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 | * Discriminative Multinomial Naive Bayes for Text Classification |
---|
19 | * Copyright (C) 2008 Jiang Su |
---|
20 | */ |
---|
21 | |
---|
22 | package weka.classifiers.bayes; |
---|
23 | |
---|
24 | |
---|
25 | import weka.classifiers.Classifier; |
---|
26 | import weka.classifiers.AbstractClassifier; |
---|
27 | import weka.core.Instance; |
---|
28 | import weka.core.Instances; |
---|
29 | import weka.core.TechnicalInformation; |
---|
30 | import weka.core.TechnicalInformationHandler; |
---|
31 | import weka.core.Utils; |
---|
32 | import weka.core.WeightedInstancesHandler; |
---|
33 | import weka.core.Capabilities.Capability; |
---|
34 | import weka.core.TechnicalInformation.Field; |
---|
35 | import weka.core.TechnicalInformation.Type; |
---|
36 | import weka.classifiers.UpdateableClassifier; |
---|
37 | import java.util.*; |
---|
38 | import java.io.Serializable; |
---|
39 | import weka.core.Capabilities; |
---|
40 | import weka.core.OptionHandler; |
---|
41 | |
---|
42 | |
---|
43 | /** |
---|
44 | <!-- globalinfo-start --> |
---|
45 | * Class for building and using a Discriminative Multinomial Naive Bayes classifier. For more information see,<br/> |
---|
46 | * <br/> |
---|
47 | * Jiang Su,Harry Zhang,Charles X. Ling,Stan Matwin: Discriminative Parameter Learning for Bayesian Networks. In: ICML 2008', 2008.<br/> |
---|
48 | * <br/> |
---|
49 | * The core equation for this classifier:<br/> |
---|
50 | * <br/> |
---|
51 | * P[Ci|D] = (P[D|Ci] x P[Ci]) / P[D] (Bayes rule)<br/> |
---|
52 | * <br/> |
---|
53 | * where Ci is class i and D is a document. |
---|
54 | * <p/> |
---|
55 | <!-- globalinfo-end --> |
---|
56 | * |
---|
57 | <!-- technical-bibtex-start --> |
---|
58 | * BibTeX: |
---|
59 | * <pre> |
---|
60 | * @inproceedings{JiangSu2008, |
---|
61 | * author = {Jiang Su,Harry Zhang,Charles X. Ling,Stan Matwin}, |
---|
62 | * booktitle = {ICML 2008'}, |
---|
63 | * title = {Discriminative Parameter Learning for Bayesian Networks}, |
---|
64 | * year = {2008} |
---|
65 | * } |
---|
66 | * </pre> |
---|
67 | * <p/> |
---|
68 | <!-- technical-bibtex-end --> |
---|
69 | * |
---|
70 | <!-- options-start --> |
---|
71 | * Valid options are: <p/> |
---|
72 | * |
---|
73 | * <pre> -D |
---|
74 | * If set, classifier is run in debug mode and |
---|
75 | * may output additional info to the console</pre> |
---|
76 | * |
---|
77 | <!-- options-end --> |
---|
78 | * |
---|
79 | * @author Jiang Su (Jiang.Su@unb.ca) 2008 |
---|
80 | * @version $Revision: 5928 $ |
---|
81 | */ |
---|
82 | public class DMNBtext extends AbstractClassifier |
---|
83 | implements OptionHandler, WeightedInstancesHandler, |
---|
84 | TechnicalInformationHandler, UpdateableClassifier { |
---|
85 | |
---|
86 | /** for serialization */ |
---|
87 | static final long serialVersionUID = 5932177450183457085L; |
---|
88 | /** The number of iterations. */ |
---|
89 | protected int m_NumIterations = 1; |
---|
90 | protected boolean m_BinaryWord = true; |
---|
91 | int m_numClasses=-1; |
---|
92 | protected Instances m_headerInfo; |
---|
93 | DNBBinary[] m_binaryClassifiers = null; |
---|
94 | |
---|
95 | /** |
---|
96 | * Returns a string describing this classifier |
---|
97 | * @return a description of the classifier suitable for |
---|
98 | * displaying in the explorer/experimenter gui |
---|
99 | */ |
---|
100 | public String globalInfo() { |
---|
101 | return |
---|
102 | "Class for building and using a Discriminative Multinomial Naive Bayes classifier. " |
---|
103 | + "For more information see,\n\n" |
---|
104 | + getTechnicalInformation().toString() + "\n\n" |
---|
105 | + "The core equation for this classifier:\n\n" |
---|
106 | + "P[Ci|D] = (P[D|Ci] x P[Ci]) / P[D] (Bayes rule)\n\n" |
---|
107 | + "where Ci is class i and D is a document."; |
---|
108 | } |
---|
109 | |
---|
110 | /** |
---|
111 | * Returns an instance of a TechnicalInformation object, containing |
---|
112 | * detailed information about the technical background of this class, |
---|
113 | * e.g., paper reference or book this class is based on. |
---|
114 | * |
---|
115 | * @return the technical information about this class |
---|
116 | */ |
---|
117 | public TechnicalInformation getTechnicalInformation() { |
---|
118 | TechnicalInformation result; |
---|
119 | |
---|
120 | result = new TechnicalInformation(Type.INPROCEEDINGS); |
---|
121 | result.setValue(Field.AUTHOR, "Jiang Su,Harry Zhang,Charles X. Ling,Stan Matwin"); |
---|
122 | result.setValue(Field.YEAR, "2008"); |
---|
123 | result.setValue(Field.TITLE, "Discriminative Parameter Learning for Bayesian Networks"); |
---|
124 | result.setValue(Field.BOOKTITLE, "ICML 2008'"); |
---|
125 | |
---|
126 | return result; |
---|
127 | } |
---|
128 | |
---|
129 | /** |
---|
130 | * Returns default capabilities of the classifier. |
---|
131 | * |
---|
132 | * @return the capabilities of this classifier |
---|
133 | */ |
---|
134 | public Capabilities getCapabilities() { |
---|
135 | Capabilities result = super.getCapabilities(); |
---|
136 | result.disableAll(); |
---|
137 | |
---|
138 | // attributes |
---|
139 | result.enable(Capability.NUMERIC_ATTRIBUTES); |
---|
140 | |
---|
141 | // class |
---|
142 | result.enable(Capability.NOMINAL_CLASS); |
---|
143 | result.enable(Capability.MISSING_CLASS_VALUES); |
---|
144 | |
---|
145 | return result; |
---|
146 | } |
---|
147 | |
---|
148 | /** |
---|
149 | * Generates the classifier. |
---|
150 | * |
---|
151 | * @param data set of instances serving as training data |
---|
152 | * @exception Exception if the classifier has not been generated successfully |
---|
153 | */ |
---|
154 | public void buildClassifier(Instances data) throws Exception { |
---|
155 | // can classifier handle the data? |
---|
156 | getCapabilities().testWithFail(data); |
---|
157 | // remove instances with missing class |
---|
158 | Instances instances = new Instances(data); |
---|
159 | instances.deleteWithMissingClass(); |
---|
160 | |
---|
161 | m_binaryClassifiers = new DNBBinary[instances.numClasses()]; |
---|
162 | m_numClasses=instances.numClasses(); |
---|
163 | m_headerInfo = new Instances(instances, 0); |
---|
164 | for (int i = 0; i < instances.numClasses(); i++) { |
---|
165 | m_binaryClassifiers[i] = new DNBBinary(); |
---|
166 | m_binaryClassifiers[i].setTargetClass(i); |
---|
167 | m_binaryClassifiers[i].initClassifier(instances); |
---|
168 | } |
---|
169 | |
---|
170 | if (instances.numInstances() == 0) |
---|
171 | return; |
---|
172 | //Iterative update |
---|
173 | Random random = new Random(); |
---|
174 | for (int it = 0; it < m_NumIterations; it++) { |
---|
175 | for (int i = 0; i < instances.numInstances(); i++) { |
---|
176 | updateClassifier(instances.instance(i)); |
---|
177 | } |
---|
178 | } |
---|
179 | |
---|
180 | // Utils.normalize(m_oldClassDis); |
---|
181 | // Utils.normalize(m_ClassDis); |
---|
182 | // m_originalPositive = m_oldClassDis[0]; |
---|
183 | // m_positive = m_ClassDis[0]; |
---|
184 | |
---|
185 | } |
---|
186 | |
---|
187 | /** |
---|
188 | * Updates the classifier with the given instance. |
---|
189 | * |
---|
190 | * @param instance the new training instance to include in the model |
---|
191 | * @exception Exception if the instance could not be incorporated in |
---|
192 | * the model. |
---|
193 | */ |
---|
194 | |
---|
195 | public void updateClassifier(Instance instance) throws Exception { |
---|
196 | |
---|
197 | if (m_numClasses == 2) { |
---|
198 | m_binaryClassifiers[0].updateClassifier(instance); |
---|
199 | } else { |
---|
200 | for (int i = 0; i < instance.numClasses(); i++) |
---|
201 | m_binaryClassifiers[i].updateClassifier(instance); |
---|
202 | } |
---|
203 | } |
---|
204 | |
---|
205 | /** |
---|
206 | * Calculates the class membership probabilities for the given test |
---|
207 | * instance. |
---|
208 | * |
---|
209 | * @param instance the instance to be classified |
---|
210 | * @return predicted class probability distribution |
---|
211 | * @exception Exception if there is a problem generating the prediction |
---|
212 | */ |
---|
213 | public double[] distributionForInstance(Instance instance) throws Exception { |
---|
214 | if (m_numClasses == 2) { |
---|
215 | // System.out.println(m_binaryClassifiers[0].getProbForTargetClass(instance)); |
---|
216 | return m_binaryClassifiers[0].distributionForInstance(instance); |
---|
217 | } |
---|
218 | double[] logDocGivenClass = new double[instance.numClasses()]; |
---|
219 | for (int i = 0; i < m_numClasses; i++) |
---|
220 | logDocGivenClass[i] = m_binaryClassifiers[i].getLogProbForTargetClass(instance); |
---|
221 | |
---|
222 | |
---|
223 | double max = logDocGivenClass[Utils.maxIndex(logDocGivenClass)]; |
---|
224 | for(int i = 0; i<m_numClasses; i++) |
---|
225 | logDocGivenClass[i] = Math.exp(logDocGivenClass[i] - max); |
---|
226 | |
---|
227 | |
---|
228 | try { |
---|
229 | Utils.normalize(logDocGivenClass); |
---|
230 | } catch (Exception e) { |
---|
231 | e.printStackTrace(); |
---|
232 | |
---|
233 | |
---|
234 | } |
---|
235 | |
---|
236 | return logDocGivenClass; |
---|
237 | } |
---|
238 | /** |
---|
239 | * Returns a string representation of the classifier. |
---|
240 | * |
---|
241 | * @return a string representation of the classifier |
---|
242 | */ |
---|
243 | public String toString() { |
---|
244 | StringBuffer result = new StringBuffer(""); |
---|
245 | result.append("The log ratio of two conditional probabilities of a word w_i: log(p(w_i)|+)/p(w_i)|-)) in decent order based on their absolute values\n"); |
---|
246 | result.append("Can be used to measure the discriminative power of each word.\n"); |
---|
247 | if (m_numClasses == 2) { |
---|
248 | // System.out.println(m_binaryClassifiers[0].getProbForTargetClass(instance)); |
---|
249 | return result.append(m_binaryClassifiers[0].toString()).toString(); |
---|
250 | } |
---|
251 | for (int i = 0; i < m_numClasses; i++) |
---|
252 | { result.append(i+" against the rest classes\n"); |
---|
253 | result.append(m_binaryClassifiers[i].toString()+"\n"); |
---|
254 | } |
---|
255 | return result.toString(); |
---|
256 | } |
---|
257 | |
---|
258 | /* |
---|
259 | * Options after -- are passed to the designated classifier.<p> |
---|
260 | * |
---|
261 | * @param options the list of options as an array of strings |
---|
262 | * @exception Exception if an option is not supported |
---|
263 | */ |
---|
264 | public void setOptions(String[] options) throws Exception { |
---|
265 | |
---|
266 | String iterations = Utils.getOption('I', options); |
---|
267 | if (iterations.length() != 0) { |
---|
268 | setNumIterations(Integer.parseInt(iterations)); |
---|
269 | } else { |
---|
270 | setNumIterations(m_NumIterations); |
---|
271 | } |
---|
272 | iterations = Utils.getOption('B', options); |
---|
273 | if (iterations.length() != 0) { |
---|
274 | setBinaryWord(Boolean.parseBoolean(iterations)); |
---|
275 | } else { |
---|
276 | setBinaryWord(m_BinaryWord); |
---|
277 | } |
---|
278 | |
---|
279 | } |
---|
280 | |
---|
281 | /** |
---|
282 | * Gets the current settings of the classifier. |
---|
283 | * |
---|
284 | * @return an array of strings suitable for passing to setOptions |
---|
285 | */ |
---|
286 | public String[] getOptions() { |
---|
287 | |
---|
288 | String[] options = new String[4]; |
---|
289 | |
---|
290 | int current = 0; |
---|
291 | options[current++] = "-I"; |
---|
292 | options[current++] = "" + getNumIterations(); |
---|
293 | |
---|
294 | options[current++] = "-B"; |
---|
295 | options[current++] = "" + getBinaryWord(); |
---|
296 | |
---|
297 | return options; |
---|
298 | } |
---|
299 | |
---|
300 | /** |
---|
301 | * Returns the tip text for this property |
---|
302 | * @return tip text for this property suitable for |
---|
303 | * displaying in the explorer/experimenter gui |
---|
304 | */ |
---|
305 | public String numIterationsTipText() { |
---|
306 | return "The number of iterations that the classifier will scan the training data"; |
---|
307 | } |
---|
308 | |
---|
309 | /** |
---|
310 | * Sets the number of iterations to be performed |
---|
311 | */ |
---|
312 | public void setNumIterations(int numIterations) { |
---|
313 | |
---|
314 | m_NumIterations = numIterations; |
---|
315 | } |
---|
316 | |
---|
317 | /** |
---|
318 | * Gets the number of iterations to be performed |
---|
319 | * |
---|
320 | * @return the iterations to be performed |
---|
321 | */ |
---|
322 | public int getNumIterations() { |
---|
323 | |
---|
324 | return m_NumIterations; |
---|
325 | } |
---|
326 | /** |
---|
327 | * Returns the tip text for this property |
---|
328 | * @return tip text for this property suitable for |
---|
329 | * displaying in the explorer/experimenter gui |
---|
330 | */ |
---|
331 | public String binaryWordTipText() { |
---|
332 | return " whether ingore the frequency information in data"; |
---|
333 | } |
---|
334 | /** |
---|
335 | * Sets whether use binary text representation |
---|
336 | */ |
---|
337 | public void setBinaryWord(boolean val) { |
---|
338 | |
---|
339 | m_BinaryWord = val; |
---|
340 | } |
---|
341 | |
---|
342 | /** |
---|
343 | * Gets whether use binary text representation |
---|
344 | * |
---|
345 | * @return whether use binary text representation |
---|
346 | */ |
---|
347 | public boolean getBinaryWord() { |
---|
348 | |
---|
349 | return m_BinaryWord; |
---|
350 | } |
---|
351 | |
---|
352 | /** |
---|
353 | * Returns the revision string. |
---|
354 | * |
---|
355 | * @return the revision |
---|
356 | */ |
---|
357 | public String getRevision() { |
---|
358 | return "$Revision: 1.0"; |
---|
359 | } |
---|
360 | |
---|
361 | public class DNBBinary implements Serializable { |
---|
362 | |
---|
363 | /** The number of iterations. */ |
---|
364 | private double[][] m_perWordPerClass; |
---|
365 | private double[] m_wordsPerClass; |
---|
366 | int m_classIndex = -1; |
---|
367 | private double[] m_classDistribution; |
---|
368 | /** number of unique words */ |
---|
369 | private int m_numAttributes; |
---|
370 | //set the target class |
---|
371 | private int m_targetClass = -1; |
---|
372 | |
---|
373 | private double m_WordLaplace=1; |
---|
374 | |
---|
375 | private double[] m_coefficient; |
---|
376 | private double m_classRatio; |
---|
377 | private double m_wordRatio; |
---|
378 | |
---|
379 | public void initClassifier(Instances instances) throws Exception { |
---|
380 | m_numAttributes = instances.numAttributes(); |
---|
381 | m_perWordPerClass = new double[2][m_numAttributes]; |
---|
382 | m_coefficient = new double[m_numAttributes]; |
---|
383 | m_wordsPerClass = new double[2]; |
---|
384 | m_classDistribution = new double[2]; |
---|
385 | m_WordLaplace = Math.log(m_numAttributes); |
---|
386 | m_classIndex = instances.classIndex(); |
---|
387 | |
---|
388 | //Laplace |
---|
389 | for (int c = 0; c < 2; c++) { |
---|
390 | m_classDistribution[c] = 1; |
---|
391 | m_wordsPerClass[c] = m_WordLaplace * m_numAttributes; |
---|
392 | java.util.Arrays.fill(m_perWordPerClass[c], m_WordLaplace); |
---|
393 | } |
---|
394 | |
---|
395 | } |
---|
396 | |
---|
397 | public void updateClassifier(Instance ins) throws |
---|
398 | Exception { |
---|
399 | //c=0 is 1, which is the target class, and c=1 is the rest |
---|
400 | int classIndex = 0; |
---|
401 | if (ins.value(ins.classIndex()) != m_targetClass) |
---|
402 | classIndex = 1; |
---|
403 | double prob = 1 - |
---|
404 | distributionForInstance(ins)[classIndex]; |
---|
405 | |
---|
406 | |
---|
407 | double weight = prob * ins.weight(); |
---|
408 | |
---|
409 | for (int a = 0; a < ins.numValues(); a++) { |
---|
410 | if (ins.index(a) != m_classIndex ) |
---|
411 | { |
---|
412 | |
---|
413 | if (m_BinaryWord) { |
---|
414 | if (ins.valueSparse(a) > 0) { |
---|
415 | m_wordsPerClass[classIndex] += |
---|
416 | weight; |
---|
417 | m_perWordPerClass[classIndex][ins. |
---|
418 | index(a)] += |
---|
419 | weight; |
---|
420 | } |
---|
421 | } else { |
---|
422 | double t = ins.valueSparse(a) * weight; |
---|
423 | m_wordsPerClass[classIndex] += t; |
---|
424 | m_perWordPerClass[classIndex][ins.index(a)] += t; |
---|
425 | } |
---|
426 | //update coefficient |
---|
427 | m_coefficient[ins.index(a)] = Math.log(m_perWordPerClass[0][ |
---|
428 | ins.index(a)] / |
---|
429 | m_perWordPerClass[1][ins.index(a)]); |
---|
430 | } |
---|
431 | } |
---|
432 | m_wordRatio = Math.log(m_wordsPerClass[0] / m_wordsPerClass[1]); |
---|
433 | m_classDistribution[classIndex] += weight; |
---|
434 | m_classRatio = Math.log(m_classDistribution[0] / |
---|
435 | m_classDistribution[1]); |
---|
436 | } |
---|
437 | |
---|
438 | |
---|
439 | /** |
---|
440 | * Calculates the class membership probabilities for the given test |
---|
441 | * instance. |
---|
442 | * |
---|
443 | * @param ins the instance to be classified |
---|
444 | * @return predicted class probability distribution |
---|
445 | * @exception Exception if there is a problem generating the prediction |
---|
446 | */ |
---|
447 | public double getLogProbForTargetClass(Instance ins) throws Exception { |
---|
448 | |
---|
449 | double probLog = m_classRatio; |
---|
450 | for (int a = 0; a < ins.numValues(); a++) { |
---|
451 | if (ins.index(a) != m_classIndex ) |
---|
452 | { |
---|
453 | |
---|
454 | if (m_BinaryWord) { |
---|
455 | if (ins.valueSparse(a) > 0) { |
---|
456 | probLog += m_coefficient[ins.index(a)] - |
---|
457 | m_wordRatio; |
---|
458 | } |
---|
459 | } else { |
---|
460 | probLog += ins.valueSparse(a) * |
---|
461 | (m_coefficient[ins.index(a)] - m_wordRatio); |
---|
462 | } |
---|
463 | } |
---|
464 | } |
---|
465 | return probLog; |
---|
466 | } |
---|
467 | |
---|
468 | /** |
---|
469 | * Calculates the class membership probabilities for the given test |
---|
470 | * instance. |
---|
471 | * |
---|
472 | * @param instance the instance to be classified |
---|
473 | * @return predicted class probability distribution |
---|
474 | * @exception Exception if there is a problem generating the prediction |
---|
475 | */ |
---|
476 | public double[] distributionForInstance(Instance instance) throws |
---|
477 | Exception { |
---|
478 | double[] probOfClassGivenDoc = new double[2]; |
---|
479 | double ratio=getLogProbForTargetClass(instance); |
---|
480 | if (ratio > 709) |
---|
481 | probOfClassGivenDoc[0]=1; |
---|
482 | else |
---|
483 | { |
---|
484 | ratio = Math.exp(ratio); |
---|
485 | probOfClassGivenDoc[0]=ratio / (1 + ratio); |
---|
486 | } |
---|
487 | |
---|
488 | probOfClassGivenDoc[1] = 1 - probOfClassGivenDoc[0]; |
---|
489 | return probOfClassGivenDoc; |
---|
490 | } |
---|
491 | |
---|
492 | /** |
---|
493 | * Returns a string representation of the classifier. |
---|
494 | * |
---|
495 | * @return a string representation of the classifier |
---|
496 | */ |
---|
497 | public String toString() { |
---|
498 | // StringBuffer result = new StringBuffer("The cofficiency of a naive Bayes classifier, can be considered as the discriminative power of a word\n--------------------------------------\n"); |
---|
499 | StringBuffer result = new StringBuffer(); |
---|
500 | |
---|
501 | result.append("\n"); |
---|
502 | TreeMap sort=new TreeMap(); |
---|
503 | double[] absCoeff=new double[m_numAttributes]; |
---|
504 | for(int w = 0; w<m_numAttributes; w++) |
---|
505 | { |
---|
506 | if(w==m_headerInfo.classIndex())continue; |
---|
507 | String val= m_headerInfo.attribute(w).name()+": "+m_coefficient[w]; |
---|
508 | sort.put((-1)*Math.abs(m_coefficient[w]),val); |
---|
509 | } |
---|
510 | Iterator it=sort.values().iterator(); |
---|
511 | while(it.hasNext()) |
---|
512 | { |
---|
513 | result.append((String)it.next()); |
---|
514 | result.append("\n"); |
---|
515 | } |
---|
516 | |
---|
517 | return result.toString(); |
---|
518 | } |
---|
519 | |
---|
520 | /** |
---|
521 | * Sets the Target Class |
---|
522 | */ |
---|
523 | public void setTargetClass(int targetClass) { |
---|
524 | |
---|
525 | m_targetClass = targetClass; |
---|
526 | } |
---|
527 | |
---|
528 | /** |
---|
529 | * Gets the Target Class |
---|
530 | * |
---|
531 | * @return the Target Class Index |
---|
532 | */ |
---|
533 | public int getTargetClass() { |
---|
534 | |
---|
535 | return m_targetClass; |
---|
536 | } |
---|
537 | |
---|
538 | } |
---|
539 | |
---|
540 | |
---|
541 | /** |
---|
542 | * Main method for testing this class. |
---|
543 | * |
---|
544 | * @param argv the options |
---|
545 | */ |
---|
546 | public static void main(String[] argv) { |
---|
547 | |
---|
548 | DMNBtext c = new DMNBtext(); |
---|
549 | |
---|
550 | runClassifier(c, argv); |
---|
551 | } |
---|
552 | } |
---|
553 | |
---|