(),
|
---|
266 | processor.getClass().getName()));
|
---|
267 | processor.apply(testdata, traindata);
|
---|
268 | }
|
---|
269 | for (ITrainingStrategy trainer : config.getTrainers()) {
|
---|
270 | Console.traceln(Level.FINE,
|
---|
271 | String.format("[%s] [%02d/%02d] %s: applying trainer %s",
|
---|
272 | config.getExperimentName(), versionCount,
|
---|
273 | testVersionCount, testVersion.getVersion(),
|
---|
274 | trainer.getName()));
|
---|
275 | trainer.apply(traindata);
|
---|
276 | }
|
---|
277 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) {
|
---|
278 | Console.traceln(Level.FINE,
|
---|
279 | String.format("[%s] [%02d/%02d] %s: applying trainer %s",
|
---|
280 | config.getExperimentName(), versionCount,
|
---|
281 | testVersionCount, testVersion.getVersion(),
|
---|
282 | trainer.getName()));
|
---|
283 | trainer.apply(testdata, traindata);
|
---|
284 | }
|
---|
285 | File resultsDir = new File(config.getResultsPath());
|
---|
286 | if (!resultsDir.exists()) {
|
---|
287 | resultsDir.mkdir();
|
---|
288 | }
|
---|
289 | for (IEvaluationStrategy evaluator : config.getEvaluators()) {
|
---|
290 | Console.traceln(Level.FINE,
|
---|
291 | String.format("[%s] [%02d/%02d] %s: applying evaluator %s",
|
---|
292 | config.getExperimentName(), versionCount,
|
---|
293 | testVersionCount, testVersion.getVersion(),
|
---|
294 | evaluator.getClass().getName()));
|
---|
295 | List<ITrainer> allTrainers = new LinkedList<>();
|
---|
296 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) {
|
---|
297 | allTrainers.add(setwiseTrainer);
|
---|
298 | }
|
---|
299 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config
|
---|
300 | .getSetWiseTestdataAwareTrainers())
|
---|
301 | {
|
---|
302 | allTrainers.add(setwiseTestdataAwareTrainer);
|
---|
303 | }
|
---|
304 | for (ITrainingStrategy trainer : config.getTrainers()) {
|
---|
305 | allTrainers.add(trainer);
|
---|
306 | }
|
---|
307 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) {
|
---|
308 | allTrainers.add(trainer);
|
---|
309 | }
|
---|
310 | if (writeHeader) {
|
---|
311 | evaluator.setParameter(config.getResultsPath() + "/" +
|
---|
312 | config.getExperimentName() + ".csv");
|
---|
313 | }
|
---|
314 | evaluator.apply(testdata, traindata, allTrainers, efforts, writeHeader,
|
---|
315 | config.getResultStorages());
|
---|
316 | writeHeader = false;
|
---|
317 | }
|
---|
318 | Console.traceln(Level.INFO,
|
---|
319 | String.format("[%s] [%02d/%02d] %s: finished",
|
---|
320 | config.getExperimentName(), versionCount,
|
---|
321 | testVersionCount, testVersion.getVersion()));
|
---|
322 | versionCount++;
|
---|
323 | }
|
---|
324 | }
|
---|
325 | }
|
---|
326 |
|
---|
327 | /**
|
---|
328 | * Helper method that checks if a version passes all filters.
|
---|
329 | *
|
---|
330 | * @param version
|
---|
331 | * version that is checked
|
---|
332 | * @param filters
|
---|
333 | * list of the filters
|
---|
334 | * @return true, if the version passes all filters, false otherwise
|
---|
335 | */
|
---|
336 | private boolean isVersion(SoftwareVersion version, List<IVersionFilter> filters) {
|
---|
337 | boolean result = true;
|
---|
338 | for (IVersionFilter filter : filters) {
|
---|
339 | result &= !filter.apply(version);
|
---|
340 | }
|
---|
341 | return result;
|
---|
342 | }
|
---|
343 |
|
---|
344 | /**
|
---|
345 | * <p>
|
---|
346 | * helper function that checks if the results are already in the data store
|
---|
347 | * </p>
|
---|
348 | *
|
---|
349 | * @param version
|
---|
350 | * version for which the results are checked
|
---|
351 | * @return
|
---|
352 | */
|
---|
353 | private int resultsAvailable(SoftwareVersion version) {
|
---|
354 | if (config.getResultStorages().isEmpty()) {
|
---|
355 | return 0;
|
---|
356 | }
|
---|
357 |
|
---|
358 | List<ITrainer> allTrainers = new LinkedList<>();
|
---|
359 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) {
|
---|
360 | allTrainers.add(setwiseTrainer);
|
---|
361 | }
|
---|
362 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config
|
---|
363 | .getSetWiseTestdataAwareTrainers())
|
---|
364 | {
|
---|
365 | allTrainers.add(setwiseTestdataAwareTrainer);
|
---|
366 | }
|
---|
367 | for (ITrainingStrategy trainer : config.getTrainers()) {
|
---|
368 | allTrainers.add(trainer);
|
---|
369 | }
|
---|
370 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) {
|
---|
371 | allTrainers.add(trainer);
|
---|
372 | }
|
---|
373 |
|
---|
374 | int available = Integer.MAX_VALUE;
|
---|
375 | for (IResultStorage storage : config.getResultStorages()) {
|
---|
376 | String classifierName = ((IWekaCompatibleTrainer) allTrainers.get(0)).getName();
|
---|
377 | int curAvailable = storage.containsResult(config.getExperimentName(),
|
---|
378 | version.getVersion(), classifierName);
|
---|
379 | if (curAvailable < available) {
|
---|
380 | available = curAvailable;
|
---|
381 | }
|
---|
382 | }
|
---|
383 | return available;
|
---|
384 | }
|
---|
385 | }
|
---|