[4] | 1 | /* |
---|
| 2 | * This program is free software; you can redistribute it and/or modify |
---|
| 3 | * it under the terms of the GNU General Public License as published by |
---|
| 4 | * the Free Software Foundation; either version 2 of the License, or |
---|
| 5 | * (at your option) any later version. |
---|
| 6 | * |
---|
| 7 | * This program is distributed in the hope that it will be useful, |
---|
| 8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
| 9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
---|
| 10 | * GNU General Public License for more details. |
---|
| 11 | * |
---|
| 12 | * You should have received a copy of the GNU General Public License |
---|
| 13 | * along with this program; if not, write to the Free Software |
---|
| 14 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
---|
| 15 | */ |
---|
| 16 | |
---|
| 17 | /* |
---|
| 18 | * IBk.java |
---|
| 19 | * Copyright (C) 1999 University of Waikato, Hamilton, New Zealand |
---|
| 20 | * |
---|
| 21 | */ |
---|
| 22 | |
---|
| 23 | package weka.classifiers.lazy; |
---|
| 24 | |
---|
| 25 | import weka.classifiers.Classifier; |
---|
| 26 | import weka.classifiers.AbstractClassifier; |
---|
| 27 | import weka.classifiers.UpdateableClassifier; |
---|
| 28 | import weka.core.Attribute; |
---|
| 29 | import weka.core.Capabilities; |
---|
| 30 | import weka.core.Instance; |
---|
| 31 | import weka.core.Instances; |
---|
| 32 | import weka.core.neighboursearch.LinearNNSearch; |
---|
| 33 | import weka.core.neighboursearch.NearestNeighbourSearch; |
---|
| 34 | import weka.core.Option; |
---|
| 35 | import weka.core.OptionHandler; |
---|
| 36 | import weka.core.RevisionUtils; |
---|
| 37 | import weka.core.SelectedTag; |
---|
| 38 | import weka.core.Tag; |
---|
| 39 | import weka.core.TechnicalInformation; |
---|
| 40 | import weka.core.TechnicalInformationHandler; |
---|
| 41 | import weka.core.Utils; |
---|
| 42 | import weka.core.WeightedInstancesHandler; |
---|
| 43 | import weka.core.Capabilities.Capability; |
---|
| 44 | import weka.core.TechnicalInformation.Field; |
---|
| 45 | import weka.core.TechnicalInformation.Type; |
---|
| 46 | import weka.core.AdditionalMeasureProducer; |
---|
| 47 | |
---|
| 48 | import java.util.Enumeration; |
---|
| 49 | import java.util.Vector; |
---|
| 50 | |
---|
| 51 | /** |
---|
| 52 | <!-- globalinfo-start --> |
---|
| 53 | * K-nearest neighbours classifier. Can select appropriate value of K based on cross-validation. Can also do distance weighting.<br/> |
---|
| 54 | * <br/> |
---|
| 55 | * For more information, see<br/> |
---|
| 56 | * <br/> |
---|
| 57 | * D. Aha, D. Kibler (1991). Instance-based learning algorithms. Machine Learning. 6:37-66. |
---|
| 58 | * <p/> |
---|
| 59 | <!-- globalinfo-end --> |
---|
| 60 | * |
---|
| 61 | <!-- technical-bibtex-start --> |
---|
| 62 | * BibTeX: |
---|
| 63 | * <pre> |
---|
| 64 | * @article{Aha1991, |
---|
| 65 | * author = {D. Aha and D. Kibler}, |
---|
| 66 | * journal = {Machine Learning}, |
---|
| 67 | * pages = {37-66}, |
---|
| 68 | * title = {Instance-based learning algorithms}, |
---|
| 69 | * volume = {6}, |
---|
| 70 | * year = {1991} |
---|
| 71 | * } |
---|
| 72 | * </pre> |
---|
| 73 | * <p/> |
---|
| 74 | <!-- technical-bibtex-end --> |
---|
| 75 | * |
---|
| 76 | <!-- options-start --> |
---|
| 77 | * Valid options are: <p/> |
---|
| 78 | * |
---|
| 79 | * <pre> -I |
---|
| 80 | * Weight neighbours by the inverse of their distance |
---|
| 81 | * (use when k > 1)</pre> |
---|
| 82 | * |
---|
| 83 | * <pre> -F |
---|
| 84 | * Weight neighbours by 1 - their distance |
---|
| 85 | * (use when k > 1)</pre> |
---|
| 86 | * |
---|
| 87 | * <pre> -K <number of neighbors> |
---|
| 88 | * Number of nearest neighbours (k) used in classification. |
---|
| 89 | * (Default = 1)</pre> |
---|
| 90 | * |
---|
| 91 | * <pre> -E |
---|
| 92 | * Minimise mean squared error rather than mean absolute |
---|
| 93 | * error when using -X option with numeric prediction.</pre> |
---|
| 94 | * |
---|
| 95 | * <pre> -W <window size> |
---|
| 96 | * Maximum number of training instances maintained. |
---|
| 97 | * Training instances are dropped FIFO. (Default = no window)</pre> |
---|
| 98 | * |
---|
| 99 | * <pre> -X |
---|
| 100 | * Select the number of nearest neighbours between 1 |
---|
| 101 | * and the k value specified using hold-one-out evaluation |
---|
| 102 | * on the training data (use when k > 1)</pre> |
---|
| 103 | * |
---|
| 104 | * <pre> -A |
---|
| 105 | * The nearest neighbour search algorithm to use (default: weka.core.neighboursearch.LinearNNSearch). |
---|
| 106 | * </pre> |
---|
| 107 | * |
---|
| 108 | <!-- options-end --> |
---|
| 109 | * |
---|
| 110 | * @author Stuart Inglis (singlis@cs.waikato.ac.nz) |
---|
| 111 | * @author Len Trigg (trigg@cs.waikato.ac.nz) |
---|
| 112 | * @author Eibe Frank (eibe@cs.waikato.ac.nz) |
---|
| 113 | * @version $Revision: 5928 $ |
---|
| 114 | */ |
---|
| 115 | public class IBk |
---|
| 116 | extends AbstractClassifier |
---|
| 117 | implements OptionHandler, UpdateableClassifier, WeightedInstancesHandler, |
---|
| 118 | TechnicalInformationHandler, AdditionalMeasureProducer { |
---|
| 119 | |
---|
| 120 | /** for serialization. */ |
---|
| 121 | static final long serialVersionUID = -3080186098777067172L; |
---|
| 122 | |
---|
| 123 | /** The training instances used for classification. */ |
---|
| 124 | protected Instances m_Train; |
---|
| 125 | |
---|
| 126 | /** The number of class values (or 1 if predicting numeric). */ |
---|
| 127 | protected int m_NumClasses; |
---|
| 128 | |
---|
| 129 | /** The class attribute type. */ |
---|
| 130 | protected int m_ClassType; |
---|
| 131 | |
---|
| 132 | /** The number of neighbours to use for classification (currently). */ |
---|
| 133 | protected int m_kNN; |
---|
| 134 | |
---|
| 135 | /** |
---|
| 136 | * The value of kNN provided by the user. This may differ from |
---|
| 137 | * m_kNN if cross-validation is being used. |
---|
| 138 | */ |
---|
| 139 | protected int m_kNNUpper; |
---|
| 140 | |
---|
| 141 | /** |
---|
| 142 | * Whether the value of k selected by cross validation has |
---|
| 143 | * been invalidated by a change in the training instances. |
---|
| 144 | */ |
---|
| 145 | protected boolean m_kNNValid; |
---|
| 146 | |
---|
| 147 | /** |
---|
| 148 | * The maximum number of training instances allowed. When |
---|
| 149 | * this limit is reached, old training instances are removed, |
---|
| 150 | * so the training data is "windowed". Set to 0 for unlimited |
---|
| 151 | * numbers of instances. |
---|
| 152 | */ |
---|
| 153 | protected int m_WindowSize; |
---|
| 154 | |
---|
| 155 | /** Whether the neighbours should be distance-weighted. */ |
---|
| 156 | protected int m_DistanceWeighting; |
---|
| 157 | |
---|
| 158 | /** Whether to select k by cross validation. */ |
---|
| 159 | protected boolean m_CrossValidate; |
---|
| 160 | |
---|
| 161 | /** |
---|
| 162 | * Whether to minimise mean squared error rather than mean absolute |
---|
| 163 | * error when cross-validating on numeric prediction tasks. |
---|
| 164 | */ |
---|
| 165 | protected boolean m_MeanSquared; |
---|
| 166 | |
---|
| 167 | /** no weighting. */ |
---|
| 168 | public static final int WEIGHT_NONE = 1; |
---|
| 169 | /** weight by 1/distance. */ |
---|
| 170 | public static final int WEIGHT_INVERSE = 2; |
---|
| 171 | /** weight by 1-distance. */ |
---|
| 172 | public static final int WEIGHT_SIMILARITY = 4; |
---|
| 173 | /** possible instance weighting methods. */ |
---|
| 174 | public static final Tag [] TAGS_WEIGHTING = { |
---|
| 175 | new Tag(WEIGHT_NONE, "No distance weighting"), |
---|
| 176 | new Tag(WEIGHT_INVERSE, "Weight by 1/distance"), |
---|
| 177 | new Tag(WEIGHT_SIMILARITY, "Weight by 1-distance") |
---|
| 178 | }; |
---|
| 179 | |
---|
| 180 | /** for nearest-neighbor search. */ |
---|
| 181 | protected NearestNeighbourSearch m_NNSearch = new LinearNNSearch(); |
---|
| 182 | |
---|
| 183 | /** The number of attributes the contribute to a prediction. */ |
---|
| 184 | protected double m_NumAttributesUsed; |
---|
| 185 | |
---|
| 186 | /** |
---|
| 187 | * IBk classifier. Simple instance-based learner that uses the class |
---|
| 188 | * of the nearest k training instances for the class of the test |
---|
| 189 | * instances. |
---|
| 190 | * |
---|
| 191 | * @param k the number of nearest neighbors to use for prediction |
---|
| 192 | */ |
---|
| 193 | public IBk(int k) { |
---|
| 194 | |
---|
| 195 | init(); |
---|
| 196 | setKNN(k); |
---|
| 197 | } |
---|
| 198 | |
---|
| 199 | /** |
---|
| 200 | * IB1 classifer. Instance-based learner. Predicts the class of the |
---|
| 201 | * single nearest training instance for each test instance. |
---|
| 202 | */ |
---|
| 203 | public IBk() { |
---|
| 204 | |
---|
| 205 | init(); |
---|
| 206 | } |
---|
| 207 | |
---|
| 208 | /** |
---|
| 209 | * Returns a string describing classifier. |
---|
| 210 | * @return a description suitable for |
---|
| 211 | * displaying in the explorer/experimenter gui |
---|
| 212 | */ |
---|
| 213 | public String globalInfo() { |
---|
| 214 | |
---|
| 215 | return "K-nearest neighbours classifier. Can " |
---|
| 216 | + "select appropriate value of K based on cross-validation. Can also do " |
---|
| 217 | + "distance weighting.\n\n" |
---|
| 218 | + "For more information, see\n\n" |
---|
| 219 | + getTechnicalInformation().toString(); |
---|
| 220 | } |
---|
| 221 | |
---|
| 222 | /** |
---|
| 223 | * Returns an instance of a TechnicalInformation object, containing |
---|
| 224 | * detailed information about the technical background of this class, |
---|
| 225 | * e.g., paper reference or book this class is based on. |
---|
| 226 | * |
---|
| 227 | * @return the technical information about this class |
---|
| 228 | */ |
---|
| 229 | public TechnicalInformation getTechnicalInformation() { |
---|
| 230 | TechnicalInformation result; |
---|
| 231 | |
---|
| 232 | result = new TechnicalInformation(Type.ARTICLE); |
---|
| 233 | result.setValue(Field.AUTHOR, "D. Aha and D. Kibler"); |
---|
| 234 | result.setValue(Field.YEAR, "1991"); |
---|
| 235 | result.setValue(Field.TITLE, "Instance-based learning algorithms"); |
---|
| 236 | result.setValue(Field.JOURNAL, "Machine Learning"); |
---|
| 237 | result.setValue(Field.VOLUME, "6"); |
---|
| 238 | result.setValue(Field.PAGES, "37-66"); |
---|
| 239 | |
---|
| 240 | return result; |
---|
| 241 | } |
---|
| 242 | |
---|
| 243 | /** |
---|
| 244 | * Returns the tip text for this property. |
---|
| 245 | * @return tip text for this property suitable for |
---|
| 246 | * displaying in the explorer/experimenter gui |
---|
| 247 | */ |
---|
| 248 | public String KNNTipText() { |
---|
| 249 | return "The number of neighbours to use."; |
---|
| 250 | } |
---|
| 251 | |
---|
| 252 | /** |
---|
| 253 | * Set the number of neighbours the learner is to use. |
---|
| 254 | * |
---|
| 255 | * @param k the number of neighbours. |
---|
| 256 | */ |
---|
| 257 | public void setKNN(int k) { |
---|
| 258 | m_kNN = k; |
---|
| 259 | m_kNNUpper = k; |
---|
| 260 | m_kNNValid = false; |
---|
| 261 | } |
---|
| 262 | |
---|
| 263 | /** |
---|
| 264 | * Gets the number of neighbours the learner will use. |
---|
| 265 | * |
---|
| 266 | * @return the number of neighbours. |
---|
| 267 | */ |
---|
| 268 | public int getKNN() { |
---|
| 269 | |
---|
| 270 | return m_kNN; |
---|
| 271 | } |
---|
| 272 | |
---|
| 273 | /** |
---|
| 274 | * Returns the tip text for this property. |
---|
| 275 | * @return tip text for this property suitable for |
---|
| 276 | * displaying in the explorer/experimenter gui |
---|
| 277 | */ |
---|
| 278 | public String windowSizeTipText() { |
---|
| 279 | return "Gets the maximum number of instances allowed in the training " + |
---|
| 280 | "pool. The addition of new instances above this value will result " + |
---|
| 281 | "in old instances being removed. A value of 0 signifies no limit " + |
---|
| 282 | "to the number of training instances."; |
---|
| 283 | } |
---|
| 284 | |
---|
| 285 | /** |
---|
| 286 | * Gets the maximum number of instances allowed in the training |
---|
| 287 | * pool. The addition of new instances above this value will result |
---|
| 288 | * in old instances being removed. A value of 0 signifies no limit |
---|
| 289 | * to the number of training instances. |
---|
| 290 | * |
---|
| 291 | * @return Value of WindowSize. |
---|
| 292 | */ |
---|
| 293 | public int getWindowSize() { |
---|
| 294 | |
---|
| 295 | return m_WindowSize; |
---|
| 296 | } |
---|
| 297 | |
---|
| 298 | /** |
---|
| 299 | * Sets the maximum number of instances allowed in the training |
---|
| 300 | * pool. The addition of new instances above this value will result |
---|
| 301 | * in old instances being removed. A value of 0 signifies no limit |
---|
| 302 | * to the number of training instances. |
---|
| 303 | * |
---|
| 304 | * @param newWindowSize Value to assign to WindowSize. |
---|
| 305 | */ |
---|
| 306 | public void setWindowSize(int newWindowSize) { |
---|
| 307 | |
---|
| 308 | m_WindowSize = newWindowSize; |
---|
| 309 | } |
---|
| 310 | |
---|
| 311 | /** |
---|
| 312 | * Returns the tip text for this property. |
---|
| 313 | * @return tip text for this property suitable for |
---|
| 314 | * displaying in the explorer/experimenter gui |
---|
| 315 | */ |
---|
| 316 | public String distanceWeightingTipText() { |
---|
| 317 | |
---|
| 318 | return "Gets the distance weighting method used."; |
---|
| 319 | } |
---|
| 320 | |
---|
| 321 | /** |
---|
| 322 | * Gets the distance weighting method used. Will be one of |
---|
| 323 | * WEIGHT_NONE, WEIGHT_INVERSE, or WEIGHT_SIMILARITY |
---|
| 324 | * |
---|
| 325 | * @return the distance weighting method used. |
---|
| 326 | */ |
---|
| 327 | public SelectedTag getDistanceWeighting() { |
---|
| 328 | |
---|
| 329 | return new SelectedTag(m_DistanceWeighting, TAGS_WEIGHTING); |
---|
| 330 | } |
---|
| 331 | |
---|
| 332 | /** |
---|
| 333 | * Sets the distance weighting method used. Values other than |
---|
| 334 | * WEIGHT_NONE, WEIGHT_INVERSE, or WEIGHT_SIMILARITY will be ignored. |
---|
| 335 | * |
---|
| 336 | * @param newMethod the distance weighting method to use |
---|
| 337 | */ |
---|
| 338 | public void setDistanceWeighting(SelectedTag newMethod) { |
---|
| 339 | |
---|
| 340 | if (newMethod.getTags() == TAGS_WEIGHTING) { |
---|
| 341 | m_DistanceWeighting = newMethod.getSelectedTag().getID(); |
---|
| 342 | } |
---|
| 343 | } |
---|
| 344 | |
---|
| 345 | /** |
---|
| 346 | * Returns the tip text for this property. |
---|
| 347 | * @return tip text for this property suitable for |
---|
| 348 | * displaying in the explorer/experimenter gui |
---|
| 349 | */ |
---|
| 350 | public String meanSquaredTipText() { |
---|
| 351 | |
---|
| 352 | return "Whether the mean squared error is used rather than mean " |
---|
| 353 | + "absolute error when doing cross-validation for regression problems."; |
---|
| 354 | } |
---|
| 355 | |
---|
| 356 | /** |
---|
| 357 | * Gets whether the mean squared error is used rather than mean |
---|
| 358 | * absolute error when doing cross-validation. |
---|
| 359 | * |
---|
| 360 | * @return true if so. |
---|
| 361 | */ |
---|
| 362 | public boolean getMeanSquared() { |
---|
| 363 | |
---|
| 364 | return m_MeanSquared; |
---|
| 365 | } |
---|
| 366 | |
---|
| 367 | /** |
---|
| 368 | * Sets whether the mean squared error is used rather than mean |
---|
| 369 | * absolute error when doing cross-validation. |
---|
| 370 | * |
---|
| 371 | * @param newMeanSquared true if so. |
---|
| 372 | */ |
---|
| 373 | public void setMeanSquared(boolean newMeanSquared) { |
---|
| 374 | |
---|
| 375 | m_MeanSquared = newMeanSquared; |
---|
| 376 | } |
---|
| 377 | |
---|
| 378 | /** |
---|
| 379 | * Returns the tip text for this property. |
---|
| 380 | * @return tip text for this property suitable for |
---|
| 381 | * displaying in the explorer/experimenter gui |
---|
| 382 | */ |
---|
| 383 | public String crossValidateTipText() { |
---|
| 384 | |
---|
| 385 | return "Whether hold-one-out cross-validation will be used " + |
---|
| 386 | "to select the best k value."; |
---|
| 387 | } |
---|
| 388 | |
---|
| 389 | /** |
---|
| 390 | * Gets whether hold-one-out cross-validation will be used |
---|
| 391 | * to select the best k value. |
---|
| 392 | * |
---|
| 393 | * @return true if cross-validation will be used. |
---|
| 394 | */ |
---|
| 395 | public boolean getCrossValidate() { |
---|
| 396 | |
---|
| 397 | return m_CrossValidate; |
---|
| 398 | } |
---|
| 399 | |
---|
| 400 | /** |
---|
| 401 | * Sets whether hold-one-out cross-validation will be used |
---|
| 402 | * to select the best k value. |
---|
| 403 | * |
---|
| 404 | * @param newCrossValidate true if cross-validation should be used. |
---|
| 405 | */ |
---|
| 406 | public void setCrossValidate(boolean newCrossValidate) { |
---|
| 407 | |
---|
| 408 | m_CrossValidate = newCrossValidate; |
---|
| 409 | } |
---|
| 410 | |
---|
| 411 | /** |
---|
| 412 | * Returns the tip text for this property. |
---|
| 413 | * @return tip text for this property suitable for |
---|
| 414 | * displaying in the explorer/experimenter gui |
---|
| 415 | */ |
---|
| 416 | public String nearestNeighbourSearchAlgorithmTipText() { |
---|
| 417 | return "The nearest neighbour search algorithm to use " + |
---|
| 418 | "(Default: weka.core.neighboursearch.LinearNNSearch)."; |
---|
| 419 | } |
---|
| 420 | |
---|
| 421 | /** |
---|
| 422 | * Returns the current nearestNeighbourSearch algorithm in use. |
---|
| 423 | * @return the NearestNeighbourSearch algorithm currently in use. |
---|
| 424 | */ |
---|
| 425 | public NearestNeighbourSearch getNearestNeighbourSearchAlgorithm() { |
---|
| 426 | return m_NNSearch; |
---|
| 427 | } |
---|
| 428 | |
---|
| 429 | /** |
---|
| 430 | * Sets the nearestNeighbourSearch algorithm to be used for finding nearest |
---|
| 431 | * neighbour(s). |
---|
| 432 | * @param nearestNeighbourSearchAlgorithm - The NearestNeighbourSearch class. |
---|
| 433 | */ |
---|
| 434 | public void setNearestNeighbourSearchAlgorithm(NearestNeighbourSearch nearestNeighbourSearchAlgorithm) { |
---|
| 435 | m_NNSearch = nearestNeighbourSearchAlgorithm; |
---|
| 436 | } |
---|
| 437 | |
---|
| 438 | /** |
---|
| 439 | * Get the number of training instances the classifier is currently using. |
---|
| 440 | * |
---|
| 441 | * @return the number of training instances the classifier is currently using |
---|
| 442 | */ |
---|
| 443 | public int getNumTraining() { |
---|
| 444 | |
---|
| 445 | return m_Train.numInstances(); |
---|
| 446 | } |
---|
| 447 | |
---|
| 448 | /** |
---|
| 449 | * Returns default capabilities of the classifier. |
---|
| 450 | * |
---|
| 451 | * @return the capabilities of this classifier |
---|
| 452 | */ |
---|
| 453 | public Capabilities getCapabilities() { |
---|
| 454 | Capabilities result = super.getCapabilities(); |
---|
| 455 | result.disableAll(); |
---|
| 456 | |
---|
| 457 | // attributes |
---|
| 458 | result.enable(Capability.NOMINAL_ATTRIBUTES); |
---|
| 459 | result.enable(Capability.NUMERIC_ATTRIBUTES); |
---|
| 460 | result.enable(Capability.DATE_ATTRIBUTES); |
---|
| 461 | result.enable(Capability.MISSING_VALUES); |
---|
| 462 | |
---|
| 463 | // class |
---|
| 464 | result.enable(Capability.NOMINAL_CLASS); |
---|
| 465 | result.enable(Capability.NUMERIC_CLASS); |
---|
| 466 | result.enable(Capability.DATE_CLASS); |
---|
| 467 | result.enable(Capability.MISSING_CLASS_VALUES); |
---|
| 468 | |
---|
| 469 | // instances |
---|
| 470 | result.setMinimumNumberInstances(0); |
---|
| 471 | |
---|
| 472 | return result; |
---|
| 473 | } |
---|
| 474 | |
---|
| 475 | /** |
---|
| 476 | * Generates the classifier. |
---|
| 477 | * |
---|
| 478 | * @param instances set of instances serving as training data |
---|
| 479 | * @throws Exception if the classifier has not been generated successfully |
---|
| 480 | */ |
---|
| 481 | public void buildClassifier(Instances instances) throws Exception { |
---|
| 482 | |
---|
| 483 | // can classifier handle the data? |
---|
| 484 | getCapabilities().testWithFail(instances); |
---|
| 485 | |
---|
| 486 | // remove instances with missing class |
---|
| 487 | instances = new Instances(instances); |
---|
| 488 | instances.deleteWithMissingClass(); |
---|
| 489 | |
---|
| 490 | m_NumClasses = instances.numClasses(); |
---|
| 491 | m_ClassType = instances.classAttribute().type(); |
---|
| 492 | m_Train = new Instances(instances, 0, instances.numInstances()); |
---|
| 493 | |
---|
| 494 | // Throw away initial instances until within the specified window size |
---|
| 495 | if ((m_WindowSize > 0) && (instances.numInstances() > m_WindowSize)) { |
---|
| 496 | m_Train = new Instances(m_Train, |
---|
| 497 | m_Train.numInstances()-m_WindowSize, |
---|
| 498 | m_WindowSize); |
---|
| 499 | } |
---|
| 500 | |
---|
| 501 | m_NumAttributesUsed = 0.0; |
---|
| 502 | for (int i = 0; i < m_Train.numAttributes(); i++) { |
---|
| 503 | if ((i != m_Train.classIndex()) && |
---|
| 504 | (m_Train.attribute(i).isNominal() || |
---|
| 505 | m_Train.attribute(i).isNumeric())) { |
---|
| 506 | m_NumAttributesUsed += 1.0; |
---|
| 507 | } |
---|
| 508 | } |
---|
| 509 | |
---|
| 510 | m_NNSearch.setInstances(m_Train); |
---|
| 511 | |
---|
| 512 | // Invalidate any currently cross-validation selected k |
---|
| 513 | m_kNNValid = false; |
---|
| 514 | } |
---|
| 515 | |
---|
| 516 | /** |
---|
| 517 | * Adds the supplied instance to the training set. |
---|
| 518 | * |
---|
| 519 | * @param instance the instance to add |
---|
| 520 | * @throws Exception if instance could not be incorporated |
---|
| 521 | * successfully |
---|
| 522 | */ |
---|
| 523 | public void updateClassifier(Instance instance) throws Exception { |
---|
| 524 | |
---|
| 525 | if (m_Train.equalHeaders(instance.dataset()) == false) { |
---|
| 526 | throw new Exception("Incompatible instance types\n" + m_Train.equalHeadersMsg(instance.dataset())); |
---|
| 527 | } |
---|
| 528 | if (instance.classIsMissing()) { |
---|
| 529 | return; |
---|
| 530 | } |
---|
| 531 | |
---|
| 532 | m_Train.add(instance); |
---|
| 533 | m_NNSearch.update(instance); |
---|
| 534 | m_kNNValid = false; |
---|
| 535 | if ((m_WindowSize > 0) && (m_Train.numInstances() > m_WindowSize)) { |
---|
| 536 | boolean deletedInstance=false; |
---|
| 537 | while (m_Train.numInstances() > m_WindowSize) { |
---|
| 538 | m_Train.delete(0); |
---|
| 539 | deletedInstance=true; |
---|
| 540 | } |
---|
| 541 | //rebuild datastructure KDTree currently can't delete |
---|
| 542 | if(deletedInstance==true) |
---|
| 543 | m_NNSearch.setInstances(m_Train); |
---|
| 544 | } |
---|
| 545 | } |
---|
| 546 | |
---|
| 547 | /** |
---|
| 548 | * Calculates the class membership probabilities for the given test instance. |
---|
| 549 | * |
---|
| 550 | * @param instance the instance to be classified |
---|
| 551 | * @return predicted class probability distribution |
---|
| 552 | * @throws Exception if an error occurred during the prediction |
---|
| 553 | */ |
---|
| 554 | public double [] distributionForInstance(Instance instance) throws Exception { |
---|
| 555 | |
---|
| 556 | if (m_Train.numInstances() == 0) { |
---|
| 557 | throw new Exception("No training instances!"); |
---|
| 558 | } |
---|
| 559 | if ((m_WindowSize > 0) && (m_Train.numInstances() > m_WindowSize)) { |
---|
| 560 | m_kNNValid = false; |
---|
| 561 | boolean deletedInstance=false; |
---|
| 562 | while (m_Train.numInstances() > m_WindowSize) { |
---|
| 563 | m_Train.delete(0); |
---|
| 564 | } |
---|
| 565 | //rebuild datastructure KDTree currently can't delete |
---|
| 566 | if(deletedInstance==true) |
---|
| 567 | m_NNSearch.setInstances(m_Train); |
---|
| 568 | } |
---|
| 569 | |
---|
| 570 | // Select k by cross validation |
---|
| 571 | if (!m_kNNValid && (m_CrossValidate) && (m_kNNUpper >= 1)) { |
---|
| 572 | crossValidate(); |
---|
| 573 | } |
---|
| 574 | |
---|
| 575 | m_NNSearch.addInstanceInfo(instance); |
---|
| 576 | |
---|
| 577 | Instances neighbours = m_NNSearch.kNearestNeighbours(instance, m_kNN); |
---|
| 578 | double [] distances = m_NNSearch.getDistances(); |
---|
| 579 | double [] distribution = makeDistribution( neighbours, distances ); |
---|
| 580 | |
---|
| 581 | return distribution; |
---|
| 582 | } |
---|
| 583 | |
---|
| 584 | /** |
---|
| 585 | * Returns an enumeration describing the available options. |
---|
| 586 | * |
---|
| 587 | * @return an enumeration of all the available options. |
---|
| 588 | */ |
---|
| 589 | public Enumeration listOptions() { |
---|
| 590 | |
---|
| 591 | Vector newVector = new Vector(8); |
---|
| 592 | |
---|
| 593 | newVector.addElement(new Option( |
---|
| 594 | "\tWeight neighbours by the inverse of their distance\n"+ |
---|
| 595 | "\t(use when k > 1)", |
---|
| 596 | "I", 0, "-I")); |
---|
| 597 | newVector.addElement(new Option( |
---|
| 598 | "\tWeight neighbours by 1 - their distance\n"+ |
---|
| 599 | "\t(use when k > 1)", |
---|
| 600 | "F", 0, "-F")); |
---|
| 601 | newVector.addElement(new Option( |
---|
| 602 | "\tNumber of nearest neighbours (k) used in classification.\n"+ |
---|
| 603 | "\t(Default = 1)", |
---|
| 604 | "K", 1,"-K <number of neighbors>")); |
---|
| 605 | newVector.addElement(new Option( |
---|
| 606 | "\tMinimise mean squared error rather than mean absolute\n"+ |
---|
| 607 | "\terror when using -X option with numeric prediction.", |
---|
| 608 | "E", 0,"-E")); |
---|
| 609 | newVector.addElement(new Option( |
---|
| 610 | "\tMaximum number of training instances maintained.\n"+ |
---|
| 611 | "\tTraining instances are dropped FIFO. (Default = no window)", |
---|
| 612 | "W", 1,"-W <window size>")); |
---|
| 613 | newVector.addElement(new Option( |
---|
| 614 | "\tSelect the number of nearest neighbours between 1\n"+ |
---|
| 615 | "\tand the k value specified using hold-one-out evaluation\n"+ |
---|
| 616 | "\ton the training data (use when k > 1)", |
---|
| 617 | "X", 0,"-X")); |
---|
| 618 | newVector.addElement(new Option( |
---|
| 619 | "\tThe nearest neighbour search algorithm to use "+ |
---|
| 620 | "(default: weka.core.neighboursearch.LinearNNSearch).\n", |
---|
| 621 | "A", 0, "-A")); |
---|
| 622 | |
---|
| 623 | return newVector.elements(); |
---|
| 624 | } |
---|
| 625 | |
---|
| 626 | /** |
---|
| 627 | * Parses a given list of options. <p/> |
---|
| 628 | * |
---|
| 629 | <!-- options-start --> |
---|
| 630 | * Valid options are: <p/> |
---|
| 631 | * |
---|
| 632 | * <pre> -I |
---|
| 633 | * Weight neighbours by the inverse of their distance |
---|
| 634 | * (use when k > 1)</pre> |
---|
| 635 | * |
---|
| 636 | * <pre> -F |
---|
| 637 | * Weight neighbours by 1 - their distance |
---|
| 638 | * (use when k > 1)</pre> |
---|
| 639 | * |
---|
| 640 | * <pre> -K <number of neighbors> |
---|
| 641 | * Number of nearest neighbours (k) used in classification. |
---|
| 642 | * (Default = 1)</pre> |
---|
| 643 | * |
---|
| 644 | * <pre> -E |
---|
| 645 | * Minimise mean squared error rather than mean absolute |
---|
| 646 | * error when using -X option with numeric prediction.</pre> |
---|
| 647 | * |
---|
| 648 | * <pre> -W <window size> |
---|
| 649 | * Maximum number of training instances maintained. |
---|
| 650 | * Training instances are dropped FIFO. (Default = no window)</pre> |
---|
| 651 | * |
---|
| 652 | * <pre> -X |
---|
| 653 | * Select the number of nearest neighbours between 1 |
---|
| 654 | * and the k value specified using hold-one-out evaluation |
---|
| 655 | * on the training data (use when k > 1)</pre> |
---|
| 656 | * |
---|
| 657 | * <pre> -A |
---|
| 658 | * The nearest neighbour search algorithm to use (default: weka.core.neighboursearch.LinearNNSearch). |
---|
| 659 | * </pre> |
---|
| 660 | * |
---|
| 661 | <!-- options-end --> |
---|
| 662 | * |
---|
| 663 | * @param options the list of options as an array of strings |
---|
| 664 | * @throws Exception if an option is not supported |
---|
| 665 | */ |
---|
| 666 | public void setOptions(String[] options) throws Exception { |
---|
| 667 | |
---|
| 668 | String knnString = Utils.getOption('K', options); |
---|
| 669 | if (knnString.length() != 0) { |
---|
| 670 | setKNN(Integer.parseInt(knnString)); |
---|
| 671 | } else { |
---|
| 672 | setKNN(1); |
---|
| 673 | } |
---|
| 674 | String windowString = Utils.getOption('W', options); |
---|
| 675 | if (windowString.length() != 0) { |
---|
| 676 | setWindowSize(Integer.parseInt(windowString)); |
---|
| 677 | } else { |
---|
| 678 | setWindowSize(0); |
---|
| 679 | } |
---|
| 680 | if (Utils.getFlag('I', options)) { |
---|
| 681 | setDistanceWeighting(new SelectedTag(WEIGHT_INVERSE, TAGS_WEIGHTING)); |
---|
| 682 | } else if (Utils.getFlag('F', options)) { |
---|
| 683 | setDistanceWeighting(new SelectedTag(WEIGHT_SIMILARITY, TAGS_WEIGHTING)); |
---|
| 684 | } else { |
---|
| 685 | setDistanceWeighting(new SelectedTag(WEIGHT_NONE, TAGS_WEIGHTING)); |
---|
| 686 | } |
---|
| 687 | setCrossValidate(Utils.getFlag('X', options)); |
---|
| 688 | setMeanSquared(Utils.getFlag('E', options)); |
---|
| 689 | |
---|
| 690 | String nnSearchClass = Utils.getOption('A', options); |
---|
| 691 | if(nnSearchClass.length() != 0) { |
---|
| 692 | String nnSearchClassSpec[] = Utils.splitOptions(nnSearchClass); |
---|
| 693 | if(nnSearchClassSpec.length == 0) { |
---|
| 694 | throw new Exception("Invalid NearestNeighbourSearch algorithm " + |
---|
| 695 | "specification string."); |
---|
| 696 | } |
---|
| 697 | String className = nnSearchClassSpec[0]; |
---|
| 698 | nnSearchClassSpec[0] = ""; |
---|
| 699 | |
---|
| 700 | setNearestNeighbourSearchAlgorithm( (NearestNeighbourSearch) |
---|
| 701 | Utils.forName( NearestNeighbourSearch.class, |
---|
| 702 | className, |
---|
| 703 | nnSearchClassSpec) |
---|
| 704 | ); |
---|
| 705 | } |
---|
| 706 | else |
---|
| 707 | this.setNearestNeighbourSearchAlgorithm(new LinearNNSearch()); |
---|
| 708 | |
---|
| 709 | Utils.checkForRemainingOptions(options); |
---|
| 710 | } |
---|
| 711 | |
---|
| 712 | /** |
---|
| 713 | * Gets the current settings of IBk. |
---|
| 714 | * |
---|
| 715 | * @return an array of strings suitable for passing to setOptions() |
---|
| 716 | */ |
---|
| 717 | public String [] getOptions() { |
---|
| 718 | |
---|
| 719 | String [] options = new String [11]; |
---|
| 720 | int current = 0; |
---|
| 721 | options[current++] = "-K"; options[current++] = "" + getKNN(); |
---|
| 722 | options[current++] = "-W"; options[current++] = "" + m_WindowSize; |
---|
| 723 | if (getCrossValidate()) { |
---|
| 724 | options[current++] = "-X"; |
---|
| 725 | } |
---|
| 726 | if (getMeanSquared()) { |
---|
| 727 | options[current++] = "-E"; |
---|
| 728 | } |
---|
| 729 | if (m_DistanceWeighting == WEIGHT_INVERSE) { |
---|
| 730 | options[current++] = "-I"; |
---|
| 731 | } else if (m_DistanceWeighting == WEIGHT_SIMILARITY) { |
---|
| 732 | options[current++] = "-F"; |
---|
| 733 | } |
---|
| 734 | |
---|
| 735 | options[current++] = "-A"; |
---|
| 736 | options[current++] = m_NNSearch.getClass().getName()+" "+Utils.joinOptions(m_NNSearch.getOptions()); |
---|
| 737 | |
---|
| 738 | while (current < options.length) { |
---|
| 739 | options[current++] = ""; |
---|
| 740 | } |
---|
| 741 | |
---|
| 742 | return options; |
---|
| 743 | } |
---|
| 744 | |
---|
| 745 | /** |
---|
| 746 | * Returns an enumeration of the additional measure names |
---|
| 747 | * produced by the neighbour search algorithm, plus the chosen K in case |
---|
| 748 | * cross-validation is enabled. |
---|
| 749 | * |
---|
| 750 | * @return an enumeration of the measure names |
---|
| 751 | */ |
---|
| 752 | public Enumeration enumerateMeasures() { |
---|
| 753 | if (m_CrossValidate) { |
---|
| 754 | Enumeration enm = m_NNSearch.enumerateMeasures(); |
---|
| 755 | Vector measures = new Vector(); |
---|
| 756 | while (enm.hasMoreElements()) |
---|
| 757 | measures.add(enm.nextElement()); |
---|
| 758 | measures.add("measureKNN"); |
---|
| 759 | return measures.elements(); |
---|
| 760 | } |
---|
| 761 | else { |
---|
| 762 | return m_NNSearch.enumerateMeasures(); |
---|
| 763 | } |
---|
| 764 | } |
---|
| 765 | |
---|
| 766 | /** |
---|
| 767 | * Returns the value of the named measure from the |
---|
| 768 | * neighbour search algorithm, plus the chosen K in case |
---|
| 769 | * cross-validation is enabled. |
---|
| 770 | * |
---|
| 771 | * @param additionalMeasureName the name of the measure to query for its value |
---|
| 772 | * @return the value of the named measure |
---|
| 773 | * @throws IllegalArgumentException if the named measure is not supported |
---|
| 774 | */ |
---|
| 775 | public double getMeasure(String additionalMeasureName) { |
---|
| 776 | if (additionalMeasureName.equals("measureKNN")) |
---|
| 777 | return m_kNN; |
---|
| 778 | else |
---|
| 779 | return m_NNSearch.getMeasure(additionalMeasureName); |
---|
| 780 | } |
---|
| 781 | |
---|
| 782 | |
---|
| 783 | /** |
---|
| 784 | * Returns a description of this classifier. |
---|
| 785 | * |
---|
| 786 | * @return a description of this classifier as a string. |
---|
| 787 | */ |
---|
| 788 | public String toString() { |
---|
| 789 | |
---|
| 790 | if (m_Train == null) { |
---|
| 791 | return "IBk: No model built yet."; |
---|
| 792 | } |
---|
| 793 | |
---|
| 794 | if (!m_kNNValid && m_CrossValidate) { |
---|
| 795 | crossValidate(); |
---|
| 796 | } |
---|
| 797 | |
---|
| 798 | String result = "IB1 instance-based classifier\n" + |
---|
| 799 | "using " + m_kNN; |
---|
| 800 | |
---|
| 801 | switch (m_DistanceWeighting) { |
---|
| 802 | case WEIGHT_INVERSE: |
---|
| 803 | result += " inverse-distance-weighted"; |
---|
| 804 | break; |
---|
| 805 | case WEIGHT_SIMILARITY: |
---|
| 806 | result += " similarity-weighted"; |
---|
| 807 | break; |
---|
| 808 | } |
---|
| 809 | result += " nearest neighbour(s) for classification\n"; |
---|
| 810 | |
---|
| 811 | if (m_WindowSize != 0) { |
---|
| 812 | result += "using a maximum of " |
---|
| 813 | + m_WindowSize + " (windowed) training instances\n"; |
---|
| 814 | } |
---|
| 815 | return result; |
---|
| 816 | } |
---|
| 817 | |
---|
| 818 | /** |
---|
| 819 | * Initialise scheme variables. |
---|
| 820 | */ |
---|
| 821 | protected void init() { |
---|
| 822 | |
---|
| 823 | setKNN(1); |
---|
| 824 | m_WindowSize = 0; |
---|
| 825 | m_DistanceWeighting = WEIGHT_NONE; |
---|
| 826 | m_CrossValidate = false; |
---|
| 827 | m_MeanSquared = false; |
---|
| 828 | } |
---|
| 829 | |
---|
| 830 | /** |
---|
| 831 | * Turn the list of nearest neighbors into a probability distribution. |
---|
| 832 | * |
---|
| 833 | * @param neighbours the list of nearest neighboring instances |
---|
| 834 | * @param distances the distances of the neighbors |
---|
| 835 | * @return the probability distribution |
---|
| 836 | * @throws Exception if computation goes wrong or has no class attribute |
---|
| 837 | */ |
---|
| 838 | protected double [] makeDistribution(Instances neighbours, double[] distances) |
---|
| 839 | throws Exception { |
---|
| 840 | |
---|
| 841 | double total = 0, weight; |
---|
| 842 | double [] distribution = new double [m_NumClasses]; |
---|
| 843 | |
---|
| 844 | // Set up a correction to the estimator |
---|
| 845 | if (m_ClassType == Attribute.NOMINAL) { |
---|
| 846 | for(int i = 0; i < m_NumClasses; i++) { |
---|
| 847 | distribution[i] = 1.0 / Math.max(1,m_Train.numInstances()); |
---|
| 848 | } |
---|
| 849 | total = (double)m_NumClasses / Math.max(1,m_Train.numInstances()); |
---|
| 850 | } |
---|
| 851 | |
---|
| 852 | for(int i=0; i < neighbours.numInstances(); i++) { |
---|
| 853 | // Collect class counts |
---|
| 854 | Instance current = neighbours.instance(i); |
---|
| 855 | distances[i] = distances[i]*distances[i]; |
---|
| 856 | distances[i] = Math.sqrt(distances[i]/m_NumAttributesUsed); |
---|
| 857 | switch (m_DistanceWeighting) { |
---|
| 858 | case WEIGHT_INVERSE: |
---|
| 859 | weight = 1.0 / (distances[i] + 0.001); // to avoid div by zero |
---|
| 860 | break; |
---|
| 861 | case WEIGHT_SIMILARITY: |
---|
| 862 | weight = 1.0 - distances[i]; |
---|
| 863 | break; |
---|
| 864 | default: // WEIGHT_NONE: |
---|
| 865 | weight = 1.0; |
---|
| 866 | break; |
---|
| 867 | } |
---|
| 868 | weight *= current.weight(); |
---|
| 869 | try { |
---|
| 870 | switch (m_ClassType) { |
---|
| 871 | case Attribute.NOMINAL: |
---|
| 872 | distribution[(int)current.classValue()] += weight; |
---|
| 873 | break; |
---|
| 874 | case Attribute.NUMERIC: |
---|
| 875 | distribution[0] += current.classValue() * weight; |
---|
| 876 | break; |
---|
| 877 | } |
---|
| 878 | } catch (Exception ex) { |
---|
| 879 | throw new Error("Data has no class attribute!"); |
---|
| 880 | } |
---|
| 881 | total += weight; |
---|
| 882 | } |
---|
| 883 | |
---|
| 884 | // Normalise distribution |
---|
| 885 | if (total > 0) { |
---|
| 886 | Utils.normalize(distribution, total); |
---|
| 887 | } |
---|
| 888 | return distribution; |
---|
| 889 | } |
---|
| 890 | |
---|
| 891 | /** |
---|
| 892 | * Select the best value for k by hold-one-out cross-validation. |
---|
| 893 | * If the class attribute is nominal, classification error is |
---|
| 894 | * minimised. If the class attribute is numeric, mean absolute |
---|
| 895 | * error is minimised |
---|
| 896 | */ |
---|
| 897 | protected void crossValidate() { |
---|
| 898 | |
---|
| 899 | try { |
---|
| 900 | if (m_NNSearch instanceof weka.core.neighboursearch.CoverTree) |
---|
| 901 | throw new Exception("CoverTree doesn't support hold-one-out "+ |
---|
| 902 | "cross-validation. Use some other NN " + |
---|
| 903 | "method."); |
---|
| 904 | |
---|
| 905 | double [] performanceStats = new double [m_kNNUpper]; |
---|
| 906 | double [] performanceStatsSq = new double [m_kNNUpper]; |
---|
| 907 | |
---|
| 908 | for(int i = 0; i < m_kNNUpper; i++) { |
---|
| 909 | performanceStats[i] = 0; |
---|
| 910 | performanceStatsSq[i] = 0; |
---|
| 911 | } |
---|
| 912 | |
---|
| 913 | |
---|
| 914 | m_kNN = m_kNNUpper; |
---|
| 915 | Instance instance; |
---|
| 916 | Instances neighbours; |
---|
| 917 | double[] origDistances, convertedDistances; |
---|
| 918 | for(int i = 0; i < m_Train.numInstances(); i++) { |
---|
| 919 | if (m_Debug && (i % 50 == 0)) { |
---|
| 920 | System.err.print("Cross validating " |
---|
| 921 | + i + "/" + m_Train.numInstances() + "\r"); |
---|
| 922 | } |
---|
| 923 | instance = m_Train.instance(i); |
---|
| 924 | neighbours = m_NNSearch.kNearestNeighbours(instance, m_kNN); |
---|
| 925 | origDistances = m_NNSearch.getDistances(); |
---|
| 926 | |
---|
| 927 | for(int j = m_kNNUpper - 1; j >= 0; j--) { |
---|
| 928 | // Update the performance stats |
---|
| 929 | convertedDistances = new double[origDistances.length]; |
---|
| 930 | System.arraycopy(origDistances, 0, |
---|
| 931 | convertedDistances, 0, origDistances.length); |
---|
| 932 | double [] distribution = makeDistribution(neighbours, |
---|
| 933 | convertedDistances); |
---|
| 934 | double thisPrediction = Utils.maxIndex(distribution); |
---|
| 935 | if (m_Train.classAttribute().isNumeric()) { |
---|
| 936 | thisPrediction = distribution[0]; |
---|
| 937 | double err = thisPrediction - instance.classValue(); |
---|
| 938 | performanceStatsSq[j] += err * err; // Squared error |
---|
| 939 | performanceStats[j] += Math.abs(err); // Absolute error |
---|
| 940 | } else { |
---|
| 941 | if (thisPrediction != instance.classValue()) { |
---|
| 942 | performanceStats[j] ++; // Classification error |
---|
| 943 | } |
---|
| 944 | } |
---|
| 945 | if (j >= 1) { |
---|
| 946 | neighbours = pruneToK(neighbours, convertedDistances, j); |
---|
| 947 | } |
---|
| 948 | } |
---|
| 949 | } |
---|
| 950 | |
---|
| 951 | // Display the results of the cross-validation |
---|
| 952 | for(int i = 0; i < m_kNNUpper; i++) { |
---|
| 953 | if (m_Debug) { |
---|
| 954 | System.err.print("Hold-one-out performance of " + (i + 1) |
---|
| 955 | + " neighbors " ); |
---|
| 956 | } |
---|
| 957 | if (m_Train.classAttribute().isNumeric()) { |
---|
| 958 | if (m_Debug) { |
---|
| 959 | if (m_MeanSquared) { |
---|
| 960 | System.err.println("(RMSE) = " |
---|
| 961 | + Math.sqrt(performanceStatsSq[i] |
---|
| 962 | / m_Train.numInstances())); |
---|
| 963 | } else { |
---|
| 964 | System.err.println("(MAE) = " |
---|
| 965 | + performanceStats[i] |
---|
| 966 | / m_Train.numInstances()); |
---|
| 967 | } |
---|
| 968 | } |
---|
| 969 | } else { |
---|
| 970 | if (m_Debug) { |
---|
| 971 | System.err.println("(%ERR) = " |
---|
| 972 | + 100.0 * performanceStats[i] |
---|
| 973 | / m_Train.numInstances()); |
---|
| 974 | } |
---|
| 975 | } |
---|
| 976 | } |
---|
| 977 | |
---|
| 978 | |
---|
| 979 | // Check through the performance stats and select the best |
---|
| 980 | // k value (or the lowest k if more than one best) |
---|
| 981 | double [] searchStats = performanceStats; |
---|
| 982 | if (m_Train.classAttribute().isNumeric() && m_MeanSquared) { |
---|
| 983 | searchStats = performanceStatsSq; |
---|
| 984 | } |
---|
| 985 | double bestPerformance = Double.NaN; |
---|
| 986 | int bestK = 1; |
---|
| 987 | for(int i = 0; i < m_kNNUpper; i++) { |
---|
| 988 | if (Double.isNaN(bestPerformance) |
---|
| 989 | || (bestPerformance > searchStats[i])) { |
---|
| 990 | bestPerformance = searchStats[i]; |
---|
| 991 | bestK = i + 1; |
---|
| 992 | } |
---|
| 993 | } |
---|
| 994 | m_kNN = bestK; |
---|
| 995 | if (m_Debug) { |
---|
| 996 | System.err.println("Selected k = " + bestK); |
---|
| 997 | } |
---|
| 998 | |
---|
| 999 | m_kNNValid = true; |
---|
| 1000 | } catch (Exception ex) { |
---|
| 1001 | throw new Error("Couldn't optimize by cross-validation: " |
---|
| 1002 | +ex.getMessage()); |
---|
| 1003 | } |
---|
| 1004 | } |
---|
| 1005 | |
---|
| 1006 | /** |
---|
| 1007 | * Prunes the list to contain the k nearest neighbors. If there are |
---|
| 1008 | * multiple neighbors at the k'th distance, all will be kept. |
---|
| 1009 | * |
---|
| 1010 | * @param neighbours the neighbour instances. |
---|
| 1011 | * @param distances the distances of the neighbours from target instance. |
---|
| 1012 | * @param k the number of neighbors to keep. |
---|
| 1013 | * @return the pruned neighbours. |
---|
| 1014 | */ |
---|
| 1015 | public Instances pruneToK(Instances neighbours, double[] distances, int k) { |
---|
| 1016 | |
---|
| 1017 | if(neighbours==null || distances==null || neighbours.numInstances()==0) { |
---|
| 1018 | return null; |
---|
| 1019 | } |
---|
| 1020 | if (k < 1) { |
---|
| 1021 | k = 1; |
---|
| 1022 | } |
---|
| 1023 | |
---|
| 1024 | int currentK = 0; |
---|
| 1025 | double currentDist; |
---|
| 1026 | for(int i=0; i < neighbours.numInstances(); i++) { |
---|
| 1027 | currentK++; |
---|
| 1028 | currentDist = distances[i]; |
---|
| 1029 | if(currentK>k && currentDist!=distances[i-1]) { |
---|
| 1030 | currentK--; |
---|
| 1031 | neighbours = new Instances(neighbours, 0, currentK); |
---|
| 1032 | break; |
---|
| 1033 | } |
---|
| 1034 | } |
---|
| 1035 | |
---|
| 1036 | return neighbours; |
---|
| 1037 | } |
---|
| 1038 | |
---|
| 1039 | /** |
---|
| 1040 | * Returns the revision string. |
---|
| 1041 | * |
---|
| 1042 | * @return the revision |
---|
| 1043 | */ |
---|
| 1044 | public String getRevision() { |
---|
| 1045 | return RevisionUtils.extract("$Revision: 5928 $"); |
---|
| 1046 | } |
---|
| 1047 | |
---|
| 1048 | /** |
---|
| 1049 | * Main method for testing this class. |
---|
| 1050 | * |
---|
| 1051 | * @param argv should contain command line options (see setOptions) |
---|
| 1052 | */ |
---|
| 1053 | public static void main(String [] argv) { |
---|
| 1054 | runClassifier(new IBk(), argv); |
---|
| 1055 | } |
---|
| 1056 | } |
---|