[86] | 1 | // Copyright 2015 Georg-August-Universität Göttingen, Germany |
---|
[41] | 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 | |
---|
[9] | 15 | package de.ugoe.cs.cpdp.training; |
---|
| 16 | |
---|
| 17 | import java.util.ArrayList; |
---|
| 18 | import java.util.HashMap; |
---|
| 19 | import java.util.HashSet; |
---|
| 20 | import java.util.Iterator; |
---|
| 21 | import java.util.Random; |
---|
| 22 | import java.util.Set; |
---|
| 23 | import java.util.logging.Level; |
---|
| 24 | |
---|
[17] | 25 | import de.ugoe.cs.cpdp.training.QuadTree; |
---|
[9] | 26 | import de.ugoe.cs.util.console.Console; |
---|
| 27 | import weka.classifiers.AbstractClassifier; |
---|
| 28 | import weka.classifiers.Classifier; |
---|
| 29 | import weka.core.DenseInstance; |
---|
| 30 | import weka.core.EuclideanDistance; |
---|
| 31 | import weka.core.Instance; |
---|
| 32 | import weka.core.Instances; |
---|
| 33 | import weka.filters.Filter; |
---|
| 34 | import weka.filters.unsupervised.attribute.Remove; |
---|
| 35 | |
---|
| 36 | /** |
---|
[135] | 37 | * <p> |
---|
[41] | 38 | * Trainer with reimplementation of WHERE clustering algorithm from: Tim Menzies, Andrew Butcher, |
---|
| 39 | * David Cok, Andrian Marcus, Lucas Layman, Forrest Shull, Burak Turhan, Thomas Zimmermann, |
---|
| 40 | * "Local versus Global Lessons for Defect Prediction and Effort Estimation," IEEE Transactions on |
---|
| 41 | * Software Engineering, vol. 39, no. 6, pp. 822-834, June, 2013 |
---|
[135] | 42 | * </p> |
---|
| 43 | * <p> |
---|
| 44 | * With WekaLocalFQTraining we do the following: |
---|
| 45 | * <ol> |
---|
| 46 | * <li>Run the Fastmap algorithm on all training data, let it calculate the 2 most significant |
---|
| 47 | * dimensions and projections of each instance to these dimensions</li> |
---|
| 48 | * <li>With these 2 dimensions we span a QuadTree which gets recursively split on median(x) and |
---|
| 49 | * median(y) values.</li> |
---|
| 50 | * <li>We cluster the QuadTree nodes together if they have similar density (50%)</li> |
---|
| 51 | * <li>We save the clusters and their training data</li> |
---|
| 52 | * <li>We only use clusters with > ALPHA instances (currently Math.sqrt(SIZE)), the rest is |
---|
| 53 | * discarded with the training data of this cluster</li> |
---|
| 54 | * <li>We train a Weka classifier for each cluster with the clusters training data</li> |
---|
| 55 | * <li>We recalculate Fastmap distances for a single instance with the old pivots and then try to |
---|
| 56 | * find a cluster containing the coords of the instance. If we can not find a cluster (due to coords |
---|
| 57 | * outside of all clusters) we find the nearest cluster.</li> |
---|
| 58 | * <li>We classify the Instance with the classifier and traindata from the Cluster we found in 7. |
---|
| 59 | * </li> |
---|
| 60 | * </p> |
---|
[9] | 61 | */ |
---|
[23] | 62 | public class WekaLocalFQTraining extends WekaBaseTraining implements ITrainingStrategy { |
---|
[17] | 63 | |
---|
[135] | 64 | /** |
---|
| 65 | * the classifier |
---|
| 66 | */ |
---|
[41] | 67 | private final TraindatasetCluster classifier = new TraindatasetCluster(); |
---|
[17] | 68 | |
---|
[135] | 69 | /* |
---|
| 70 | * (non-Javadoc) |
---|
| 71 | * |
---|
| 72 | * @see de.ugoe.cs.cpdp.training.WekaBaseTraining#getClassifier() |
---|
| 73 | */ |
---|
[41] | 74 | @Override |
---|
| 75 | public Classifier getClassifier() { |
---|
| 76 | return classifier; |
---|
| 77 | } |
---|
[17] | 78 | |
---|
[135] | 79 | /* |
---|
| 80 | * (non-Javadoc) |
---|
| 81 | * |
---|
| 82 | * @see de.ugoe.cs.cpdp.training.ITrainingStrategy#apply(weka.core.Instances) |
---|
| 83 | */ |
---|
[41] | 84 | @Override |
---|
| 85 | public void apply(Instances traindata) { |
---|
| 86 | try { |
---|
| 87 | classifier.buildClassifier(traindata); |
---|
| 88 | } |
---|
| 89 | catch (Exception e) { |
---|
| 90 | throw new RuntimeException(e); |
---|
| 91 | } |
---|
| 92 | } |
---|
[17] | 93 | |
---|
[135] | 94 | /** |
---|
| 95 | * <p> |
---|
| 96 | * Weka classifier for the local model with WHERE clustering |
---|
| 97 | * </p> |
---|
| 98 | * |
---|
| 99 | * @author Alexander Trautsch |
---|
| 100 | */ |
---|
[41] | 101 | public class TraindatasetCluster extends AbstractClassifier { |
---|
[17] | 102 | |
---|
[135] | 103 | /** |
---|
| 104 | * default serialization ID |
---|
| 105 | */ |
---|
[41] | 106 | private static final long serialVersionUID = 1L; |
---|
[17] | 107 | |
---|
[135] | 108 | /** |
---|
| 109 | * classifiers for each cluster |
---|
| 110 | */ |
---|
[41] | 111 | private HashMap<Integer, Classifier> cclassifier; |
---|
[9] | 112 | |
---|
[135] | 113 | /** |
---|
| 114 | * training data for each cluster |
---|
| 115 | */ |
---|
[41] | 116 | private HashMap<Integer, Instances> ctraindata; |
---|
[9] | 117 | |
---|
[135] | 118 | /** |
---|
[41] | 119 | * holds the instances and indices of the pivot objects of the Fastmap calculation in |
---|
| 120 | * buildClassifier |
---|
| 121 | */ |
---|
| 122 | private HashMap<Integer, Instance> cpivots; |
---|
[17] | 123 | |
---|
[135] | 124 | /** |
---|
| 125 | * holds the indices of the pivot objects for x,y and the dimension [x,y][dimension] |
---|
| 126 | */ |
---|
[41] | 127 | private int[][] cpivotindices; |
---|
[14] | 128 | |
---|
[135] | 129 | /** |
---|
| 130 | * holds the sizes of the cluster multiple "boxes" per cluster |
---|
| 131 | */ |
---|
[41] | 132 | private HashMap<Integer, ArrayList<Double[][]>> csize; |
---|
[12] | 133 | |
---|
[135] | 134 | /** |
---|
| 135 | * debug variable |
---|
| 136 | */ |
---|
[41] | 137 | @SuppressWarnings("unused") |
---|
| 138 | private boolean show_biggest = true; |
---|
| 139 | |
---|
[135] | 140 | /** |
---|
| 141 | * debug variable |
---|
| 142 | */ |
---|
[41] | 143 | @SuppressWarnings("unused") |
---|
| 144 | private int CFOUND = 0; |
---|
[135] | 145 | |
---|
| 146 | /** |
---|
| 147 | * debug variable |
---|
| 148 | */ |
---|
[41] | 149 | @SuppressWarnings("unused") |
---|
| 150 | private int CNOTFOUND = 0; |
---|
| 151 | |
---|
[135] | 152 | /** |
---|
| 153 | * <p> |
---|
| 154 | * copies an instance such that is is compatible with the local model |
---|
| 155 | * </p> |
---|
| 156 | * |
---|
| 157 | * @param instances |
---|
| 158 | * instance format |
---|
| 159 | * @param instance |
---|
| 160 | * instance that is copied |
---|
| 161 | * @return |
---|
| 162 | */ |
---|
[41] | 163 | private Instance createInstance(Instances instances, Instance instance) { |
---|
| 164 | // attributes for feeding instance to classifier |
---|
| 165 | Set<String> attributeNames = new HashSet<>(); |
---|
| 166 | for (int j = 0; j < instances.numAttributes(); j++) { |
---|
| 167 | attributeNames.add(instances.attribute(j).name()); |
---|
| 168 | } |
---|
| 169 | |
---|
| 170 | double[] values = new double[instances.numAttributes()]; |
---|
| 171 | int index = 0; |
---|
| 172 | for (int j = 0; j < instance.numAttributes(); j++) { |
---|
| 173 | if (attributeNames.contains(instance.attribute(j).name())) { |
---|
| 174 | values[index] = instance.value(j); |
---|
| 175 | index++; |
---|
| 176 | } |
---|
| 177 | } |
---|
| 178 | |
---|
| 179 | Instances tmp = new Instances(instances); |
---|
| 180 | tmp.clear(); |
---|
| 181 | Instance instCopy = new DenseInstance(instance.weight(), values); |
---|
| 182 | instCopy.setDataset(tmp); |
---|
| 183 | |
---|
| 184 | return instCopy; |
---|
| 185 | } |
---|
| 186 | |
---|
| 187 | /** |
---|
[135] | 188 | * <p> |
---|
[41] | 189 | * Because Fastmap saves only the image not the values of the attributes it used we can not |
---|
| 190 | * use the old data directly to classify single instances to clusters. |
---|
[135] | 191 | * </p> |
---|
| 192 | * <p> |
---|
| 193 | * To classify a single instance we do a new Fastmap computation with only the instance and |
---|
[41] | 194 | * the old pivot elements. |
---|
[135] | 195 | * </p> |
---|
| 196 | * </p> |
---|
| 197 | * After that we find the cluster with our Fastmap result for x and y. |
---|
| 198 | * </p> |
---|
[41] | 199 | * |
---|
[135] | 200 | * @param instance |
---|
| 201 | * instance that is classified |
---|
| 202 | * @see weka.classifiers.AbstractClassifier#classifyInstance(weka.core.Instance) |
---|
[41] | 203 | */ |
---|
| 204 | @Override |
---|
| 205 | public double classifyInstance(Instance instance) { |
---|
| 206 | |
---|
| 207 | double ret = 0; |
---|
| 208 | try { |
---|
| 209 | // classinstance gets passed to classifier |
---|
| 210 | Instances traindata = ctraindata.get(0); |
---|
| 211 | Instance classInstance = createInstance(traindata, instance); |
---|
| 212 | |
---|
| 213 | // this one keeps the class attribute |
---|
| 214 | Instances traindata2 = ctraindata.get(1); |
---|
| 215 | |
---|
| 216 | // remove class attribute before clustering |
---|
| 217 | Remove filter = new Remove(); |
---|
| 218 | filter.setAttributeIndices("" + (traindata.classIndex() + 1)); |
---|
| 219 | filter.setInputFormat(traindata); |
---|
| 220 | traindata = Filter.useFilter(traindata, filter); |
---|
| 221 | Instance clusterInstance = createInstance(traindata, instance); |
---|
| 222 | |
---|
| 223 | Fastmap FMAP = new Fastmap(2); |
---|
| 224 | EuclideanDistance dist = new EuclideanDistance(traindata); |
---|
| 225 | |
---|
| 226 | // we set our pivot indices [x=0,y=1][dimension] |
---|
| 227 | int[][] npivotindices = new int[2][2]; |
---|
| 228 | npivotindices[0][0] = 1; |
---|
| 229 | npivotindices[1][0] = 2; |
---|
| 230 | npivotindices[0][1] = 3; |
---|
| 231 | npivotindices[1][1] = 4; |
---|
| 232 | |
---|
| 233 | // build temp dist matrix (2 pivots per dimension + 1 instance we want to classify) |
---|
| 234 | // the instance we want to classify comes first after that the pivot elements in the |
---|
| 235 | // order defined above |
---|
| 236 | double[][] distmat = new double[2 * FMAP.target_dims + 1][2 * FMAP.target_dims + 1]; |
---|
| 237 | distmat[0][0] = 0; |
---|
[135] | 238 | distmat[0][1] = dist.distance(clusterInstance, |
---|
| 239 | this.cpivots.get((Integer) this.cpivotindices[0][0])); |
---|
| 240 | distmat[0][2] = dist.distance(clusterInstance, |
---|
| 241 | this.cpivots.get((Integer) this.cpivotindices[1][0])); |
---|
| 242 | distmat[0][3] = dist.distance(clusterInstance, |
---|
| 243 | this.cpivots.get((Integer) this.cpivotindices[0][1])); |
---|
| 244 | distmat[0][4] = dist.distance(clusterInstance, |
---|
| 245 | this.cpivots.get((Integer) this.cpivotindices[1][1])); |
---|
[41] | 246 | |
---|
[135] | 247 | distmat[1][0] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][0]), |
---|
| 248 | clusterInstance); |
---|
[41] | 249 | distmat[1][1] = 0; |
---|
[135] | 250 | distmat[1][2] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][0]), |
---|
| 251 | this.cpivots.get((Integer) this.cpivotindices[1][0])); |
---|
| 252 | distmat[1][3] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][0]), |
---|
| 253 | this.cpivots.get((Integer) this.cpivotindices[0][1])); |
---|
| 254 | distmat[1][4] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][0]), |
---|
| 255 | this.cpivots.get((Integer) this.cpivotindices[1][1])); |
---|
[41] | 256 | |
---|
[135] | 257 | distmat[2][0] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][0]), |
---|
| 258 | clusterInstance); |
---|
| 259 | distmat[2][1] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][0]), |
---|
| 260 | this.cpivots.get((Integer) this.cpivotindices[0][0])); |
---|
[41] | 261 | distmat[2][2] = 0; |
---|
[135] | 262 | distmat[2][3] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][0]), |
---|
| 263 | this.cpivots.get((Integer) this.cpivotindices[0][1])); |
---|
| 264 | distmat[2][4] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][0]), |
---|
| 265 | this.cpivots.get((Integer) this.cpivotindices[1][1])); |
---|
[41] | 266 | |
---|
[135] | 267 | distmat[3][0] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][1]), |
---|
| 268 | clusterInstance); |
---|
| 269 | distmat[3][1] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][1]), |
---|
| 270 | this.cpivots.get((Integer) this.cpivotindices[0][0])); |
---|
| 271 | distmat[3][2] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][1]), |
---|
| 272 | this.cpivots.get((Integer) this.cpivotindices[1][0])); |
---|
[41] | 273 | distmat[3][3] = 0; |
---|
[135] | 274 | distmat[3][4] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[0][1]), |
---|
| 275 | this.cpivots.get((Integer) this.cpivotindices[1][1])); |
---|
[41] | 276 | |
---|
[135] | 277 | distmat[4][0] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][1]), |
---|
| 278 | clusterInstance); |
---|
| 279 | distmat[4][1] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][1]), |
---|
| 280 | this.cpivots.get((Integer) this.cpivotindices[0][0])); |
---|
| 281 | distmat[4][2] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][1]), |
---|
| 282 | this.cpivots.get((Integer) this.cpivotindices[1][0])); |
---|
| 283 | distmat[4][3] = dist.distance(this.cpivots.get((Integer) this.cpivotindices[1][1]), |
---|
| 284 | this.cpivots.get((Integer) this.cpivotindices[0][1])); |
---|
[41] | 285 | distmat[4][4] = 0; |
---|
| 286 | |
---|
| 287 | /* |
---|
| 288 | * debug output: show biggest distance found within the new distance matrix double |
---|
| 289 | * biggest = 0; for(int i=0; i < distmat.length; i++) { for(int j=0; j < |
---|
| 290 | * distmat[0].length; j++) { if(biggest < distmat[i][j]) { biggest = distmat[i][j]; |
---|
| 291 | * } } } if(this.show_biggest) { Console.traceln(Level.INFO, |
---|
[135] | 292 | * String.format(""+clusterInstance)); Console.traceln(Level.INFO, String.format( |
---|
| 293 | * "biggest distances: "+ biggest)); this.show_biggest = false; } |
---|
[41] | 294 | */ |
---|
| 295 | |
---|
| 296 | FMAP.setDistmat(distmat); |
---|
| 297 | FMAP.setPivots(npivotindices); |
---|
| 298 | FMAP.calculate(); |
---|
| 299 | double[][] x = FMAP.getX(); |
---|
| 300 | double[] proj = x[0]; |
---|
| 301 | |
---|
| 302 | // debug output: show the calculated distance matrix, our result vektor for the |
---|
| 303 | // instance and the complete result matrix |
---|
| 304 | /* |
---|
| 305 | * Console.traceln(Level.INFO, "distmat:"); for(int i=0; i<distmat.length; i++){ |
---|
| 306 | * for(int j=0; j<distmat[0].length; j++){ Console.trace(Level.INFO, |
---|
| 307 | * String.format("%20s", distmat[i][j])); } Console.traceln(Level.INFO, ""); } |
---|
| 308 | * |
---|
| 309 | * Console.traceln(Level.INFO, "vector:"); for(int i=0; i < proj.length; i++) { |
---|
| 310 | * Console.trace(Level.INFO, String.format("%20s", proj[i])); } |
---|
| 311 | * Console.traceln(Level.INFO, ""); |
---|
| 312 | * |
---|
| 313 | * Console.traceln(Level.INFO, "resultmat:"); for(int i=0; i<x.length; i++){ for(int |
---|
| 314 | * j=0; j<x[0].length; j++){ Console.trace(Level.INFO, String.format("%20s", |
---|
| 315 | * x[i][j])); } Console.traceln(Level.INFO, ""); } |
---|
| 316 | */ |
---|
| 317 | |
---|
| 318 | // now we iterate over all clusters (well, boxes of sizes per cluster really) and |
---|
| 319 | // save the number of the |
---|
| 320 | // cluster in which we are |
---|
| 321 | int cnumber; |
---|
| 322 | int found_cnumber = -1; |
---|
| 323 | Iterator<Integer> clusternumber = this.csize.keySet().iterator(); |
---|
| 324 | while (clusternumber.hasNext() && found_cnumber == -1) { |
---|
| 325 | cnumber = clusternumber.next(); |
---|
| 326 | |
---|
| 327 | // now iterate over the boxes of the cluster and hope we find one (cluster could |
---|
| 328 | // have been removed) |
---|
| 329 | // or we are too far away from any cluster because of the fastmap calculation |
---|
| 330 | // with the initial pivot objects |
---|
| 331 | for (int box = 0; box < this.csize.get(cnumber).size(); box++) { |
---|
| 332 | Double[][] current = this.csize.get(cnumber).get(box); |
---|
| 333 | |
---|
| 334 | if (proj[0] >= current[0][0] && proj[0] <= current[0][1] && // x |
---|
| 335 | proj[1] >= current[1][0] && proj[1] <= current[1][1]) |
---|
| 336 | { // y |
---|
| 337 | found_cnumber = cnumber; |
---|
| 338 | } |
---|
| 339 | } |
---|
| 340 | } |
---|
| 341 | |
---|
| 342 | // we want to count how often we are really inside a cluster |
---|
| 343 | // if ( found_cnumber == -1 ) { |
---|
| 344 | // CNOTFOUND += 1; |
---|
| 345 | // }else { |
---|
| 346 | // CFOUND += 1; |
---|
| 347 | // } |
---|
| 348 | |
---|
| 349 | // now it can happen that we do not find a cluster because we deleted it previously |
---|
| 350 | // (too few instances) |
---|
| 351 | // or we get bigger distance measures from weka so that we are completely outside of |
---|
| 352 | // our clusters. |
---|
| 353 | // in these cases we just find the nearest cluster to our instance and use it for |
---|
| 354 | // classification. |
---|
| 355 | // to do that we use the EuclideanDistance again to compare our distance to all |
---|
| 356 | // other Instances |
---|
| 357 | // then we take the cluster of the closest weka instance |
---|
| 358 | dist = new EuclideanDistance(traindata2); |
---|
| 359 | if (!this.ctraindata.containsKey(found_cnumber)) { |
---|
| 360 | double min_distance = Double.MAX_VALUE; |
---|
| 361 | clusternumber = ctraindata.keySet().iterator(); |
---|
| 362 | while (clusternumber.hasNext()) { |
---|
| 363 | cnumber = clusternumber.next(); |
---|
| 364 | for (int i = 0; i < ctraindata.get(cnumber).size(); i++) { |
---|
[135] | 365 | if (dist.distance(instance, |
---|
| 366 | ctraindata.get(cnumber).get(i)) <= min_distance) |
---|
[41] | 367 | { |
---|
| 368 | found_cnumber = cnumber; |
---|
| 369 | min_distance = |
---|
| 370 | dist.distance(instance, ctraindata.get(cnumber).get(i)); |
---|
| 371 | } |
---|
| 372 | } |
---|
| 373 | } |
---|
| 374 | } |
---|
| 375 | |
---|
| 376 | // here we have the cluster where an instance has the minimum distance between |
---|
| 377 | // itself and the |
---|
| 378 | // instance we want to classify |
---|
| 379 | // if we still have not found a cluster we exit because something is really wrong |
---|
| 380 | if (found_cnumber == -1) { |
---|
| 381 | Console.traceln(Level.INFO, String |
---|
| 382 | .format("ERROR matching instance to cluster with full search!")); |
---|
| 383 | throw new RuntimeException("cluster not found with full search"); |
---|
| 384 | } |
---|
| 385 | |
---|
| 386 | // classify the passed instance with the cluster we found and its training data |
---|
| 387 | ret = cclassifier.get(found_cnumber).classifyInstance(classInstance); |
---|
| 388 | |
---|
| 389 | } |
---|
| 390 | catch (Exception e) { |
---|
| 391 | Console.traceln(Level.INFO, String.format("ERROR matching instance to cluster!")); |
---|
| 392 | throw new RuntimeException(e); |
---|
| 393 | } |
---|
| 394 | return ret; |
---|
| 395 | } |
---|
| 396 | |
---|
[135] | 397 | /* |
---|
| 398 | * (non-Javadoc) |
---|
| 399 | * |
---|
| 400 | * @see weka.classifiers.Classifier#buildClassifier(weka.core.Instances) |
---|
| 401 | */ |
---|
[41] | 402 | @Override |
---|
| 403 | public void buildClassifier(Instances traindata) throws Exception { |
---|
| 404 | |
---|
| 405 | // Console.traceln(Level.INFO, String.format("found: "+ CFOUND + ", notfound: " + |
---|
| 406 | // CNOTFOUND)); |
---|
| 407 | this.show_biggest = true; |
---|
| 408 | |
---|
| 409 | cclassifier = new HashMap<Integer, Classifier>(); |
---|
| 410 | ctraindata = new HashMap<Integer, Instances>(); |
---|
| 411 | cpivots = new HashMap<Integer, Instance>(); |
---|
| 412 | cpivotindices = new int[2][2]; |
---|
| 413 | |
---|
| 414 | // 1. copy traindata |
---|
| 415 | Instances train = new Instances(traindata); |
---|
| 416 | Instances train2 = new Instances(traindata); // this one keeps the class attribute |
---|
| 417 | |
---|
| 418 | // 2. remove class attribute for clustering |
---|
| 419 | Remove filter = new Remove(); |
---|
| 420 | filter.setAttributeIndices("" + (train.classIndex() + 1)); |
---|
| 421 | filter.setInputFormat(train); |
---|
| 422 | train = Filter.useFilter(train, filter); |
---|
| 423 | |
---|
| 424 | // 3. calculate distance matrix (needed for Fastmap because it starts at dimension 1) |
---|
| 425 | double biggest = 0; |
---|
| 426 | EuclideanDistance dist = new EuclideanDistance(train); |
---|
| 427 | double[][] distmat = new double[train.size()][train.size()]; |
---|
| 428 | for (int i = 0; i < train.size(); i++) { |
---|
| 429 | for (int j = 0; j < train.size(); j++) { |
---|
| 430 | distmat[i][j] = dist.distance(train.get(i), train.get(j)); |
---|
| 431 | if (distmat[i][j] > biggest) { |
---|
| 432 | biggest = distmat[i][j]; |
---|
| 433 | } |
---|
| 434 | } |
---|
| 435 | } |
---|
| 436 | // Console.traceln(Level.INFO, String.format("biggest distances: "+ biggest)); |
---|
| 437 | |
---|
| 438 | // 4. run fastmap for 2 dimensions on the distance matrix |
---|
| 439 | Fastmap FMAP = new Fastmap(2); |
---|
| 440 | FMAP.setDistmat(distmat); |
---|
| 441 | FMAP.calculate(); |
---|
| 442 | |
---|
| 443 | cpivotindices = FMAP.getPivots(); |
---|
| 444 | |
---|
| 445 | double[][] X = FMAP.getX(); |
---|
| 446 | |
---|
| 447 | // quadtree payload generation |
---|
| 448 | ArrayList<QuadTreePayload<Instance>> qtp = new ArrayList<QuadTreePayload<Instance>>(); |
---|
| 449 | |
---|
| 450 | // we need these for the sizes of the quadrants |
---|
| 451 | double[] big = |
---|
| 452 | { 0, 0 }; |
---|
| 453 | double[] small = |
---|
| 454 | { Double.MAX_VALUE, Double.MAX_VALUE }; |
---|
| 455 | |
---|
| 456 | // set quadtree payload values and get max and min x and y values for size |
---|
| 457 | for (int i = 0; i < X.length; i++) { |
---|
| 458 | if (X[i][0] >= big[0]) { |
---|
| 459 | big[0] = X[i][0]; |
---|
| 460 | } |
---|
| 461 | if (X[i][1] >= big[1]) { |
---|
| 462 | big[1] = X[i][1]; |
---|
| 463 | } |
---|
| 464 | if (X[i][0] <= small[0]) { |
---|
| 465 | small[0] = X[i][0]; |
---|
| 466 | } |
---|
| 467 | if (X[i][1] <= small[1]) { |
---|
| 468 | small[1] = X[i][1]; |
---|
| 469 | } |
---|
| 470 | QuadTreePayload<Instance> tmp = |
---|
| 471 | new QuadTreePayload<Instance>(X[i][0], X[i][1], train2.get(i)); |
---|
| 472 | qtp.add(tmp); |
---|
| 473 | } |
---|
| 474 | |
---|
| 475 | // Console.traceln(Level.INFO, |
---|
[135] | 476 | // String.format("size for cluster ("+small[0]+","+small[1]+") - |
---|
| 477 | // ("+big[0]+","+big[1]+")")); |
---|
[41] | 478 | |
---|
| 479 | // 5. generate quadtree |
---|
| 480 | QuadTree TREE = new QuadTree(null, qtp); |
---|
| 481 | QuadTree.size = train.size(); |
---|
| 482 | QuadTree.alpha = Math.sqrt(train.size()); |
---|
| 483 | QuadTree.ccluster = new ArrayList<ArrayList<QuadTreePayload<Instance>>>(); |
---|
| 484 | QuadTree.csize = new HashMap<Integer, ArrayList<Double[][]>>(); |
---|
| 485 | |
---|
| 486 | // Console.traceln(Level.INFO, String.format("Generate QuadTree with "+ QuadTree.size + |
---|
| 487 | // " size, Alpha: "+ QuadTree.alpha+ "")); |
---|
| 488 | |
---|
| 489 | // set the size and then split the tree recursively at the median value for x, y |
---|
| 490 | TREE.setSize(new double[] |
---|
| 491 | { small[0], big[0] }, new double[] |
---|
| 492 | { small[1], big[1] }); |
---|
| 493 | |
---|
| 494 | // recursive split und grid clustering eher static |
---|
[135] | 495 | QuadTree.recursiveSplit(TREE); |
---|
[41] | 496 | |
---|
| 497 | // generate list of nodes sorted by density (childs only) |
---|
| 498 | ArrayList<QuadTree> l = new ArrayList<QuadTree>(TREE.getList(TREE)); |
---|
| 499 | |
---|
| 500 | // recursive grid clustering (tree pruning), the values are stored in ccluster |
---|
| 501 | TREE.gridClustering(l); |
---|
| 502 | |
---|
| 503 | // wir iterieren durch die cluster und sammeln uns die instanzen daraus |
---|
| 504 | // ctraindata.clear(); |
---|
| 505 | for (int i = 0; i < QuadTree.ccluster.size(); i++) { |
---|
| 506 | ArrayList<QuadTreePayload<Instance>> current = QuadTree.ccluster.get(i); |
---|
| 507 | |
---|
| 508 | // i is the clusternumber |
---|
| 509 | // we only allow clusters with Instances > ALPHA, other clusters are not considered! |
---|
| 510 | // if(current.size() > QuadTree.alpha) { |
---|
| 511 | if (current.size() > 4) { |
---|
| 512 | for (int j = 0; j < current.size(); j++) { |
---|
| 513 | if (!ctraindata.containsKey(i)) { |
---|
| 514 | ctraindata.put(i, new Instances(train2)); |
---|
| 515 | ctraindata.get(i).delete(); |
---|
| 516 | } |
---|
| 517 | ctraindata.get(i).add(current.get(j).getInst()); |
---|
| 518 | } |
---|
| 519 | } |
---|
| 520 | else { |
---|
[135] | 521 | Console.traceln(Level.INFO, String |
---|
| 522 | .format("drop cluster, only: " + current.size() + " instances")); |
---|
[41] | 523 | } |
---|
| 524 | } |
---|
| 525 | |
---|
| 526 | // here we keep things we need later on |
---|
| 527 | // QuadTree sizes for later use (matching new instances) |
---|
| 528 | this.csize = new HashMap<Integer, ArrayList<Double[][]>>(QuadTree.csize); |
---|
| 529 | |
---|
| 530 | // pivot elements |
---|
| 531 | // this.cpivots.clear(); |
---|
| 532 | for (int i = 0; i < FMAP.PA[0].length; i++) { |
---|
| 533 | this.cpivots.put(FMAP.PA[0][i], (Instance) train.get(FMAP.PA[0][i]).copy()); |
---|
| 534 | } |
---|
| 535 | for (int j = 0; j < FMAP.PA[0].length; j++) { |
---|
| 536 | this.cpivots.put(FMAP.PA[1][j], (Instance) train.get(FMAP.PA[1][j]).copy()); |
---|
| 537 | } |
---|
| 538 | |
---|
| 539 | /* |
---|
| 540 | * debug output int pnumber; Iterator<Integer> pivotnumber = |
---|
| 541 | * cpivots.keySet().iterator(); while ( pivotnumber.hasNext() ) { pnumber = |
---|
| 542 | * pivotnumber.next(); Console.traceln(Level.INFO, String.format("pivot: "+pnumber+ |
---|
| 543 | * " inst: "+cpivots.get(pnumber))); } |
---|
| 544 | */ |
---|
| 545 | |
---|
| 546 | // train one classifier per cluster, we get the cluster number from the traindata |
---|
| 547 | int cnumber; |
---|
| 548 | Iterator<Integer> clusternumber = ctraindata.keySet().iterator(); |
---|
| 549 | // cclassifier.clear(); |
---|
| 550 | |
---|
| 551 | // int traindata_count = 0; |
---|
| 552 | while (clusternumber.hasNext()) { |
---|
| 553 | cnumber = clusternumber.next(); |
---|
| 554 | cclassifier.put(cnumber, setupClassifier()); // this is the classifier used for the |
---|
| 555 | // cluster |
---|
| 556 | cclassifier.get(cnumber).buildClassifier(ctraindata.get(cnumber)); |
---|
| 557 | // Console.traceln(Level.INFO, String.format("classifier in cluster "+cnumber)); |
---|
| 558 | // traindata_count += ctraindata.get(cnumber).size(); |
---|
| 559 | // Console.traceln(Level.INFO, |
---|
[135] | 560 | // String.format("building classifier in cluster "+cnumber +" with "+ |
---|
[41] | 561 | // ctraindata.get(cnumber).size() +" traindata instances")); |
---|
| 562 | } |
---|
| 563 | |
---|
| 564 | // add all traindata |
---|
| 565 | // Console.traceln(Level.INFO, String.format("traindata in all clusters: " + |
---|
| 566 | // traindata_count)); |
---|
| 567 | } |
---|
| 568 | } |
---|
| 569 | |
---|
| 570 | /** |
---|
[135] | 571 | * <p> |
---|
| 572 | * Payload for the QuadTree. x and y are the calculated Fastmap values. T is a Weka instance. |
---|
| 573 | * </p> |
---|
| 574 | * |
---|
| 575 | * @author Alexander Trautsch |
---|
[136] | 576 | * @param <T> type of the instance |
---|
[41] | 577 | */ |
---|
| 578 | public class QuadTreePayload<T> { |
---|
| 579 | |
---|
[135] | 580 | /** |
---|
| 581 | * x-value |
---|
| 582 | */ |
---|
| 583 | public final double x; |
---|
| 584 | |
---|
| 585 | /** |
---|
| 586 | * y-value |
---|
| 587 | */ |
---|
| 588 | public final double y; |
---|
| 589 | |
---|
| 590 | /** |
---|
| 591 | * associated instance |
---|
| 592 | */ |
---|
[41] | 593 | private T inst; |
---|
| 594 | |
---|
[135] | 595 | /** |
---|
| 596 | * <p> |
---|
| 597 | * Constructor. Creates the payload. |
---|
| 598 | * </p> |
---|
| 599 | * |
---|
| 600 | * @param x |
---|
| 601 | * x-value |
---|
| 602 | * @param y |
---|
| 603 | * y-value |
---|
| 604 | * @param value |
---|
| 605 | * associated instace |
---|
| 606 | */ |
---|
[41] | 607 | public QuadTreePayload(double x, double y, T value) { |
---|
| 608 | this.x = x; |
---|
| 609 | this.y = y; |
---|
| 610 | this.inst = value; |
---|
| 611 | } |
---|
| 612 | |
---|
[135] | 613 | /** |
---|
| 614 | * <p> |
---|
| 615 | * returns the instance |
---|
| 616 | * </p> |
---|
| 617 | * |
---|
[136] | 618 | * @return the instance |
---|
[135] | 619 | */ |
---|
[41] | 620 | public T getInst() { |
---|
| 621 | return this.inst; |
---|
| 622 | } |
---|
| 623 | } |
---|
| 624 | |
---|
| 625 | /** |
---|
[135] | 626 | * <p> |
---|
| 627 | * Fastmap implementation after:<br> |
---|
| 628 | * * Faloutsos, C., & Lin, K. I. (1995). FastMap: A fast algorithm for indexing, data-mining and |
---|
[41] | 629 | * visualization of traditional and multimedia datasets (Vol. 24, No. 2, pp. 163-174). ACM. |
---|
[135] | 630 | * </p> |
---|
[41] | 631 | */ |
---|
| 632 | public class Fastmap { |
---|
| 633 | |
---|
[135] | 634 | /** |
---|
| 635 | * N x k Array, at the end, the i-th row will be the image of the i-th object |
---|
| 636 | */ |
---|
[41] | 637 | private double[][] X; |
---|
| 638 | |
---|
[135] | 639 | /** |
---|
| 640 | * 2 x k pivot Array one pair per recursive call |
---|
| 641 | */ |
---|
[41] | 642 | private int[][] PA; |
---|
| 643 | |
---|
[135] | 644 | /** |
---|
| 645 | * Objects we got (distance matrix) |
---|
| 646 | */ |
---|
[41] | 647 | private double[][] O; |
---|
| 648 | |
---|
[135] | 649 | /** |
---|
| 650 | * column of X currently updated (also the dimension) |
---|
| 651 | */ |
---|
[41] | 652 | private int col = 0; |
---|
| 653 | |
---|
[135] | 654 | /** |
---|
| 655 | * number of dimensions we want |
---|
| 656 | */ |
---|
[41] | 657 | private int target_dims = 0; |
---|
| 658 | |
---|
[135] | 659 | /** |
---|
| 660 | * if we already have the pivot elements |
---|
| 661 | */ |
---|
[41] | 662 | private boolean pivot_set = false; |
---|
| 663 | |
---|
[135] | 664 | /** |
---|
| 665 | * <p> |
---|
| 666 | * Constructor. Creates a new Fastmap object. |
---|
| 667 | * </p> |
---|
| 668 | * |
---|
| 669 | * @param k |
---|
| 670 | */ |
---|
[41] | 671 | public Fastmap(int k) { |
---|
| 672 | this.target_dims = k; |
---|
| 673 | } |
---|
| 674 | |
---|
| 675 | /** |
---|
[135] | 676 | * <p> |
---|
| 677 | * Sets the distance matrix and params that depend on this. |
---|
| 678 | * </p> |
---|
[41] | 679 | * |
---|
| 680 | * @param O |
---|
[135] | 681 | * distance matrix |
---|
[41] | 682 | */ |
---|
| 683 | public void setDistmat(double[][] O) { |
---|
| 684 | this.O = O; |
---|
| 685 | int N = O.length; |
---|
| 686 | this.X = new double[N][this.target_dims]; |
---|
| 687 | this.PA = new int[2][this.target_dims]; |
---|
| 688 | } |
---|
| 689 | |
---|
| 690 | /** |
---|
[135] | 691 | * <p> |
---|
[41] | 692 | * Set pivot elements, we need that to classify instances after the calculation is complete |
---|
| 693 | * (because we then want to reuse only the pivot elements). |
---|
[135] | 694 | * </p> |
---|
[41] | 695 | * |
---|
| 696 | * @param pi |
---|
[135] | 697 | * the pivots |
---|
[41] | 698 | */ |
---|
| 699 | public void setPivots(int[][] pi) { |
---|
| 700 | this.pivot_set = true; |
---|
| 701 | this.PA = pi; |
---|
| 702 | } |
---|
| 703 | |
---|
| 704 | /** |
---|
[135] | 705 | * <p> |
---|
[41] | 706 | * Return the pivot elements that were chosen during the calculation |
---|
[135] | 707 | * </p> |
---|
[41] | 708 | * |
---|
[135] | 709 | * @return the pivots |
---|
[41] | 710 | */ |
---|
| 711 | public int[][] getPivots() { |
---|
| 712 | return this.PA; |
---|
| 713 | } |
---|
| 714 | |
---|
| 715 | /** |
---|
[135] | 716 | * <p> |
---|
| 717 | * The distance function for euclidean distance. Acts according to equation 4 of the Fastmap |
---|
| 718 | * paper. |
---|
| 719 | * </p> |
---|
[41] | 720 | * |
---|
| 721 | * @param x |
---|
| 722 | * x index of x image (if k==0 x object) |
---|
| 723 | * @param y |
---|
| 724 | * y index of y image (if k==0 y object) |
---|
[135] | 725 | * @param k |
---|
| 726 | * dimensionality |
---|
| 727 | * @return the distance |
---|
[41] | 728 | */ |
---|
| 729 | private double dist(int x, int y, int k) { |
---|
| 730 | |
---|
| 731 | // basis is object distance, we get this from our distance matrix |
---|
| 732 | double tmp = this.O[x][y] * this.O[x][y]; |
---|
| 733 | |
---|
| 734 | // decrease by projections |
---|
| 735 | for (int i = 0; i < k; i++) { |
---|
| 736 | double tmp2 = (this.X[x][i] - this.X[y][i]); |
---|
| 737 | tmp -= tmp2 * tmp2; |
---|
| 738 | } |
---|
| 739 | |
---|
| 740 | return Math.abs(tmp); |
---|
| 741 | } |
---|
| 742 | |
---|
| 743 | /** |
---|
[135] | 744 | * <p> |
---|
| 745 | * Find the object farthest from the given index. This method is a helper Method for |
---|
| 746 | * findDistandObjects. |
---|
| 747 | * </p> |
---|
[41] | 748 | * |
---|
| 749 | * @param index |
---|
| 750 | * of the object |
---|
| 751 | * @return index of the farthest object from the given index |
---|
| 752 | */ |
---|
| 753 | private int findFarthest(int index) { |
---|
| 754 | double furthest = Double.MIN_VALUE; |
---|
| 755 | int ret = 0; |
---|
| 756 | |
---|
| 757 | for (int i = 0; i < O.length; i++) { |
---|
| 758 | double dist = this.dist(i, index, this.col); |
---|
| 759 | if (i != index && dist > furthest) { |
---|
| 760 | furthest = dist; |
---|
| 761 | ret = i; |
---|
| 762 | } |
---|
| 763 | } |
---|
| 764 | return ret; |
---|
| 765 | } |
---|
| 766 | |
---|
| 767 | /** |
---|
[135] | 768 | * <p> |
---|
| 769 | * Finds the pivot objects. This method is basically algorithm 1 of the Fastmap paper. |
---|
| 770 | * </p> |
---|
[41] | 771 | * |
---|
[135] | 772 | * @return 2 indexes of the chosen pivot objects |
---|
[41] | 773 | */ |
---|
| 774 | private int[] findDistantObjects() { |
---|
| 775 | // 1. choose object randomly |
---|
| 776 | Random r = new Random(); |
---|
| 777 | int obj = r.nextInt(this.O.length); |
---|
| 778 | |
---|
| 779 | // 2. find farthest object from randomly chosen object |
---|
| 780 | int idx1 = this.findFarthest(obj); |
---|
| 781 | |
---|
| 782 | // 3. find farthest object from previously farthest object |
---|
| 783 | int idx2 = this.findFarthest(idx1); |
---|
| 784 | |
---|
| 785 | return new int[] |
---|
| 786 | { idx1, idx2 }; |
---|
| 787 | } |
---|
| 788 | |
---|
| 789 | /** |
---|
[135] | 790 | * <p> |
---|
| 791 | * Calculates the new k-vector values (projections) This is basically algorithm 2 of the |
---|
| 792 | * fastmap paper. We just added the possibility to pre-set the pivot elements because we |
---|
| 793 | * need to classify single instances after the computation is already done. |
---|
| 794 | * </p> |
---|
[41] | 795 | */ |
---|
| 796 | public void calculate() { |
---|
| 797 | |
---|
| 798 | for (int k = 0; k < this.target_dims; k++) { |
---|
| 799 | // 2) choose pivot objects |
---|
| 800 | if (!this.pivot_set) { |
---|
| 801 | int[] pivots = this.findDistantObjects(); |
---|
| 802 | |
---|
| 803 | // 3) record ids of pivot objects |
---|
| 804 | this.PA[0][this.col] = pivots[0]; |
---|
| 805 | this.PA[1][this.col] = pivots[1]; |
---|
| 806 | } |
---|
| 807 | |
---|
| 808 | // 4) inter object distances are zero (this.X is initialized with 0 so we just |
---|
| 809 | // continue) |
---|
| 810 | if (this.dist(this.PA[0][this.col], this.PA[1][this.col], this.col) == 0) { |
---|
| 811 | continue; |
---|
| 812 | } |
---|
| 813 | |
---|
| 814 | // 5) project the objects on the line between the pivots |
---|
| 815 | double dxy = this.dist(this.PA[0][this.col], this.PA[1][this.col], this.col); |
---|
| 816 | for (int i = 0; i < this.O.length; i++) { |
---|
| 817 | |
---|
| 818 | double dix = this.dist(i, this.PA[0][this.col], this.col); |
---|
| 819 | double diy = this.dist(i, this.PA[1][this.col], this.col); |
---|
| 820 | |
---|
| 821 | double tmp = (dix + dxy - diy) / (2 * Math.sqrt(dxy)); |
---|
| 822 | |
---|
| 823 | // save the projection |
---|
| 824 | this.X[i][this.col] = tmp; |
---|
| 825 | } |
---|
| 826 | |
---|
| 827 | this.col += 1; |
---|
| 828 | } |
---|
| 829 | } |
---|
| 830 | |
---|
| 831 | /** |
---|
[135] | 832 | * <p> |
---|
[41] | 833 | * returns the result matrix of the projections |
---|
[135] | 834 | * </p> |
---|
[41] | 835 | * |
---|
| 836 | * @return calculated result |
---|
| 837 | */ |
---|
| 838 | public double[][] getX() { |
---|
| 839 | return this.X; |
---|
| 840 | } |
---|
| 841 | } |
---|
[9] | 842 | } |
---|