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 | * ClassifierDecList.java |
---|
19 | * Copyright (C) 1999 University of Waikato, Hamilton, New Zealand |
---|
20 | * |
---|
21 | */ |
---|
22 | |
---|
23 | package weka.classifiers.rules.part; |
---|
24 | |
---|
25 | import weka.classifiers.trees.j48.ClassifierSplitModel; |
---|
26 | import weka.classifiers.trees.j48.Distribution; |
---|
27 | import weka.classifiers.trees.j48.EntropySplitCrit; |
---|
28 | import weka.classifiers.trees.j48.ModelSelection; |
---|
29 | import weka.classifiers.trees.j48.NoSplit; |
---|
30 | import weka.core.Instance; |
---|
31 | import weka.core.Instances; |
---|
32 | import weka.core.RevisionHandler; |
---|
33 | import weka.core.RevisionUtils; |
---|
34 | import weka.core.Utils; |
---|
35 | |
---|
36 | import java.io.Serializable; |
---|
37 | |
---|
38 | /** |
---|
39 | * Class for handling a rule (partial tree) for a decision list. |
---|
40 | * |
---|
41 | * @author Eibe Frank (eibe@cs.waikato.ac.nz) |
---|
42 | * @version $Revision: 1.13 $ |
---|
43 | */ |
---|
44 | public class ClassifierDecList |
---|
45 | implements Serializable, RevisionHandler { |
---|
46 | |
---|
47 | /** for serialization */ |
---|
48 | private static final long serialVersionUID = 7284358349711992497L; |
---|
49 | |
---|
50 | /** Minimum number of objects */ |
---|
51 | protected int m_minNumObj; |
---|
52 | |
---|
53 | /** To compute the entropy. */ |
---|
54 | protected static EntropySplitCrit m_splitCrit = new EntropySplitCrit(); |
---|
55 | |
---|
56 | /** The model selection method. */ |
---|
57 | protected ModelSelection m_toSelectModel; |
---|
58 | |
---|
59 | /** Local model at node. */ |
---|
60 | protected ClassifierSplitModel m_localModel; |
---|
61 | |
---|
62 | /** References to sons. */ |
---|
63 | protected ClassifierDecList [] m_sons; |
---|
64 | |
---|
65 | /** True if node is leaf. */ |
---|
66 | protected boolean m_isLeaf; |
---|
67 | |
---|
68 | /** True if node is empty. */ |
---|
69 | protected boolean m_isEmpty; |
---|
70 | |
---|
71 | /** The training instances. */ |
---|
72 | protected Instances m_train; |
---|
73 | |
---|
74 | /** The pruning instances. */ |
---|
75 | protected Distribution m_test; |
---|
76 | |
---|
77 | /** Which son to expand? */ |
---|
78 | protected int indeX; |
---|
79 | |
---|
80 | /** |
---|
81 | * Constructor - just calls constructor of class DecList. |
---|
82 | */ |
---|
83 | public ClassifierDecList(ModelSelection toSelectLocModel, int minNum){ |
---|
84 | |
---|
85 | m_toSelectModel = toSelectLocModel; |
---|
86 | m_minNumObj = minNum; |
---|
87 | } |
---|
88 | |
---|
89 | /** |
---|
90 | * Method for building a pruned partial tree. |
---|
91 | * |
---|
92 | * @exception Exception if something goes wrong |
---|
93 | */ |
---|
94 | public void buildRule(Instances data) throws Exception { |
---|
95 | |
---|
96 | buildDecList(data, false); |
---|
97 | |
---|
98 | cleanup(new Instances(data, 0)); |
---|
99 | } |
---|
100 | |
---|
101 | /** |
---|
102 | * Builds the partial tree without hold out set. |
---|
103 | * |
---|
104 | * @exception Exception if something goes wrong |
---|
105 | */ |
---|
106 | public void buildDecList(Instances data, boolean leaf) throws Exception { |
---|
107 | |
---|
108 | Instances [] localInstances,localPruneInstances; |
---|
109 | int index,ind; |
---|
110 | int i,j; |
---|
111 | double sumOfWeights; |
---|
112 | NoSplit noSplit; |
---|
113 | |
---|
114 | m_train = null; |
---|
115 | m_test = null; |
---|
116 | m_isLeaf = false; |
---|
117 | m_isEmpty = false; |
---|
118 | m_sons = null; |
---|
119 | indeX = 0; |
---|
120 | sumOfWeights = data.sumOfWeights(); |
---|
121 | noSplit = new NoSplit (new Distribution((Instances)data)); |
---|
122 | if (leaf) |
---|
123 | m_localModel = noSplit; |
---|
124 | else |
---|
125 | m_localModel = m_toSelectModel.selectModel(data); |
---|
126 | if (m_localModel.numSubsets() > 1) { |
---|
127 | localInstances = m_localModel.split(data); |
---|
128 | data = null; |
---|
129 | m_sons = new ClassifierDecList [m_localModel.numSubsets()]; |
---|
130 | i = 0; |
---|
131 | do { |
---|
132 | i++; |
---|
133 | ind = chooseIndex(); |
---|
134 | if (ind == -1) { |
---|
135 | for (j = 0; j < m_sons.length; j++) |
---|
136 | if (m_sons[j] == null) |
---|
137 | m_sons[j] = getNewDecList(localInstances[j],true); |
---|
138 | if (i < 2) { |
---|
139 | m_localModel = noSplit; |
---|
140 | m_isLeaf = true; |
---|
141 | m_sons = null; |
---|
142 | if (Utils.eq(sumOfWeights,0)) |
---|
143 | m_isEmpty = true; |
---|
144 | return; |
---|
145 | } |
---|
146 | ind = 0; |
---|
147 | break; |
---|
148 | } else |
---|
149 | m_sons[ind] = getNewDecList(localInstances[ind],false); |
---|
150 | } while ((i < m_sons.length) && (m_sons[ind].m_isLeaf)); |
---|
151 | |
---|
152 | // Choose rule |
---|
153 | indeX = chooseLastIndex(); |
---|
154 | }else{ |
---|
155 | m_isLeaf = true; |
---|
156 | if (Utils.eq(sumOfWeights, 0)) |
---|
157 | m_isEmpty = true; |
---|
158 | } |
---|
159 | } |
---|
160 | |
---|
161 | /** |
---|
162 | * Classifies an instance. |
---|
163 | * |
---|
164 | * @exception Exception if something goes wrong |
---|
165 | */ |
---|
166 | public double classifyInstance(Instance instance) |
---|
167 | throws Exception { |
---|
168 | |
---|
169 | double maxProb = -1; |
---|
170 | double currentProb; |
---|
171 | int maxIndex = 0; |
---|
172 | int j; |
---|
173 | |
---|
174 | for (j = 0; j < instance.numClasses(); |
---|
175 | j++){ |
---|
176 | currentProb = getProbs(j,instance,1); |
---|
177 | if (Utils.gr(currentProb,maxProb)){ |
---|
178 | maxIndex = j; |
---|
179 | maxProb = currentProb; |
---|
180 | } |
---|
181 | } |
---|
182 | if (Utils.eq(maxProb,0)) |
---|
183 | return -1.0; |
---|
184 | else |
---|
185 | return (double)maxIndex; |
---|
186 | } |
---|
187 | |
---|
188 | /** |
---|
189 | * Returns class probabilities for a weighted instance. |
---|
190 | * |
---|
191 | * @exception Exception if something goes wrong |
---|
192 | */ |
---|
193 | public final double [] distributionForInstance(Instance instance) |
---|
194 | throws Exception { |
---|
195 | |
---|
196 | |
---|
197 | double [] doubles = |
---|
198 | new double[instance.numClasses()]; |
---|
199 | |
---|
200 | for (int i = 0; i < doubles.length; i++) |
---|
201 | doubles[i] = getProbs(i,instance,1); |
---|
202 | |
---|
203 | return doubles; |
---|
204 | } |
---|
205 | |
---|
206 | /** |
---|
207 | * Returns the weight a rule assigns to an instance. |
---|
208 | * |
---|
209 | * @exception Exception if something goes wrong |
---|
210 | */ |
---|
211 | public double weight(Instance instance) throws Exception { |
---|
212 | |
---|
213 | int subset; |
---|
214 | |
---|
215 | if (m_isLeaf) |
---|
216 | return 1; |
---|
217 | subset = m_localModel.whichSubset(instance); |
---|
218 | if (subset == -1) |
---|
219 | return (m_localModel.weights(instance))[indeX]* |
---|
220 | m_sons[indeX].weight(instance); |
---|
221 | if (subset == indeX) |
---|
222 | return m_sons[indeX].weight(instance); |
---|
223 | return 0; |
---|
224 | } |
---|
225 | |
---|
226 | /** |
---|
227 | * Cleanup in order to save memory. |
---|
228 | */ |
---|
229 | public final void cleanup(Instances justHeaderInfo) { |
---|
230 | |
---|
231 | m_train = justHeaderInfo; |
---|
232 | m_test = null; |
---|
233 | if (!m_isLeaf) |
---|
234 | for (int i = 0; i < m_sons.length; i++) |
---|
235 | if (m_sons[i] != null) |
---|
236 | m_sons[i].cleanup(justHeaderInfo); |
---|
237 | } |
---|
238 | |
---|
239 | /** |
---|
240 | * Prints rules. |
---|
241 | */ |
---|
242 | public String toString(){ |
---|
243 | |
---|
244 | try { |
---|
245 | StringBuffer text; |
---|
246 | |
---|
247 | text = new StringBuffer(); |
---|
248 | if (m_isLeaf){ |
---|
249 | text.append(": "); |
---|
250 | text.append(m_localModel.dumpLabel(0,m_train)+"\n"); |
---|
251 | }else{ |
---|
252 | dumpDecList(text); |
---|
253 | //dumpTree(0,text); |
---|
254 | } |
---|
255 | return text.toString(); |
---|
256 | } catch (Exception e) { |
---|
257 | return "Can't print rule."; |
---|
258 | } |
---|
259 | } |
---|
260 | |
---|
261 | /** |
---|
262 | * Returns a newly created tree. |
---|
263 | * |
---|
264 | * @exception Exception if something goes wrong |
---|
265 | */ |
---|
266 | protected ClassifierDecList getNewDecList(Instances train, boolean leaf) |
---|
267 | throws Exception { |
---|
268 | |
---|
269 | ClassifierDecList newDecList = new ClassifierDecList(m_toSelectModel, |
---|
270 | m_minNumObj); |
---|
271 | newDecList.buildDecList(train,leaf); |
---|
272 | |
---|
273 | return newDecList; |
---|
274 | } |
---|
275 | |
---|
276 | /** |
---|
277 | * Method for choosing a subset to expand. |
---|
278 | */ |
---|
279 | public final int chooseIndex() { |
---|
280 | |
---|
281 | int minIndex = -1; |
---|
282 | double estimated, min = Double.MAX_VALUE; |
---|
283 | int i, j; |
---|
284 | |
---|
285 | for (i = 0; i < m_sons.length; i++) |
---|
286 | if (son(i) == null) { |
---|
287 | if (Utils.sm(localModel().distribution().perBag(i), |
---|
288 | (double)m_minNumObj)) |
---|
289 | estimated = Double.MAX_VALUE; |
---|
290 | else{ |
---|
291 | estimated = 0; |
---|
292 | for (j = 0; j < localModel().distribution().numClasses(); j++) |
---|
293 | estimated -= m_splitCrit.logFunc(localModel().distribution(). |
---|
294 | perClassPerBag(i,j)); |
---|
295 | estimated += m_splitCrit.logFunc(localModel().distribution(). |
---|
296 | perBag(i)); |
---|
297 | estimated /= localModel().distribution().perBag(i); |
---|
298 | } |
---|
299 | if (Utils.smOrEq(estimated,0)) |
---|
300 | return i; |
---|
301 | if (Utils.sm(estimated,min)) { |
---|
302 | min = estimated; |
---|
303 | minIndex = i; |
---|
304 | } |
---|
305 | } |
---|
306 | |
---|
307 | return minIndex; |
---|
308 | } |
---|
309 | |
---|
310 | /** |
---|
311 | * Choose last index (ie. choose rule). |
---|
312 | */ |
---|
313 | public final int chooseLastIndex() { |
---|
314 | |
---|
315 | int minIndex = 0; |
---|
316 | double estimated, min = Double.MAX_VALUE; |
---|
317 | |
---|
318 | if (!m_isLeaf) |
---|
319 | for (int i = 0; i < m_sons.length; i++) |
---|
320 | if (son(i) != null) { |
---|
321 | if (Utils.grOrEq(localModel().distribution().perBag(i), |
---|
322 | (double)m_minNumObj)) { |
---|
323 | estimated = son(i).getSizeOfBranch(); |
---|
324 | if (Utils.sm(estimated,min)) { |
---|
325 | min = estimated; |
---|
326 | minIndex = i; |
---|
327 | } |
---|
328 | } |
---|
329 | } |
---|
330 | |
---|
331 | return minIndex; |
---|
332 | } |
---|
333 | |
---|
334 | /** |
---|
335 | * Returns the number of instances covered by a branch |
---|
336 | */ |
---|
337 | protected double getSizeOfBranch() { |
---|
338 | |
---|
339 | if (m_isLeaf) { |
---|
340 | return -localModel().distribution().total(); |
---|
341 | } else |
---|
342 | return son(indeX).getSizeOfBranch(); |
---|
343 | } |
---|
344 | |
---|
345 | /** |
---|
346 | * Help method for printing tree structure. |
---|
347 | */ |
---|
348 | private void dumpDecList(StringBuffer text) throws Exception { |
---|
349 | |
---|
350 | text.append(m_localModel.leftSide(m_train)); |
---|
351 | text.append(m_localModel.rightSide(indeX, m_train)); |
---|
352 | if (m_sons[indeX].m_isLeaf){ |
---|
353 | text.append(": "); |
---|
354 | text.append(m_localModel.dumpLabel(indeX,m_train)+"\n"); |
---|
355 | }else{ |
---|
356 | text.append(" AND\n"); |
---|
357 | m_sons[indeX].dumpDecList(text); |
---|
358 | } |
---|
359 | } |
---|
360 | |
---|
361 | /** |
---|
362 | * Dumps the partial tree (only used for debugging) |
---|
363 | * |
---|
364 | * @exception Exception Exception if something goes wrong |
---|
365 | */ |
---|
366 | private void dumpTree(int depth,StringBuffer text) |
---|
367 | throws Exception { |
---|
368 | |
---|
369 | int i,j; |
---|
370 | |
---|
371 | for (i=0;i<m_sons.length;i++){ |
---|
372 | text.append("\n");; |
---|
373 | for (j=0;j<depth;j++) |
---|
374 | text.append("| "); |
---|
375 | text.append(m_localModel.leftSide(m_train)); |
---|
376 | text.append(m_localModel.rightSide(i, m_train)); |
---|
377 | if (m_sons[i] == null) |
---|
378 | text.append("null"); |
---|
379 | else if (m_sons[i].m_isLeaf){ |
---|
380 | text.append(": "); |
---|
381 | text.append(m_localModel.dumpLabel(i,m_train)); |
---|
382 | }else |
---|
383 | m_sons[i].dumpTree(depth+1,text); |
---|
384 | } |
---|
385 | } |
---|
386 | |
---|
387 | /** |
---|
388 | * Help method for computing class probabilities of |
---|
389 | * a given instance. |
---|
390 | * |
---|
391 | * @exception Exception Exception if something goes wrong |
---|
392 | */ |
---|
393 | private double getProbs(int classIndex,Instance instance, |
---|
394 | double weight) throws Exception { |
---|
395 | |
---|
396 | double [] weights; |
---|
397 | int treeIndex; |
---|
398 | |
---|
399 | if (m_isLeaf) { |
---|
400 | return weight * localModel().classProb(classIndex, instance, -1); |
---|
401 | } else { |
---|
402 | treeIndex = localModel().whichSubset(instance); |
---|
403 | if (treeIndex == -1) { |
---|
404 | weights = localModel().weights(instance); |
---|
405 | return son(indeX).getProbs(classIndex, instance, |
---|
406 | weights[indeX] * weight); |
---|
407 | }else{ |
---|
408 | if (treeIndex == indeX) { |
---|
409 | return son(indeX).getProbs(classIndex, instance, weight); |
---|
410 | } else { |
---|
411 | return 0; |
---|
412 | } |
---|
413 | } |
---|
414 | } |
---|
415 | } |
---|
416 | |
---|
417 | /** |
---|
418 | * Method just exists to make program easier to read. |
---|
419 | */ |
---|
420 | protected ClassifierSplitModel localModel(){ |
---|
421 | |
---|
422 | return (ClassifierSplitModel)m_localModel; |
---|
423 | } |
---|
424 | |
---|
425 | /** |
---|
426 | * Method just exists to make program easier to read. |
---|
427 | */ |
---|
428 | protected ClassifierDecList son(int index){ |
---|
429 | |
---|
430 | return m_sons[index]; |
---|
431 | } |
---|
432 | |
---|
433 | /** |
---|
434 | * Returns the revision string. |
---|
435 | * |
---|
436 | * @return the revision |
---|
437 | */ |
---|
438 | public String getRevision() { |
---|
439 | return RevisionUtils.extract("$Revision: 1.13 $"); |
---|
440 | } |
---|
441 | } |
---|