| 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 | * LMTNode.java |
|---|
| 19 | * Copyright (C) 2003 University of Waikato, Hamilton, New Zealand |
|---|
| 20 | * |
|---|
| 21 | */ |
|---|
| 22 | |
|---|
| 23 | package weka.classifiers.trees.lmt; |
|---|
| 24 | |
|---|
| 25 | import weka.classifiers.Evaluation; |
|---|
| 26 | import weka.classifiers.functions.SimpleLinearRegression; |
|---|
| 27 | import weka.classifiers.trees.j48.ClassifierSplitModel; |
|---|
| 28 | import weka.classifiers.trees.j48.ModelSelection; |
|---|
| 29 | import weka.core.Instance; |
|---|
| 30 | import weka.core.Instances; |
|---|
| 31 | import weka.core.RevisionHandler; |
|---|
| 32 | import weka.core.RevisionUtils; |
|---|
| 33 | import weka.filters.Filter; |
|---|
| 34 | import weka.filters.supervised.attribute.NominalToBinary; |
|---|
| 35 | |
|---|
| 36 | import java.util.Collections; |
|---|
| 37 | import java.util.Comparator; |
|---|
| 38 | import java.util.Vector; |
|---|
| 39 | |
|---|
| 40 | /** |
|---|
| 41 | * Auxiliary class for list of LMTNodes |
|---|
| 42 | */ |
|---|
| 43 | class CompareNode |
|---|
| 44 | implements Comparator, RevisionHandler { |
|---|
| 45 | |
|---|
| 46 | /** |
|---|
| 47 | * Compares its two arguments for order. |
|---|
| 48 | * |
|---|
| 49 | * @param o1 first object |
|---|
| 50 | * @param o2 second object |
|---|
| 51 | * @return a negative integer, zero, or a positive integer as the first |
|---|
| 52 | * argument is less than, equal to, or greater than the second. |
|---|
| 53 | */ |
|---|
| 54 | public int compare(Object o1, Object o2) { |
|---|
| 55 | if ( ((LMTNode)o1).m_alpha < ((LMTNode)o2).m_alpha) return -1; |
|---|
| 56 | if ( ((LMTNode)o1).m_alpha > ((LMTNode)o2).m_alpha) return 1; |
|---|
| 57 | return 0; |
|---|
| 58 | } |
|---|
| 59 | |
|---|
| 60 | /** |
|---|
| 61 | * Returns the revision string. |
|---|
| 62 | * |
|---|
| 63 | * @return the revision |
|---|
| 64 | */ |
|---|
| 65 | public String getRevision() { |
|---|
| 66 | return RevisionUtils.extract("$Revision: 1.8 $"); |
|---|
| 67 | } |
|---|
| 68 | } |
|---|
| 69 | |
|---|
| 70 | /** |
|---|
| 71 | * Class for logistic model tree structure. |
|---|
| 72 | * |
|---|
| 73 | * |
|---|
| 74 | * @author Niels Landwehr |
|---|
| 75 | * @author Marc Sumner |
|---|
| 76 | * @version $Revision: 1.8 $ |
|---|
| 77 | */ |
|---|
| 78 | public class LMTNode |
|---|
| 79 | extends LogisticBase { |
|---|
| 80 | |
|---|
| 81 | /** for serialization */ |
|---|
| 82 | static final long serialVersionUID = 1862737145870398755L; |
|---|
| 83 | |
|---|
| 84 | /** Total number of training instances. */ |
|---|
| 85 | protected double m_totalInstanceWeight; |
|---|
| 86 | |
|---|
| 87 | /** Node id*/ |
|---|
| 88 | protected int m_id; |
|---|
| 89 | |
|---|
| 90 | /** ID of logistic model at leaf*/ |
|---|
| 91 | protected int m_leafModelNum; |
|---|
| 92 | |
|---|
| 93 | /** Alpha-value (for pruning) at the node*/ |
|---|
| 94 | public double m_alpha; |
|---|
| 95 | |
|---|
| 96 | /** Weighted number of training examples currently misclassified by the logistic model at the node*/ |
|---|
| 97 | public double m_numIncorrectModel; |
|---|
| 98 | |
|---|
| 99 | /** Weighted number of training examples currently misclassified by the subtree rooted at the node*/ |
|---|
| 100 | public double m_numIncorrectTree; |
|---|
| 101 | |
|---|
| 102 | /**minimum number of instances at which a node is considered for splitting*/ |
|---|
| 103 | protected int m_minNumInstances; |
|---|
| 104 | |
|---|
| 105 | /**ModelSelection object (for splitting)*/ |
|---|
| 106 | protected ModelSelection m_modelSelection; |
|---|
| 107 | |
|---|
| 108 | /**Filter to convert nominal attributes to binary*/ |
|---|
| 109 | protected NominalToBinary m_nominalToBinary; |
|---|
| 110 | |
|---|
| 111 | /**Simple regression functions fit by LogitBoost at higher levels in the tree*/ |
|---|
| 112 | protected SimpleLinearRegression[][] m_higherRegressions; |
|---|
| 113 | |
|---|
| 114 | /**Number of simple regression functions fit by LogitBoost at higher levels in the tree*/ |
|---|
| 115 | protected int m_numHigherRegressions = 0; |
|---|
| 116 | |
|---|
| 117 | /**Number of folds for CART pruning*/ |
|---|
| 118 | protected static int m_numFoldsPruning = 5; |
|---|
| 119 | |
|---|
| 120 | /**Use heuristic that determines the number of LogitBoost iterations only once in the beginning? */ |
|---|
| 121 | protected boolean m_fastRegression; |
|---|
| 122 | |
|---|
| 123 | /**Number of instances at the node*/ |
|---|
| 124 | protected int m_numInstances; |
|---|
| 125 | |
|---|
| 126 | /**The ClassifierSplitModel (for splitting)*/ |
|---|
| 127 | protected ClassifierSplitModel m_localModel; |
|---|
| 128 | |
|---|
| 129 | /**Array of children of the node*/ |
|---|
| 130 | protected LMTNode[] m_sons; |
|---|
| 131 | |
|---|
| 132 | /**True if node is leaf*/ |
|---|
| 133 | protected boolean m_isLeaf; |
|---|
| 134 | |
|---|
| 135 | /** |
|---|
| 136 | * Constructor for logistic model tree node. |
|---|
| 137 | * |
|---|
| 138 | * @param modelSelection selection method for local splitting model |
|---|
| 139 | * @param numBoostingIterations sets the numBoostingIterations parameter |
|---|
| 140 | * @param fastRegression sets the fastRegression parameter |
|---|
| 141 | * @param errorOnProbabilities Use error on probabilities for stopping criterion of LogitBoost? |
|---|
| 142 | * @param minNumInstances minimum number of instances at which a node is considered for splitting |
|---|
| 143 | */ |
|---|
| 144 | public LMTNode(ModelSelection modelSelection, int numBoostingIterations, |
|---|
| 145 | boolean fastRegression, |
|---|
| 146 | boolean errorOnProbabilities, int minNumInstances, |
|---|
| 147 | double weightTrimBeta, boolean useAIC) { |
|---|
| 148 | m_modelSelection = modelSelection; |
|---|
| 149 | m_fixedNumIterations = numBoostingIterations; |
|---|
| 150 | m_fastRegression = fastRegression; |
|---|
| 151 | m_errorOnProbabilities = errorOnProbabilities; |
|---|
| 152 | m_minNumInstances = minNumInstances; |
|---|
| 153 | m_maxIterations = 200; |
|---|
| 154 | setWeightTrimBeta(weightTrimBeta); |
|---|
| 155 | setUseAIC(useAIC); |
|---|
| 156 | } |
|---|
| 157 | |
|---|
| 158 | /** |
|---|
| 159 | * Method for building a logistic model tree (only called for the root node). |
|---|
| 160 | * Grows an initial logistic model tree and prunes it back using the CART pruning scheme. |
|---|
| 161 | * |
|---|
| 162 | * @param data the data to train with |
|---|
| 163 | * @throws Exception if something goes wrong |
|---|
| 164 | */ |
|---|
| 165 | public void buildClassifier(Instances data) throws Exception{ |
|---|
| 166 | |
|---|
| 167 | //heuristic to avoid cross-validating the number of LogitBoost iterations |
|---|
| 168 | //at every node: build standalone logistic model and take its optimum number |
|---|
| 169 | //of iteration everywhere in the tree. |
|---|
| 170 | if (m_fastRegression && (m_fixedNumIterations < 0)) m_fixedNumIterations = tryLogistic(data); |
|---|
| 171 | |
|---|
| 172 | //Need to cross-validate alpha-parameter for CART-pruning |
|---|
| 173 | Instances cvData = new Instances(data); |
|---|
| 174 | cvData.stratify(m_numFoldsPruning); |
|---|
| 175 | |
|---|
| 176 | double[][] alphas = new double[m_numFoldsPruning][]; |
|---|
| 177 | double[][] errors = new double[m_numFoldsPruning][]; |
|---|
| 178 | |
|---|
| 179 | for (int i = 0; i < m_numFoldsPruning; i++) { |
|---|
| 180 | //for every fold, grow tree on training set... |
|---|
| 181 | Instances train = cvData.trainCV(m_numFoldsPruning, i); |
|---|
| 182 | Instances test = cvData.testCV(m_numFoldsPruning, i); |
|---|
| 183 | |
|---|
| 184 | buildTree(train, null, train.numInstances() , 0); |
|---|
| 185 | |
|---|
| 186 | int numNodes = getNumInnerNodes(); |
|---|
| 187 | alphas[i] = new double[numNodes + 2]; |
|---|
| 188 | errors[i] = new double[numNodes + 2]; |
|---|
| 189 | |
|---|
| 190 | //... then prune back and log alpha-values and errors on test set |
|---|
| 191 | prune(alphas[i], errors[i], test); |
|---|
| 192 | } |
|---|
| 193 | |
|---|
| 194 | //build tree using all the data |
|---|
| 195 | buildTree(data, null, data.numInstances(), 0); |
|---|
| 196 | int numNodes = getNumInnerNodes(); |
|---|
| 197 | |
|---|
| 198 | double[] treeAlphas = new double[numNodes + 2]; |
|---|
| 199 | |
|---|
| 200 | //prune back and log alpha-values |
|---|
| 201 | int iterations = prune(treeAlphas, null, null); |
|---|
| 202 | |
|---|
| 203 | double[] treeErrors = new double[numNodes + 2]; |
|---|
| 204 | |
|---|
| 205 | for (int i = 0; i <= iterations; i++){ |
|---|
| 206 | //compute midpoint alphas |
|---|
| 207 | double alpha = Math.sqrt(treeAlphas[i] * treeAlphas[i+1]); |
|---|
| 208 | double error = 0; |
|---|
| 209 | |
|---|
| 210 | //compute error estimate for final trees from the midpoint-alphas and the error estimates gotten in |
|---|
| 211 | //the cross-validation |
|---|
| 212 | for (int k = 0; k < m_numFoldsPruning; k++) { |
|---|
| 213 | int l = 0; |
|---|
| 214 | while (alphas[k][l] <= alpha) l++; |
|---|
| 215 | error += errors[k][l - 1]; |
|---|
| 216 | } |
|---|
| 217 | |
|---|
| 218 | treeErrors[i] = error; |
|---|
| 219 | } |
|---|
| 220 | |
|---|
| 221 | //find best alpha |
|---|
| 222 | int best = -1; |
|---|
| 223 | double bestError = Double.MAX_VALUE; |
|---|
| 224 | for (int i = iterations; i >= 0; i--) { |
|---|
| 225 | if (treeErrors[i] < bestError) { |
|---|
| 226 | bestError = treeErrors[i]; |
|---|
| 227 | best = i; |
|---|
| 228 | } |
|---|
| 229 | } |
|---|
| 230 | |
|---|
| 231 | double bestAlpha = Math.sqrt(treeAlphas[best] * treeAlphas[best + 1]); |
|---|
| 232 | |
|---|
| 233 | //"unprune" final tree (faster than regrowing it) |
|---|
| 234 | unprune(); |
|---|
| 235 | |
|---|
| 236 | //CART-prune it with best alpha |
|---|
| 237 | prune(bestAlpha); |
|---|
| 238 | cleanup(); |
|---|
| 239 | } |
|---|
| 240 | |
|---|
| 241 | /** |
|---|
| 242 | * Method for building the tree structure. |
|---|
| 243 | * Builds a logistic model, splits the node and recursively builds tree for child nodes. |
|---|
| 244 | * @param data the training data passed on to this node |
|---|
| 245 | * @param higherRegressions An array of regression functions produced by LogitBoost at higher |
|---|
| 246 | * levels in the tree. They represent a logistic regression model that is refined locally |
|---|
| 247 | * at this node. |
|---|
| 248 | * @param totalInstanceWeight the total number of training examples |
|---|
| 249 | * @param higherNumParameters effective number of parameters in the logistic regression model built |
|---|
| 250 | * in parent nodes |
|---|
| 251 | * @throws Exception if something goes wrong |
|---|
| 252 | */ |
|---|
| 253 | public void buildTree(Instances data, SimpleLinearRegression[][] higherRegressions, |
|---|
| 254 | double totalInstanceWeight, double higherNumParameters) throws Exception{ |
|---|
| 255 | |
|---|
| 256 | //save some stuff |
|---|
| 257 | m_totalInstanceWeight = totalInstanceWeight; |
|---|
| 258 | m_train = new Instances(data); |
|---|
| 259 | |
|---|
| 260 | m_isLeaf = true; |
|---|
| 261 | m_sons = null; |
|---|
| 262 | |
|---|
| 263 | m_numInstances = m_train.numInstances(); |
|---|
| 264 | m_numClasses = m_train.numClasses(); |
|---|
| 265 | |
|---|
| 266 | //init |
|---|
| 267 | m_numericData = getNumericData(m_train); |
|---|
| 268 | m_numericDataHeader = new Instances(m_numericData, 0); |
|---|
| 269 | |
|---|
| 270 | m_regressions = initRegressions(); |
|---|
| 271 | m_numRegressions = 0; |
|---|
| 272 | |
|---|
| 273 | if (higherRegressions != null) m_higherRegressions = higherRegressions; |
|---|
| 274 | else m_higherRegressions = new SimpleLinearRegression[m_numClasses][0]; |
|---|
| 275 | |
|---|
| 276 | m_numHigherRegressions = m_higherRegressions[0].length; |
|---|
| 277 | |
|---|
| 278 | m_numParameters = higherNumParameters; |
|---|
| 279 | |
|---|
| 280 | //build logistic model |
|---|
| 281 | if (m_numInstances >= m_numFoldsBoosting) { |
|---|
| 282 | if (m_fixedNumIterations > 0){ |
|---|
| 283 | performBoosting(m_fixedNumIterations); |
|---|
| 284 | } else if (getUseAIC()) { |
|---|
| 285 | performBoostingInfCriterion(); |
|---|
| 286 | } else { |
|---|
| 287 | performBoostingCV(); |
|---|
| 288 | } |
|---|
| 289 | } |
|---|
| 290 | |
|---|
| 291 | m_numParameters += m_numRegressions; |
|---|
| 292 | |
|---|
| 293 | //only keep the simple regression functions that correspond to the selected number of LogitBoost iterations |
|---|
| 294 | m_regressions = selectRegressions(m_regressions); |
|---|
| 295 | |
|---|
| 296 | boolean grow; |
|---|
| 297 | //split node if more than minNumInstances... |
|---|
| 298 | if (m_numInstances > m_minNumInstances) { |
|---|
| 299 | //split node: either splitting on class value (a la C4.5) or splitting on residuals |
|---|
| 300 | if (m_modelSelection instanceof ResidualModelSelection) { |
|---|
| 301 | //need ps/Ys/Zs/weights |
|---|
| 302 | double[][] probs = getProbs(getFs(m_numericData)); |
|---|
| 303 | double[][] trainYs = getYs(m_train); |
|---|
| 304 | double[][] dataZs = getZs(probs, trainYs); |
|---|
| 305 | double[][] dataWs = getWs(probs, trainYs); |
|---|
| 306 | m_localModel = ((ResidualModelSelection)m_modelSelection).selectModel(m_train, dataZs, dataWs); |
|---|
| 307 | } else { |
|---|
| 308 | m_localModel = m_modelSelection.selectModel(m_train); |
|---|
| 309 | } |
|---|
| 310 | //... and valid split found |
|---|
| 311 | grow = (m_localModel.numSubsets() > 1); |
|---|
| 312 | } else { |
|---|
| 313 | grow = false; |
|---|
| 314 | } |
|---|
| 315 | |
|---|
| 316 | if (grow) { |
|---|
| 317 | //create and build children of node |
|---|
| 318 | m_isLeaf = false; |
|---|
| 319 | Instances[] localInstances = m_localModel.split(m_train); |
|---|
| 320 | m_sons = new LMTNode[m_localModel.numSubsets()]; |
|---|
| 321 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 322 | m_sons[i] = new LMTNode(m_modelSelection, m_fixedNumIterations, |
|---|
| 323 | m_fastRegression, |
|---|
| 324 | m_errorOnProbabilities,m_minNumInstances, |
|---|
| 325 | getWeightTrimBeta(), getUseAIC()); |
|---|
| 326 | //the "higherRegressions" (partial logistic model fit at higher levels in the tree) passed |
|---|
| 327 | //on to the children are the "higherRegressions" at this node plus the regressions added |
|---|
| 328 | //at this node (m_regressions). |
|---|
| 329 | m_sons[i].buildTree(localInstances[i], |
|---|
| 330 | mergeArrays(m_regressions, m_higherRegressions), m_totalInstanceWeight, m_numParameters); |
|---|
| 331 | localInstances[i] = null; |
|---|
| 332 | } |
|---|
| 333 | } |
|---|
| 334 | } |
|---|
| 335 | |
|---|
| 336 | /** |
|---|
| 337 | * Prunes a logistic model tree using the CART pruning scheme, given a |
|---|
| 338 | * cost-complexity parameter alpha. |
|---|
| 339 | * |
|---|
| 340 | * @param alpha the cost-complexity measure |
|---|
| 341 | * @throws Exception if something goes wrong |
|---|
| 342 | */ |
|---|
| 343 | public void prune(double alpha) throws Exception { |
|---|
| 344 | |
|---|
| 345 | Vector nodeList; |
|---|
| 346 | CompareNode comparator = new CompareNode(); |
|---|
| 347 | |
|---|
| 348 | //determine training error of logistic models and subtrees, and calculate alpha-values from them |
|---|
| 349 | modelErrors(); |
|---|
| 350 | treeErrors(); |
|---|
| 351 | calculateAlphas(); |
|---|
| 352 | |
|---|
| 353 | //get list of all inner nodes in the tree |
|---|
| 354 | nodeList = getNodes(); |
|---|
| 355 | |
|---|
| 356 | boolean prune = (nodeList.size() > 0); |
|---|
| 357 | |
|---|
| 358 | while (prune) { |
|---|
| 359 | |
|---|
| 360 | //select node with minimum alpha |
|---|
| 361 | LMTNode nodeToPrune = (LMTNode)Collections.min(nodeList,comparator); |
|---|
| 362 | |
|---|
| 363 | //want to prune if its alpha is smaller than alpha |
|---|
| 364 | if (nodeToPrune.m_alpha > alpha) break; |
|---|
| 365 | |
|---|
| 366 | nodeToPrune.m_isLeaf = true; |
|---|
| 367 | nodeToPrune.m_sons = null; |
|---|
| 368 | |
|---|
| 369 | //update tree errors and alphas |
|---|
| 370 | treeErrors(); |
|---|
| 371 | calculateAlphas(); |
|---|
| 372 | |
|---|
| 373 | nodeList = getNodes(); |
|---|
| 374 | prune = (nodeList.size() > 0); |
|---|
| 375 | } |
|---|
| 376 | } |
|---|
| 377 | |
|---|
| 378 | /** |
|---|
| 379 | * Method for performing one fold in the cross-validation of the cost-complexity parameter. |
|---|
| 380 | * Generates a sequence of alpha-values with error estimates for the corresponding (partially pruned) |
|---|
| 381 | * trees, given the test set of that fold. |
|---|
| 382 | * @param alphas array to hold the generated alpha-values |
|---|
| 383 | * @param errors array to hold the corresponding error estimates |
|---|
| 384 | * @param test test set of that fold (to obtain error estimates) |
|---|
| 385 | * @throws Exception if something goes wrong |
|---|
| 386 | */ |
|---|
| 387 | public int prune(double[] alphas, double[] errors, Instances test) throws Exception { |
|---|
| 388 | |
|---|
| 389 | Vector nodeList; |
|---|
| 390 | |
|---|
| 391 | CompareNode comparator = new CompareNode(); |
|---|
| 392 | |
|---|
| 393 | //determine training error of logistic models and subtrees, and calculate alpha-values from them |
|---|
| 394 | modelErrors(); |
|---|
| 395 | treeErrors(); |
|---|
| 396 | calculateAlphas(); |
|---|
| 397 | |
|---|
| 398 | //get list of all inner nodes in the tree |
|---|
| 399 | nodeList = getNodes(); |
|---|
| 400 | |
|---|
| 401 | boolean prune = (nodeList.size() > 0); |
|---|
| 402 | |
|---|
| 403 | //alpha_0 is always zero (unpruned tree) |
|---|
| 404 | alphas[0] = 0; |
|---|
| 405 | |
|---|
| 406 | Evaluation eval; |
|---|
| 407 | |
|---|
| 408 | //error of unpruned tree |
|---|
| 409 | if (errors != null) { |
|---|
| 410 | eval = new Evaluation(test); |
|---|
| 411 | eval.evaluateModel(this, test); |
|---|
| 412 | errors[0] = eval.errorRate(); |
|---|
| 413 | } |
|---|
| 414 | |
|---|
| 415 | int iteration = 0; |
|---|
| 416 | while (prune) { |
|---|
| 417 | |
|---|
| 418 | iteration++; |
|---|
| 419 | |
|---|
| 420 | //get node with minimum alpha |
|---|
| 421 | LMTNode nodeToPrune = (LMTNode)Collections.min(nodeList,comparator); |
|---|
| 422 | |
|---|
| 423 | nodeToPrune.m_isLeaf = true; |
|---|
| 424 | //Do not set m_sons null, want to unprune |
|---|
| 425 | |
|---|
| 426 | //get alpha-value of node |
|---|
| 427 | alphas[iteration] = nodeToPrune.m_alpha; |
|---|
| 428 | |
|---|
| 429 | //log error |
|---|
| 430 | if (errors != null) { |
|---|
| 431 | eval = new Evaluation(test); |
|---|
| 432 | eval.evaluateModel(this, test); |
|---|
| 433 | errors[iteration] = eval.errorRate(); |
|---|
| 434 | } |
|---|
| 435 | |
|---|
| 436 | //update errors/alphas |
|---|
| 437 | treeErrors(); |
|---|
| 438 | calculateAlphas(); |
|---|
| 439 | |
|---|
| 440 | nodeList = getNodes(); |
|---|
| 441 | prune = (nodeList.size() > 0); |
|---|
| 442 | } |
|---|
| 443 | |
|---|
| 444 | //set last alpha 1 to indicate end |
|---|
| 445 | alphas[iteration + 1] = 1.0; |
|---|
| 446 | return iteration; |
|---|
| 447 | } |
|---|
| 448 | |
|---|
| 449 | |
|---|
| 450 | /** |
|---|
| 451 | *Method to "unprune" a logistic model tree. |
|---|
| 452 | *Sets all leaf-fields to false. |
|---|
| 453 | *Faster than re-growing the tree because the logistic models do not have to be fit again. |
|---|
| 454 | */ |
|---|
| 455 | protected void unprune() { |
|---|
| 456 | if (m_sons != null) { |
|---|
| 457 | m_isLeaf = false; |
|---|
| 458 | for (int i = 0; i < m_sons.length; i++) m_sons[i].unprune(); |
|---|
| 459 | } |
|---|
| 460 | } |
|---|
| 461 | |
|---|
| 462 | /** |
|---|
| 463 | *Determines the optimum number of LogitBoost iterations to perform by building a standalone logistic |
|---|
| 464 | *regression function on the training data. Used for the heuristic that avoids cross-validating this |
|---|
| 465 | *number again at every node. |
|---|
| 466 | *@param data training instances for the logistic model |
|---|
| 467 | *@throws Exception if something goes wrong |
|---|
| 468 | */ |
|---|
| 469 | protected int tryLogistic(Instances data) throws Exception{ |
|---|
| 470 | |
|---|
| 471 | //convert nominal attributes |
|---|
| 472 | Instances filteredData = new Instances(data); |
|---|
| 473 | NominalToBinary nominalToBinary = new NominalToBinary(); |
|---|
| 474 | nominalToBinary.setInputFormat(filteredData); |
|---|
| 475 | filteredData = Filter.useFilter(filteredData, nominalToBinary); |
|---|
| 476 | |
|---|
| 477 | LogisticBase logistic = new LogisticBase(0,true,m_errorOnProbabilities); |
|---|
| 478 | |
|---|
| 479 | //limit LogitBoost to 200 iterations (speed) |
|---|
| 480 | logistic.setMaxIterations(200); |
|---|
| 481 | logistic.setWeightTrimBeta(getWeightTrimBeta()); // Not in Marc's code. Added by Eibe. |
|---|
| 482 | logistic.setUseAIC(getUseAIC()); |
|---|
| 483 | logistic.buildClassifier(filteredData); |
|---|
| 484 | |
|---|
| 485 | //return best number of iterations |
|---|
| 486 | return logistic.getNumRegressions(); |
|---|
| 487 | } |
|---|
| 488 | |
|---|
| 489 | /** |
|---|
| 490 | * Method to count the number of inner nodes in the tree |
|---|
| 491 | * @return the number of inner nodes |
|---|
| 492 | */ |
|---|
| 493 | public int getNumInnerNodes(){ |
|---|
| 494 | if (m_isLeaf) return 0; |
|---|
| 495 | int numNodes = 1; |
|---|
| 496 | for (int i = 0; i < m_sons.length; i++) numNodes += m_sons[i].getNumInnerNodes(); |
|---|
| 497 | return numNodes; |
|---|
| 498 | } |
|---|
| 499 | |
|---|
| 500 | /** |
|---|
| 501 | * Returns the number of leaves in the tree. |
|---|
| 502 | * Leaves are only counted if their logistic model has changed compared to the one of the parent node. |
|---|
| 503 | * @return the number of leaves |
|---|
| 504 | */ |
|---|
| 505 | public int getNumLeaves(){ |
|---|
| 506 | int numLeaves; |
|---|
| 507 | if (!m_isLeaf) { |
|---|
| 508 | numLeaves = 0; |
|---|
| 509 | int numEmptyLeaves = 0; |
|---|
| 510 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 511 | numLeaves += m_sons[i].getNumLeaves(); |
|---|
| 512 | if (m_sons[i].m_isLeaf && !m_sons[i].hasModels()) numEmptyLeaves++; |
|---|
| 513 | } |
|---|
| 514 | if (numEmptyLeaves > 1) { |
|---|
| 515 | numLeaves -= (numEmptyLeaves - 1); |
|---|
| 516 | } |
|---|
| 517 | } else { |
|---|
| 518 | numLeaves = 1; |
|---|
| 519 | } |
|---|
| 520 | return numLeaves; |
|---|
| 521 | } |
|---|
| 522 | |
|---|
| 523 | /** |
|---|
| 524 | *Updates the numIncorrectModel field for all nodes. This is needed for calculating the alpha-values. |
|---|
| 525 | */ |
|---|
| 526 | public void modelErrors() throws Exception{ |
|---|
| 527 | |
|---|
| 528 | Evaluation eval = new Evaluation(m_train); |
|---|
| 529 | |
|---|
| 530 | if (!m_isLeaf) { |
|---|
| 531 | m_isLeaf = true; |
|---|
| 532 | eval.evaluateModel(this, m_train); |
|---|
| 533 | m_isLeaf = false; |
|---|
| 534 | m_numIncorrectModel = eval.incorrect(); |
|---|
| 535 | for (int i = 0; i < m_sons.length; i++) m_sons[i].modelErrors(); |
|---|
| 536 | } else { |
|---|
| 537 | eval.evaluateModel(this, m_train); |
|---|
| 538 | m_numIncorrectModel = eval.incorrect(); |
|---|
| 539 | } |
|---|
| 540 | } |
|---|
| 541 | |
|---|
| 542 | /** |
|---|
| 543 | *Updates the numIncorrectTree field for all nodes. This is needed for calculating the alpha-values. |
|---|
| 544 | */ |
|---|
| 545 | public void treeErrors(){ |
|---|
| 546 | if (m_isLeaf) { |
|---|
| 547 | m_numIncorrectTree = m_numIncorrectModel; |
|---|
| 548 | } else { |
|---|
| 549 | m_numIncorrectTree = 0; |
|---|
| 550 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 551 | m_sons[i].treeErrors(); |
|---|
| 552 | m_numIncorrectTree += m_sons[i].m_numIncorrectTree; |
|---|
| 553 | } |
|---|
| 554 | } |
|---|
| 555 | } |
|---|
| 556 | |
|---|
| 557 | /** |
|---|
| 558 | *Updates the alpha field for all nodes. |
|---|
| 559 | */ |
|---|
| 560 | public void calculateAlphas() throws Exception { |
|---|
| 561 | |
|---|
| 562 | if (!m_isLeaf) { |
|---|
| 563 | double errorDiff = m_numIncorrectModel - m_numIncorrectTree; |
|---|
| 564 | |
|---|
| 565 | if (errorDiff <= 0) { |
|---|
| 566 | //split increases training error (should not normally happen). |
|---|
| 567 | //prune it instantly. |
|---|
| 568 | m_isLeaf = true; |
|---|
| 569 | m_sons = null; |
|---|
| 570 | m_alpha = Double.MAX_VALUE; |
|---|
| 571 | } else { |
|---|
| 572 | //compute alpha |
|---|
| 573 | errorDiff /= m_totalInstanceWeight; |
|---|
| 574 | m_alpha = errorDiff / (double)(getNumLeaves() - 1); |
|---|
| 575 | |
|---|
| 576 | for (int i = 0; i < m_sons.length; i++) m_sons[i].calculateAlphas(); |
|---|
| 577 | } |
|---|
| 578 | } else { |
|---|
| 579 | //alpha = infinite for leaves (do not want to prune) |
|---|
| 580 | m_alpha = Double.MAX_VALUE; |
|---|
| 581 | } |
|---|
| 582 | } |
|---|
| 583 | |
|---|
| 584 | /** |
|---|
| 585 | * Merges two arrays of regression functions into one |
|---|
| 586 | * @param a1 one array |
|---|
| 587 | * @param a2 the other array |
|---|
| 588 | * |
|---|
| 589 | * @return an array that contains all entries from both input arrays |
|---|
| 590 | */ |
|---|
| 591 | protected SimpleLinearRegression[][] mergeArrays(SimpleLinearRegression[][] a1, |
|---|
| 592 | SimpleLinearRegression[][] a2){ |
|---|
| 593 | int numModels1 = a1[0].length; |
|---|
| 594 | int numModels2 = a2[0].length; |
|---|
| 595 | |
|---|
| 596 | SimpleLinearRegression[][] result = |
|---|
| 597 | new SimpleLinearRegression[m_numClasses][numModels1 + numModels2]; |
|---|
| 598 | |
|---|
| 599 | for (int i = 0; i < m_numClasses; i++) |
|---|
| 600 | for (int j = 0; j < numModels1; j++) { |
|---|
| 601 | result[i][j] = a1[i][j]; |
|---|
| 602 | } |
|---|
| 603 | for (int i = 0; i < m_numClasses; i++) |
|---|
| 604 | for (int j = 0; j < numModels2; j++) result[i][j+numModels1] = a2[i][j]; |
|---|
| 605 | return result; |
|---|
| 606 | } |
|---|
| 607 | |
|---|
| 608 | /** |
|---|
| 609 | * Return a list of all inner nodes in the tree |
|---|
| 610 | * @return the list of nodes |
|---|
| 611 | */ |
|---|
| 612 | public Vector getNodes(){ |
|---|
| 613 | Vector nodeList = new Vector(); |
|---|
| 614 | getNodes(nodeList); |
|---|
| 615 | return nodeList; |
|---|
| 616 | } |
|---|
| 617 | |
|---|
| 618 | /** |
|---|
| 619 | * Fills a list with all inner nodes in the tree |
|---|
| 620 | * |
|---|
| 621 | * @param nodeList the list to be filled |
|---|
| 622 | */ |
|---|
| 623 | public void getNodes(Vector nodeList) { |
|---|
| 624 | if (!m_isLeaf) { |
|---|
| 625 | nodeList.add(this); |
|---|
| 626 | for (int i = 0; i < m_sons.length; i++) m_sons[i].getNodes(nodeList); |
|---|
| 627 | } |
|---|
| 628 | } |
|---|
| 629 | |
|---|
| 630 | /** |
|---|
| 631 | * Returns a numeric version of a set of instances. |
|---|
| 632 | * All nominal attributes are replaced by binary ones, and the class variable is replaced |
|---|
| 633 | * by a pseudo-class variable that is used by LogitBoost. |
|---|
| 634 | */ |
|---|
| 635 | protected Instances getNumericData(Instances train) throws Exception{ |
|---|
| 636 | |
|---|
| 637 | Instances filteredData = new Instances(train); |
|---|
| 638 | m_nominalToBinary = new NominalToBinary(); |
|---|
| 639 | m_nominalToBinary.setInputFormat(filteredData); |
|---|
| 640 | filteredData = Filter.useFilter(filteredData, m_nominalToBinary); |
|---|
| 641 | |
|---|
| 642 | return super.getNumericData(filteredData); |
|---|
| 643 | } |
|---|
| 644 | |
|---|
| 645 | /** |
|---|
| 646 | * Computes the F-values of LogitBoost for an instance from the current logistic model at the node |
|---|
| 647 | * Note that this also takes into account the (partial) logistic model fit at higher levels in |
|---|
| 648 | * the tree. |
|---|
| 649 | * @param instance the instance |
|---|
| 650 | * @return the array of F-values |
|---|
| 651 | */ |
|---|
| 652 | protected double[] getFs(Instance instance) throws Exception{ |
|---|
| 653 | |
|---|
| 654 | double [] pred = new double [m_numClasses]; |
|---|
| 655 | |
|---|
| 656 | //Need to take into account partial model fit at higher levels in the tree (m_higherRegressions) |
|---|
| 657 | //and the part of the model fit at this node (m_regressions). |
|---|
| 658 | |
|---|
| 659 | //Fs from m_regressions (use method of LogisticBase) |
|---|
| 660 | double [] instanceFs = super.getFs(instance); |
|---|
| 661 | |
|---|
| 662 | //Fs from m_higherRegressions |
|---|
| 663 | for (int i = 0; i < m_numHigherRegressions; i++) { |
|---|
| 664 | double predSum = 0; |
|---|
| 665 | for (int j = 0; j < m_numClasses; j++) { |
|---|
| 666 | pred[j] = m_higherRegressions[j][i].classifyInstance(instance); |
|---|
| 667 | predSum += pred[j]; |
|---|
| 668 | } |
|---|
| 669 | predSum /= m_numClasses; |
|---|
| 670 | for (int j = 0; j < m_numClasses; j++) { |
|---|
| 671 | instanceFs[j] += (pred[j] - predSum) * (m_numClasses - 1) |
|---|
| 672 | / m_numClasses; |
|---|
| 673 | } |
|---|
| 674 | } |
|---|
| 675 | return instanceFs; |
|---|
| 676 | } |
|---|
| 677 | |
|---|
| 678 | /** |
|---|
| 679 | *Returns true if the logistic regression model at this node has changed compared to the |
|---|
| 680 | *one at the parent node. |
|---|
| 681 | *@return whether it has changed |
|---|
| 682 | */ |
|---|
| 683 | public boolean hasModels() { |
|---|
| 684 | return (m_numRegressions > 0); |
|---|
| 685 | } |
|---|
| 686 | |
|---|
| 687 | /** |
|---|
| 688 | * Returns the class probabilities for an instance according to the logistic model at the node. |
|---|
| 689 | * @param instance the instance |
|---|
| 690 | * @return the array of probabilities |
|---|
| 691 | */ |
|---|
| 692 | public double[] modelDistributionForInstance(Instance instance) throws Exception { |
|---|
| 693 | |
|---|
| 694 | //make copy and convert nominal attributes |
|---|
| 695 | instance = (Instance)instance.copy(); |
|---|
| 696 | m_nominalToBinary.input(instance); |
|---|
| 697 | instance = m_nominalToBinary.output(); |
|---|
| 698 | |
|---|
| 699 | //saet numeric pseudo-class |
|---|
| 700 | instance.setDataset(m_numericDataHeader); |
|---|
| 701 | |
|---|
| 702 | return probs(getFs(instance)); |
|---|
| 703 | } |
|---|
| 704 | |
|---|
| 705 | /** |
|---|
| 706 | * Returns the class probabilities for an instance given by the logistic model tree. |
|---|
| 707 | * @param instance the instance |
|---|
| 708 | * @return the array of probabilities |
|---|
| 709 | */ |
|---|
| 710 | public double[] distributionForInstance(Instance instance) throws Exception { |
|---|
| 711 | |
|---|
| 712 | double[] probs; |
|---|
| 713 | |
|---|
| 714 | if (m_isLeaf) { |
|---|
| 715 | //leaf: use logistic model |
|---|
| 716 | probs = modelDistributionForInstance(instance); |
|---|
| 717 | } else { |
|---|
| 718 | //sort into appropiate child node |
|---|
| 719 | int branch = m_localModel.whichSubset(instance); |
|---|
| 720 | probs = m_sons[branch].distributionForInstance(instance); |
|---|
| 721 | } |
|---|
| 722 | return probs; |
|---|
| 723 | } |
|---|
| 724 | |
|---|
| 725 | /** |
|---|
| 726 | * Returns the number of leaves (normal count). |
|---|
| 727 | * @return the number of leaves |
|---|
| 728 | */ |
|---|
| 729 | public int numLeaves() { |
|---|
| 730 | if (m_isLeaf) return 1; |
|---|
| 731 | int numLeaves = 0; |
|---|
| 732 | for (int i = 0; i < m_sons.length; i++) numLeaves += m_sons[i].numLeaves(); |
|---|
| 733 | return numLeaves; |
|---|
| 734 | } |
|---|
| 735 | |
|---|
| 736 | /** |
|---|
| 737 | * Returns the number of nodes. |
|---|
| 738 | * @return the number of nodes |
|---|
| 739 | */ |
|---|
| 740 | public int numNodes() { |
|---|
| 741 | if (m_isLeaf) return 1; |
|---|
| 742 | int numNodes = 1; |
|---|
| 743 | for (int i = 0; i < m_sons.length; i++) numNodes += m_sons[i].numNodes(); |
|---|
| 744 | return numNodes; |
|---|
| 745 | } |
|---|
| 746 | |
|---|
| 747 | /** |
|---|
| 748 | * Returns a description of the logistic model tree (tree structure and logistic models) |
|---|
| 749 | * @return describing string |
|---|
| 750 | */ |
|---|
| 751 | public String toString(){ |
|---|
| 752 | //assign numbers to logistic regression functions at leaves |
|---|
| 753 | assignLeafModelNumbers(0); |
|---|
| 754 | try{ |
|---|
| 755 | StringBuffer text = new StringBuffer(); |
|---|
| 756 | |
|---|
| 757 | if (m_isLeaf) { |
|---|
| 758 | text.append(": "); |
|---|
| 759 | text.append("LM_"+m_leafModelNum+":"+getModelParameters()); |
|---|
| 760 | } else { |
|---|
| 761 | dumpTree(0,text); |
|---|
| 762 | } |
|---|
| 763 | text.append("\n\nNumber of Leaves : \t"+numLeaves()+"\n"); |
|---|
| 764 | text.append("\nSize of the Tree : \t"+numNodes()+"\n"); |
|---|
| 765 | |
|---|
| 766 | //This prints logistic models after the tree, comment out if only tree should be printed |
|---|
| 767 | text.append(modelsToString()); |
|---|
| 768 | return text.toString(); |
|---|
| 769 | } catch (Exception e){ |
|---|
| 770 | return "Can't print logistic model tree"; |
|---|
| 771 | } |
|---|
| 772 | |
|---|
| 773 | |
|---|
| 774 | } |
|---|
| 775 | |
|---|
| 776 | /** |
|---|
| 777 | * Returns a string describing the number of LogitBoost iterations performed at this node, the total number |
|---|
| 778 | * of LogitBoost iterations performed (including iterations at higher levels in the tree), and the number |
|---|
| 779 | * of training examples at this node. |
|---|
| 780 | * @return the describing string |
|---|
| 781 | */ |
|---|
| 782 | public String getModelParameters(){ |
|---|
| 783 | |
|---|
| 784 | StringBuffer text = new StringBuffer(); |
|---|
| 785 | int numModels = m_numRegressions+m_numHigherRegressions; |
|---|
| 786 | text.append(m_numRegressions+"/"+numModels+" ("+m_numInstances+")"); |
|---|
| 787 | return text.toString(); |
|---|
| 788 | } |
|---|
| 789 | |
|---|
| 790 | |
|---|
| 791 | /** |
|---|
| 792 | * Help method for printing tree structure. |
|---|
| 793 | * |
|---|
| 794 | * @throws Exception if something goes wrong |
|---|
| 795 | */ |
|---|
| 796 | protected void dumpTree(int depth,StringBuffer text) |
|---|
| 797 | throws Exception { |
|---|
| 798 | |
|---|
| 799 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 800 | text.append("\n"); |
|---|
| 801 | for (int j = 0; j < depth; j++) |
|---|
| 802 | text.append("| "); |
|---|
| 803 | text.append(m_localModel.leftSide(m_train)); |
|---|
| 804 | text.append(m_localModel.rightSide(i, m_train)); |
|---|
| 805 | if (m_sons[i].m_isLeaf) { |
|---|
| 806 | text.append(": "); |
|---|
| 807 | text.append("LM_"+m_sons[i].m_leafModelNum+":"+m_sons[i].getModelParameters()); |
|---|
| 808 | }else |
|---|
| 809 | m_sons[i].dumpTree(depth+1,text); |
|---|
| 810 | } |
|---|
| 811 | } |
|---|
| 812 | |
|---|
| 813 | /** |
|---|
| 814 | * Assigns unique IDs to all nodes in the tree |
|---|
| 815 | */ |
|---|
| 816 | public int assignIDs(int lastID) { |
|---|
| 817 | |
|---|
| 818 | int currLastID = lastID + 1; |
|---|
| 819 | |
|---|
| 820 | m_id = currLastID; |
|---|
| 821 | if (m_sons != null) { |
|---|
| 822 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 823 | currLastID = m_sons[i].assignIDs(currLastID); |
|---|
| 824 | } |
|---|
| 825 | } |
|---|
| 826 | return currLastID; |
|---|
| 827 | } |
|---|
| 828 | |
|---|
| 829 | /** |
|---|
| 830 | * Assigns numbers to the logistic regression models at the leaves of the tree |
|---|
| 831 | */ |
|---|
| 832 | public int assignLeafModelNumbers(int leafCounter) { |
|---|
| 833 | if (!m_isLeaf) { |
|---|
| 834 | m_leafModelNum = 0; |
|---|
| 835 | for (int i = 0; i < m_sons.length; i++){ |
|---|
| 836 | leafCounter = m_sons[i].assignLeafModelNumbers(leafCounter); |
|---|
| 837 | } |
|---|
| 838 | } else { |
|---|
| 839 | leafCounter++; |
|---|
| 840 | m_leafModelNum = leafCounter; |
|---|
| 841 | } |
|---|
| 842 | return leafCounter; |
|---|
| 843 | } |
|---|
| 844 | |
|---|
| 845 | /** |
|---|
| 846 | * Returns an array containing the coefficients of the logistic regression function at this node. |
|---|
| 847 | * @return the array of coefficients, first dimension is the class, second the attribute. |
|---|
| 848 | */ |
|---|
| 849 | protected double[][] getCoefficients(){ |
|---|
| 850 | |
|---|
| 851 | //Need to take into account partial model fit at higher levels in the tree (m_higherRegressions) |
|---|
| 852 | //and the part of the model fit at this node (m_regressions). |
|---|
| 853 | |
|---|
| 854 | //get coefficients from m_regressions: use method of LogisticBase |
|---|
| 855 | double[][] coefficients = super.getCoefficients(); |
|---|
| 856 | //get coefficients from m_higherRegressions: |
|---|
| 857 | double constFactor = (double)(m_numClasses - 1) / (double)m_numClasses; // (J - 1)/J |
|---|
| 858 | for (int j = 0; j < m_numClasses; j++) { |
|---|
| 859 | for (int i = 0; i < m_numHigherRegressions; i++) { |
|---|
| 860 | double slope = m_higherRegressions[j][i].getSlope(); |
|---|
| 861 | double intercept = m_higherRegressions[j][i].getIntercept(); |
|---|
| 862 | int attribute = m_higherRegressions[j][i].getAttributeIndex(); |
|---|
| 863 | coefficients[j][0] += constFactor * intercept; |
|---|
| 864 | coefficients[j][attribute + 1] += constFactor * slope; |
|---|
| 865 | } |
|---|
| 866 | } |
|---|
| 867 | |
|---|
| 868 | return coefficients; |
|---|
| 869 | } |
|---|
| 870 | |
|---|
| 871 | /** |
|---|
| 872 | * Returns a string describing the logistic regression function at the node. |
|---|
| 873 | */ |
|---|
| 874 | public String modelsToString(){ |
|---|
| 875 | |
|---|
| 876 | StringBuffer text = new StringBuffer(); |
|---|
| 877 | if (m_isLeaf) { |
|---|
| 878 | text.append("LM_"+m_leafModelNum+":"+super.toString()); |
|---|
| 879 | } else { |
|---|
| 880 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 881 | text.append("\n"+m_sons[i].modelsToString()); |
|---|
| 882 | } |
|---|
| 883 | } |
|---|
| 884 | return text.toString(); |
|---|
| 885 | } |
|---|
| 886 | |
|---|
| 887 | /** |
|---|
| 888 | * Returns graph describing the tree. |
|---|
| 889 | * |
|---|
| 890 | * @throws Exception if something goes wrong |
|---|
| 891 | */ |
|---|
| 892 | public String graph() throws Exception { |
|---|
| 893 | |
|---|
| 894 | StringBuffer text = new StringBuffer(); |
|---|
| 895 | |
|---|
| 896 | assignIDs(-1); |
|---|
| 897 | assignLeafModelNumbers(0); |
|---|
| 898 | text.append("digraph LMTree {\n"); |
|---|
| 899 | if (m_isLeaf) { |
|---|
| 900 | text.append("N" + m_id + " [label=\"LM_"+m_leafModelNum+":"+getModelParameters()+"\" " + |
|---|
| 901 | "shape=box style=filled"); |
|---|
| 902 | text.append("]\n"); |
|---|
| 903 | }else { |
|---|
| 904 | text.append("N" + m_id |
|---|
| 905 | + " [label=\"" + |
|---|
| 906 | m_localModel.leftSide(m_train) + "\" "); |
|---|
| 907 | text.append("]\n"); |
|---|
| 908 | graphTree(text); |
|---|
| 909 | } |
|---|
| 910 | |
|---|
| 911 | return text.toString() +"}\n"; |
|---|
| 912 | } |
|---|
| 913 | |
|---|
| 914 | /** |
|---|
| 915 | * Helper function for graph description of tree |
|---|
| 916 | * |
|---|
| 917 | * @throws Exception if something goes wrong |
|---|
| 918 | */ |
|---|
| 919 | private void graphTree(StringBuffer text) throws Exception { |
|---|
| 920 | |
|---|
| 921 | for (int i = 0; i < m_sons.length; i++) { |
|---|
| 922 | text.append("N" + m_id |
|---|
| 923 | + "->" + |
|---|
| 924 | "N" + m_sons[i].m_id + |
|---|
| 925 | " [label=\"" + m_localModel.rightSide(i,m_train).trim() + |
|---|
| 926 | "\"]\n"); |
|---|
| 927 | if (m_sons[i].m_isLeaf) { |
|---|
| 928 | text.append("N" +m_sons[i].m_id + " [label=\"LM_"+m_sons[i].m_leafModelNum+":"+ |
|---|
| 929 | m_sons[i].getModelParameters()+"\" " + "shape=box style=filled"); |
|---|
| 930 | text.append("]\n"); |
|---|
| 931 | } else { |
|---|
| 932 | text.append("N" + m_sons[i].m_id + |
|---|
| 933 | " [label=\""+m_sons[i].m_localModel.leftSide(m_train) + |
|---|
| 934 | "\" "); |
|---|
| 935 | text.append("]\n"); |
|---|
| 936 | m_sons[i].graphTree(text); |
|---|
| 937 | } |
|---|
| 938 | } |
|---|
| 939 | } |
|---|
| 940 | |
|---|
| 941 | /** |
|---|
| 942 | * Cleanup in order to save memory. |
|---|
| 943 | */ |
|---|
| 944 | public void cleanup() { |
|---|
| 945 | super.cleanup(); |
|---|
| 946 | if (!m_isLeaf) { |
|---|
| 947 | for (int i = 0; i < m_sons.length; i++) m_sons[i].cleanup(); |
|---|
| 948 | } |
|---|
| 949 | } |
|---|
| 950 | |
|---|
| 951 | /** |
|---|
| 952 | * Returns the revision string. |
|---|
| 953 | * |
|---|
| 954 | * @return the revision |
|---|
| 955 | */ |
|---|
| 956 | public String getRevision() { |
|---|
| 957 | return RevisionUtils.extract("$Revision: 1.8 $"); |
|---|
| 958 | } |
|---|
| 959 | } |
|---|