[142] | 1 | package de.ugoe.cs.cpdp.execution; |
---|
| 2 | |
---|
| 3 | import java.io.File; |
---|
| 4 | import java.util.Collections; |
---|
| 5 | import java.util.LinkedList; |
---|
| 6 | import java.util.List; |
---|
| 7 | import java.util.logging.Level; |
---|
| 8 | |
---|
| 9 | import org.apache.commons.collections4.list.SetUniqueList; |
---|
| 10 | |
---|
| 11 | import de.ugoe.cs.cpdp.ExperimentConfiguration; |
---|
| 12 | import de.ugoe.cs.cpdp.dataprocessing.IProcessesingStrategy; |
---|
| 13 | import de.ugoe.cs.cpdp.dataprocessing.ISetWiseProcessingStrategy; |
---|
| 14 | import de.ugoe.cs.cpdp.dataselection.IPointWiseDataselectionStrategy; |
---|
| 15 | import de.ugoe.cs.cpdp.dataselection.ISetWiseDataselectionStrategy; |
---|
| 16 | import de.ugoe.cs.cpdp.eval.IEvaluationStrategy; |
---|
| 17 | import de.ugoe.cs.cpdp.eval.IResultStorage; |
---|
| 18 | import de.ugoe.cs.cpdp.loader.IVersionLoader; |
---|
| 19 | import de.ugoe.cs.cpdp.training.ISetWiseTestdataAwareTrainingStrategy; |
---|
| 20 | import de.ugoe.cs.cpdp.training.ISetWiseTrainingStrategy; |
---|
| 21 | import de.ugoe.cs.cpdp.training.ITestAwareTrainingStrategy; |
---|
| 22 | import de.ugoe.cs.cpdp.training.ITrainer; |
---|
| 23 | import de.ugoe.cs.cpdp.training.ITrainingStrategy; |
---|
| 24 | import de.ugoe.cs.cpdp.training.IWekaCompatibleTrainer; |
---|
| 25 | import de.ugoe.cs.cpdp.versions.IVersionFilter; |
---|
| 26 | import de.ugoe.cs.cpdp.versions.SoftwareVersion; |
---|
| 27 | import de.ugoe.cs.util.console.Console; |
---|
| 28 | import weka.core.Instances; |
---|
| 29 | |
---|
| 30 | public class HeterogeneousExperiment implements IExecutionStrategy { |
---|
| 31 | |
---|
| 32 | /** |
---|
| 33 | * configuration of the experiment |
---|
| 34 | */ |
---|
| 35 | protected final ExperimentConfiguration config; |
---|
| 36 | |
---|
| 37 | /** |
---|
| 38 | * Constructor. Creates a new experiment based on a configuration. |
---|
| 39 | * |
---|
| 40 | * @param config |
---|
| 41 | * configuration of the experiment |
---|
| 42 | */ |
---|
| 43 | public HeterogeneousExperiment(ExperimentConfiguration config) { |
---|
| 44 | this.config = config; |
---|
| 45 | } |
---|
| 46 | |
---|
| 47 | |
---|
| 48 | /** |
---|
| 49 | * DUBLICATE FROM AbstractCrossProjectExperiment |
---|
| 50 | */ |
---|
| 51 | private boolean isVersion(SoftwareVersion version, List<IVersionFilter> filters) { |
---|
| 52 | boolean result = true; |
---|
| 53 | for (IVersionFilter filter : filters) { |
---|
| 54 | result &= !filter.apply(version); |
---|
| 55 | } |
---|
| 56 | return result; |
---|
| 57 | } |
---|
| 58 | |
---|
| 59 | /** |
---|
| 60 | * DUBLICATE FROM AbstractCrossProjectExperiment |
---|
| 61 | */ |
---|
| 62 | public static Instances makeSingleTrainingSet(SetUniqueList<Instances> traindataSet) { |
---|
| 63 | Instances traindataFull = null; |
---|
| 64 | for (Instances traindata : traindataSet) { |
---|
| 65 | if (traindataFull == null) { |
---|
| 66 | traindataFull = new Instances(traindata); |
---|
| 67 | } |
---|
| 68 | else { |
---|
| 69 | for (int i = 0; i < traindata.numInstances(); i++) { |
---|
| 70 | traindataFull.add(traindata.instance(i)); |
---|
| 71 | } |
---|
| 72 | } |
---|
| 73 | } |
---|
| 74 | return traindataFull; |
---|
| 75 | } |
---|
| 76 | |
---|
| 77 | /** |
---|
| 78 | * <p> |
---|
| 79 | * Defines which products are allowed for training. |
---|
| 80 | * </p> |
---|
| 81 | * |
---|
| 82 | * @param trainingVersion |
---|
| 83 | * training version |
---|
| 84 | * @param testVersion |
---|
| 85 | * test candidate |
---|
| 86 | * @return true if test candidate can be used for training |
---|
| 87 | */ |
---|
| 88 | protected boolean isTrainingVersion(SoftwareVersion trainingVersion, |
---|
| 89 | SoftwareVersion testVersion) { |
---|
| 90 | if(testVersion.getDataset().equals(trainingVersion.getDataset())) { |
---|
| 91 | return false; |
---|
| 92 | } |
---|
| 93 | |
---|
| 94 | return true; |
---|
| 95 | } |
---|
| 96 | |
---|
| 97 | @Override |
---|
| 98 | public void run() { |
---|
| 99 | final List<SoftwareVersion> versions = new LinkedList<>(); |
---|
| 100 | |
---|
| 101 | for (IVersionLoader loader : config.getLoaders()) { |
---|
| 102 | versions.addAll(loader.load()); |
---|
| 103 | } |
---|
| 104 | |
---|
| 105 | for (IVersionFilter filter : config.getVersionFilters()) { |
---|
| 106 | filter.apply(versions); |
---|
| 107 | } |
---|
| 108 | boolean writeHeader = true; |
---|
| 109 | int versionCount = 1; |
---|
| 110 | int testVersionCount = 0; |
---|
| 111 | |
---|
| 112 | // cahnged |
---|
| 113 | for (SoftwareVersion testVersion : versions) { |
---|
| 114 | if (isVersion(testVersion, config.getTestVersionFilters())) { |
---|
| 115 | for (SoftwareVersion trainingVersion : versions) { |
---|
| 116 | if (isVersion(trainingVersion, config.getTrainingVersionFilters())) { |
---|
| 117 | testVersionCount++; |
---|
| 118 | } |
---|
| 119 | } |
---|
| 120 | } |
---|
| 121 | } |
---|
| 122 | |
---|
| 123 | // sort versions |
---|
| 124 | Collections.sort(versions); |
---|
| 125 | |
---|
| 126 | // todo: test version check problematic |
---|
| 127 | // |
---|
| 128 | for (SoftwareVersion testVersion : versions) { |
---|
| 129 | if (isVersion(testVersion, config.getTestVersionFilters())) { |
---|
| 130 | |
---|
| 131 | |
---|
| 132 | // now iterate trainVersions |
---|
| 133 | for (SoftwareVersion trainingVersion : versions) { |
---|
| 134 | if (isVersion(trainingVersion, config.getTrainingVersionFilters())) { |
---|
| 135 | if (trainingVersion != testVersion) { |
---|
| 136 | if (isTrainingVersion(trainingVersion, testVersion)) { // checks if they are the same dataset |
---|
| 137 | |
---|
| 138 | Console.traceln(Level.INFO, |
---|
| 139 | String.format("[%s] [%02d/%02d] %s:%s starting", |
---|
| 140 | config.getExperimentName(), versionCount, |
---|
| 141 | testVersionCount, testVersion.getVersion(), trainingVersion.getVersion())); |
---|
| 142 | int numResultsAvailable = resultsAvailable(testVersion, trainingVersion); |
---|
| 143 | if (numResultsAvailable >= config.getRepetitions()) { |
---|
| 144 | Console.traceln(Level.INFO, |
---|
| 145 | String.format( |
---|
| 146 | "[%s] [%02d/%02d] %s:%s results already available; skipped", |
---|
| 147 | config.getExperimentName(), versionCount, |
---|
| 148 | testVersionCount, testVersion.getVersion(), trainingVersion.getVersion())); |
---|
| 149 | versionCount++; |
---|
| 150 | continue; |
---|
| 151 | } |
---|
| 152 | |
---|
| 153 | // Setup testdata and training data |
---|
| 154 | Instances testdata = testVersion.getInstances(); |
---|
| 155 | List<Double> efforts = testVersion.getEfforts(); |
---|
| 156 | Instances traindata = trainingVersion.getInstances(); |
---|
| 157 | |
---|
| 158 | // only one set |
---|
| 159 | SetUniqueList<Instances> traindataSet = |
---|
| 160 | SetUniqueList.setUniqueList(new LinkedList<Instances>()); |
---|
| 161 | traindataSet.add(traindata); |
---|
| 162 | |
---|
| 163 | for (ISetWiseProcessingStrategy processor : config.getSetWisePreprocessors()) { |
---|
| 164 | Console.traceln(Level.FINE, |
---|
| 165 | String.format( |
---|
| 166 | "[%s] [%02d/%02d] %s:%s applying setwise preprocessor %s", |
---|
| 167 | config.getExperimentName(), versionCount, |
---|
| 168 | testVersionCount, testVersion.getVersion(), trainingVersion.getVersion(), |
---|
| 169 | processor.getClass().getName())); |
---|
| 170 | processor.apply(testdata, traindataSet); |
---|
| 171 | } |
---|
| 172 | for (ISetWiseDataselectionStrategy dataselector : config.getSetWiseSelectors()) { |
---|
| 173 | Console |
---|
| 174 | .traceln(Level.FINE, |
---|
| 175 | String.format("[%s] [%02d/%02d] %s: applying setwise selection %s", |
---|
| 176 | config.getExperimentName(), versionCount, |
---|
| 177 | testVersionCount, testVersion.getVersion(), |
---|
| 178 | dataselector.getClass().getName())); |
---|
| 179 | dataselector.apply(testdata, traindataSet); |
---|
| 180 | } |
---|
| 181 | for (ISetWiseProcessingStrategy processor : config.getSetWisePostprocessors()) { |
---|
| 182 | Console.traceln(Level.FINE, |
---|
| 183 | String.format( |
---|
| 184 | "[%s] [%02d/%02d] %s: applying setwise postprocessor %s", |
---|
| 185 | config.getExperimentName(), versionCount, |
---|
| 186 | testVersionCount, testVersion.getVersion(), |
---|
| 187 | processor.getClass().getName())); |
---|
| 188 | processor.apply(testdata, traindataSet); |
---|
| 189 | } |
---|
| 190 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) { |
---|
| 191 | Console |
---|
| 192 | .traceln(Level.FINE, |
---|
| 193 | String.format("[%s] [%02d/%02d] %s: applying setwise trainer %s", |
---|
| 194 | config.getExperimentName(), versionCount, |
---|
| 195 | testVersionCount, testVersion.getVersion(), |
---|
| 196 | setwiseTrainer.getName())); |
---|
| 197 | setwiseTrainer.apply(traindataSet); |
---|
| 198 | } |
---|
| 199 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config |
---|
| 200 | .getSetWiseTestdataAwareTrainers()) |
---|
| 201 | { |
---|
| 202 | Console.traceln(Level.FINE, |
---|
| 203 | String.format( |
---|
| 204 | "[%s] [%02d/%02d] %s:%s applying testdata aware setwise trainer %s", |
---|
| 205 | config.getExperimentName(), versionCount, |
---|
| 206 | testVersionCount, testVersion.getVersion(), trainingVersion.getVersion(), |
---|
| 207 | setwiseTestdataAwareTrainer.getName())); |
---|
| 208 | setwiseTestdataAwareTrainer.apply(traindataSet, testdata); |
---|
| 209 | } |
---|
| 210 | |
---|
| 211 | |
---|
| 212 | // this part will not work in heterogeneous |
---|
| 213 | //Instances traindata = makeSingleTrainingSet(traindataSet); |
---|
| 214 | for (IProcessesingStrategy processor : config.getPreProcessors()) { |
---|
| 215 | Console.traceln(Level.FINE, |
---|
| 216 | String.format("[%s] [%02d/%02d] %s: applying preprocessor %s", |
---|
| 217 | config.getExperimentName(), versionCount, |
---|
| 218 | testVersionCount, testVersion.getVersion(), |
---|
| 219 | processor.getClass().getName())); |
---|
| 220 | processor.apply(testdata, traindata); |
---|
| 221 | } |
---|
| 222 | for (IPointWiseDataselectionStrategy dataselector : config |
---|
| 223 | .getPointWiseSelectors()) |
---|
| 224 | { |
---|
| 225 | Console.traceln(Level.FINE, |
---|
| 226 | String.format( |
---|
| 227 | "[%s] [%02d/%02d] %s: applying pointwise selection %s", |
---|
| 228 | config.getExperimentName(), versionCount, |
---|
| 229 | testVersionCount, testVersion.getVersion(), |
---|
| 230 | dataselector.getClass().getName())); |
---|
| 231 | traindata = dataselector.apply(testdata, traindata); |
---|
| 232 | } |
---|
| 233 | for (IProcessesingStrategy processor : config.getPostProcessors()) { |
---|
| 234 | Console.traceln(Level.FINE, |
---|
| 235 | String.format( |
---|
| 236 | "[%s] [%02d/%02d] %s: applying setwise postprocessor %s", |
---|
| 237 | config.getExperimentName(), versionCount, |
---|
| 238 | testVersionCount, testVersion.getVersion(), |
---|
| 239 | processor.getClass().getName())); |
---|
| 240 | processor.apply(testdata, traindata); |
---|
| 241 | } |
---|
| 242 | for (ITrainingStrategy trainer : config.getTrainers()) { |
---|
| 243 | Console.traceln(Level.FINE, |
---|
| 244 | String.format("[%s] [%02d/%02d] %s: applying trainer %s", |
---|
| 245 | config.getExperimentName(), versionCount, |
---|
| 246 | testVersionCount, testVersion.getVersion(), |
---|
| 247 | trainer.getName())); |
---|
| 248 | trainer.apply(traindata); |
---|
| 249 | } |
---|
| 250 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) { |
---|
| 251 | Console.traceln(Level.FINE, |
---|
| 252 | String.format("[%s] [%02d/%02d] %s: applying trainer %s", |
---|
| 253 | config.getExperimentName(), versionCount, |
---|
| 254 | testVersionCount, testVersion.getVersion(), |
---|
| 255 | trainer.getName())); |
---|
| 256 | trainer.apply(testdata, traindata); |
---|
| 257 | } |
---|
| 258 | File resultsDir = new File(config.getResultsPath()); |
---|
| 259 | if (!resultsDir.exists()) { |
---|
| 260 | resultsDir.mkdir(); |
---|
| 261 | } |
---|
| 262 | for (IEvaluationStrategy evaluator : config.getEvaluators()) { |
---|
| 263 | Console.traceln(Level.FINE, |
---|
| 264 | String.format("[%s] [%02d/%02d] %s:%s applying evaluator %s", |
---|
| 265 | config.getExperimentName(), versionCount, |
---|
| 266 | testVersionCount, testVersion.getVersion(), trainingVersion.getVersion(), |
---|
| 267 | evaluator.getClass().getName())); |
---|
| 268 | List<ITrainer> allTrainers = new LinkedList<>(); |
---|
| 269 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) { |
---|
| 270 | allTrainers.add(setwiseTrainer); |
---|
| 271 | } |
---|
| 272 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config |
---|
| 273 | .getSetWiseTestdataAwareTrainers()) |
---|
| 274 | { |
---|
| 275 | allTrainers.add(setwiseTestdataAwareTrainer); |
---|
| 276 | } |
---|
| 277 | for (ITrainingStrategy trainer : config.getTrainers()) { |
---|
| 278 | allTrainers.add(trainer); |
---|
| 279 | } |
---|
| 280 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) { |
---|
| 281 | allTrainers.add(trainer); |
---|
| 282 | } |
---|
| 283 | if (writeHeader) { |
---|
| 284 | evaluator.setParameter(config.getResultsPath() + "/" + |
---|
| 285 | config.getExperimentName() + ".csv"); |
---|
| 286 | } |
---|
| 287 | evaluator.apply(testdata, traindata, allTrainers, efforts, writeHeader, |
---|
| 288 | config.getResultStorages()); |
---|
| 289 | writeHeader = false; |
---|
| 290 | } |
---|
| 291 | Console.traceln(Level.INFO, |
---|
| 292 | String.format("[%s] [%02d/%02d] %s: finished", |
---|
| 293 | config.getExperimentName(), versionCount, |
---|
| 294 | testVersionCount, testVersion.getVersion())); |
---|
| 295 | versionCount++; |
---|
| 296 | |
---|
| 297 | |
---|
| 298 | |
---|
| 299 | } |
---|
| 300 | } |
---|
| 301 | } |
---|
| 302 | } |
---|
| 303 | |
---|
| 304 | } /* end if check training*/ |
---|
| 305 | } /* end for iteration test version */ |
---|
| 306 | } |
---|
| 307 | |
---|
| 308 | /** |
---|
| 309 | * DUBLICATE FROM AbstractCrossProjectExperiment |
---|
| 310 | */ |
---|
| 311 | private int resultsAvailable(SoftwareVersion version, SoftwareVersion trainVersion) { |
---|
| 312 | if (config.getResultStorages().isEmpty()) { |
---|
| 313 | return 0; |
---|
| 314 | } |
---|
| 315 | |
---|
| 316 | List<ITrainer> allTrainers = new LinkedList<>(); |
---|
| 317 | for (ISetWiseTrainingStrategy setwiseTrainer : config.getSetWiseTrainers()) { |
---|
| 318 | allTrainers.add(setwiseTrainer); |
---|
| 319 | } |
---|
| 320 | for (ISetWiseTestdataAwareTrainingStrategy setwiseTestdataAwareTrainer : config |
---|
| 321 | .getSetWiseTestdataAwareTrainers()) |
---|
| 322 | { |
---|
| 323 | allTrainers.add(setwiseTestdataAwareTrainer); |
---|
| 324 | } |
---|
| 325 | for (ITrainingStrategy trainer : config.getTrainers()) { |
---|
| 326 | allTrainers.add(trainer); |
---|
| 327 | } |
---|
| 328 | for (ITestAwareTrainingStrategy trainer : config.getTestAwareTrainers()) { |
---|
| 329 | allTrainers.add(trainer); |
---|
| 330 | } |
---|
| 331 | |
---|
| 332 | int available = Integer.MAX_VALUE; |
---|
| 333 | for (IResultStorage storage : config.getResultStorages()) { |
---|
| 334 | String classifierName = ((IWekaCompatibleTrainer) allTrainers.get(0)).getName(); |
---|
| 335 | int curAvailable = storage.containsHeterogeneousResult(config.getExperimentName(), |
---|
| 336 | version.getVersion(), classifierName, trainVersion.getVersion()); |
---|
| 337 | if (curAvailable < available) { |
---|
| 338 | available = curAvailable; |
---|
| 339 | } |
---|
| 340 | } |
---|
| 341 | return available; |
---|
| 342 | } |
---|
| 343 | } |
---|