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 | * sIB.java |
---|
19 | * Copyright (C) 2008 University of Waikato, Hamilton, New Zealand |
---|
20 | * |
---|
21 | */ |
---|
22 | |
---|
23 | package weka.clusterers; |
---|
24 | |
---|
25 | import weka.core.Capabilities; |
---|
26 | import weka.core.Instance; |
---|
27 | import weka.core.DenseInstance; |
---|
28 | import weka.core.Instances; |
---|
29 | import weka.core.Option; |
---|
30 | import weka.core.RevisionHandler; |
---|
31 | import weka.core.RevisionUtils; |
---|
32 | import weka.core.TechnicalInformation; |
---|
33 | import weka.core.TechnicalInformationHandler; |
---|
34 | import weka.core.Utils; |
---|
35 | import weka.core.Capabilities.Capability; |
---|
36 | import weka.core.TechnicalInformation.Field; |
---|
37 | import weka.core.TechnicalInformation.Type; |
---|
38 | import weka.core.matrix.Matrix; |
---|
39 | import weka.filters.unsupervised.attribute.ReplaceMissingValues; |
---|
40 | |
---|
41 | import java.io.Serializable; |
---|
42 | import java.util.ArrayList; |
---|
43 | import java.util.Enumeration; |
---|
44 | import java.util.Random; |
---|
45 | import java.util.Vector; |
---|
46 | |
---|
47 | /** |
---|
48 | <!-- globalinfo-start --> |
---|
49 | * Cluster data using the sequential information bottleneck algorithm.<br/> |
---|
50 | * <br/> |
---|
51 | * Note: only hard clustering scheme is supported. sIB assign for each instance the cluster that have the minimum cost/distance to the instance. The trade-off beta is set to infinite so 1/beta is zero.<br/> |
---|
52 | * <br/> |
---|
53 | * For more information, see:<br/> |
---|
54 | * <br/> |
---|
55 | * Noam Slonim, Nir Friedman, Naftali Tishby: Unsupervised document classification using sequential information maximization. In: Proceedings of the 25th International ACM SIGIR Conference on Research and Development in Information Retrieval, 129-136, 2002. |
---|
56 | * <p/> |
---|
57 | <!-- globalinfo-end --> |
---|
58 | * |
---|
59 | <!-- technical-bibtex-start --> |
---|
60 | * BibTeX: |
---|
61 | * <pre> |
---|
62 | * @inproceedings{Slonim2002, |
---|
63 | * author = {Noam Slonim and Nir Friedman and Naftali Tishby}, |
---|
64 | * booktitle = {Proceedings of the 25th International ACM SIGIR Conference on Research and Development in Information Retrieval}, |
---|
65 | * pages = {129-136}, |
---|
66 | * title = {Unsupervised document classification using sequential information maximization}, |
---|
67 | * year = {2002} |
---|
68 | * } |
---|
69 | * </pre> |
---|
70 | * <p/> |
---|
71 | <!-- technical-bibtex-end --> |
---|
72 | * |
---|
73 | <!-- options-start --> |
---|
74 | * Valid options are: <p/> |
---|
75 | * |
---|
76 | * <pre> -I <num> |
---|
77 | * maximum number of iterations |
---|
78 | * (default 100).</pre> |
---|
79 | * |
---|
80 | * <pre> -M <num> |
---|
81 | * minimum number of changes in a single iteration |
---|
82 | * (default 0).</pre> |
---|
83 | * |
---|
84 | * <pre> -N <num> |
---|
85 | * number of clusters. |
---|
86 | * (default 2).</pre> |
---|
87 | * |
---|
88 | * <pre> -R <num> |
---|
89 | * number of restarts. |
---|
90 | * (default 5).</pre> |
---|
91 | * |
---|
92 | * <pre> -U |
---|
93 | * set not to normalize the data |
---|
94 | * (default true).</pre> |
---|
95 | * |
---|
96 | * <pre> -V |
---|
97 | * set to output debug info |
---|
98 | * (default false).</pre> |
---|
99 | * |
---|
100 | * <pre> -S <num> |
---|
101 | * Random number seed. |
---|
102 | * (default 1)</pre> |
---|
103 | * |
---|
104 | <!-- options-end --> |
---|
105 | * |
---|
106 | * @author Noam Slonim |
---|
107 | * @author <a href="mailto:lh92@cs.waikato.ac.nz">Anna Huang</a> |
---|
108 | * @version $Revision: 5987 $ |
---|
109 | */ |
---|
110 | public class sIB |
---|
111 | extends RandomizableClusterer |
---|
112 | implements TechnicalInformationHandler { |
---|
113 | |
---|
114 | /** for serialization. */ |
---|
115 | private static final long serialVersionUID = -8652125897352654213L; |
---|
116 | |
---|
117 | /** |
---|
118 | * Inner class handling status of the input data |
---|
119 | * |
---|
120 | * @see Serializable |
---|
121 | */ |
---|
122 | private class Input |
---|
123 | implements Serializable, RevisionHandler { |
---|
124 | |
---|
125 | /** for serialization */ |
---|
126 | static final long serialVersionUID = -2464453171263384037L; |
---|
127 | |
---|
128 | /** Prior probability of each instance */ |
---|
129 | private double[] Px; |
---|
130 | |
---|
131 | /** Prior probability of each attribute */ |
---|
132 | private double[] Py; |
---|
133 | |
---|
134 | /** Joint distribution of attribute and instance */ |
---|
135 | private Matrix Pyx; |
---|
136 | |
---|
137 | /** P[y|x] */ |
---|
138 | private Matrix Py_x; |
---|
139 | |
---|
140 | /** Mutual information between the instances and the attributes */ |
---|
141 | private double Ixy; |
---|
142 | |
---|
143 | /** Entropy of the attributes */ |
---|
144 | private double Hy; |
---|
145 | |
---|
146 | /** Entropy of the instances */ |
---|
147 | private double Hx; |
---|
148 | |
---|
149 | /** Sum values of the dataset */ |
---|
150 | private double sumVals; |
---|
151 | |
---|
152 | /** |
---|
153 | * Returns the revision string. |
---|
154 | * |
---|
155 | * @return the revision |
---|
156 | */ |
---|
157 | public String getRevision() { |
---|
158 | return RevisionUtils.extract("$Revision: 5987 $"); |
---|
159 | } |
---|
160 | } |
---|
161 | |
---|
162 | /** |
---|
163 | * Internal class handling the whole partition |
---|
164 | * |
---|
165 | * @see Serializable |
---|
166 | */ |
---|
167 | private class Partition |
---|
168 | implements Serializable, RevisionHandler { |
---|
169 | |
---|
170 | /** for serialization */ |
---|
171 | static final long serialVersionUID = 4957194978951259946L; |
---|
172 | |
---|
173 | /** Cluster assignment for each instance */ |
---|
174 | private int[] Pt_x; |
---|
175 | |
---|
176 | /** Prior probability of each cluster */ |
---|
177 | private double[] Pt; |
---|
178 | |
---|
179 | /** sIB equation score, to evaluate the quality of the partition */ |
---|
180 | private double L; |
---|
181 | |
---|
182 | /** Number of changes during the generation of this partition */ |
---|
183 | private int counter; |
---|
184 | |
---|
185 | /** Attribute probablities for each cluster */ |
---|
186 | private Matrix Py_t; |
---|
187 | |
---|
188 | /** |
---|
189 | * Create a new empty <code>Partition</code> instance. |
---|
190 | */ |
---|
191 | public Partition() { |
---|
192 | Pt_x = new int[m_numInstances]; |
---|
193 | for (int i = 0; i < m_numInstances; i++) { |
---|
194 | Pt_x[i] = -1; |
---|
195 | } |
---|
196 | Pt = new double[m_numCluster]; |
---|
197 | Py_t = new Matrix(m_numAttributes, m_numCluster); |
---|
198 | counter = 0; |
---|
199 | } |
---|
200 | |
---|
201 | /** |
---|
202 | * Find all the instances that have been assigned to cluster i |
---|
203 | * @param i index of the cluster |
---|
204 | * @return an arraylist of the instance ids that have been assigned to cluster i |
---|
205 | */ |
---|
206 | private ArrayList<Integer> find(int i) { |
---|
207 | ArrayList<Integer> indices = new ArrayList<Integer>(); |
---|
208 | for (int x = 0; x < Pt_x.length; x++) { |
---|
209 | if (Pt_x[x] == i) { |
---|
210 | indices.add(x); |
---|
211 | } |
---|
212 | } |
---|
213 | return indices; |
---|
214 | } |
---|
215 | |
---|
216 | /** |
---|
217 | * Find the size of the cluster i |
---|
218 | * @param i index of the cluster |
---|
219 | * @return the size of cluster i |
---|
220 | */ |
---|
221 | private int size(int i) { |
---|
222 | int count = 0; |
---|
223 | for (int x = 0; x < Pt_x.length; x++) { |
---|
224 | if (Pt_x[x] == i) { |
---|
225 | count++; |
---|
226 | } |
---|
227 | } |
---|
228 | return count; |
---|
229 | } |
---|
230 | |
---|
231 | /** |
---|
232 | * Copy the current partition into T |
---|
233 | * @param T the target partition object |
---|
234 | */ |
---|
235 | private void copy(Partition T) { |
---|
236 | if (T == null) { |
---|
237 | T = new Partition(); |
---|
238 | } |
---|
239 | System.arraycopy(Pt_x, 0, T.Pt_x, 0, Pt_x.length); |
---|
240 | System.arraycopy(Pt, 0, T.Pt, 0, Pt.length); |
---|
241 | T.L = L; |
---|
242 | T.counter = counter; |
---|
243 | |
---|
244 | double[][] mArray = Py_t.getArray(); |
---|
245 | double[][] tgtArray = T.Py_t.getArray(); |
---|
246 | for (int i = 0; i < mArray.length; i++) { |
---|
247 | System.arraycopy(mArray[i], 0, tgtArray[i], 0, mArray[0].length); |
---|
248 | } |
---|
249 | } |
---|
250 | |
---|
251 | /** |
---|
252 | * Output the current partition |
---|
253 | * @param insts |
---|
254 | * @return a string that describes the partition |
---|
255 | */ |
---|
256 | public String toString() { |
---|
257 | StringBuffer text = new StringBuffer(); |
---|
258 | text.append("score (L) : " + Utils.doubleToString(L, 4) + "\n"); |
---|
259 | text.append("number of changes : " + counter +"\n"); |
---|
260 | for (int i = 0; i < m_numCluster; i++) { |
---|
261 | text.append("\nCluster "+i+"\n"); |
---|
262 | text.append("size : "+size(i)+"\n"); |
---|
263 | text.append("prior prob : "+Utils.doubleToString(Pt[i], 4)+"\n"); |
---|
264 | } |
---|
265 | return text.toString(); |
---|
266 | } |
---|
267 | |
---|
268 | /** |
---|
269 | * Returns the revision string. |
---|
270 | * |
---|
271 | * @return the revision |
---|
272 | */ |
---|
273 | public String getRevision() { |
---|
274 | return RevisionUtils.extract("$Revision: 5987 $"); |
---|
275 | } |
---|
276 | } |
---|
277 | |
---|
278 | /** Training data */ |
---|
279 | private Instances m_data; |
---|
280 | |
---|
281 | /** Number of clusters */ |
---|
282 | private int m_numCluster = 2; |
---|
283 | |
---|
284 | /** Number of restarts */ |
---|
285 | private int m_numRestarts = 5; |
---|
286 | |
---|
287 | /** Verbose? */ |
---|
288 | private boolean m_verbose = false; |
---|
289 | |
---|
290 | /** Uniform prior probability of the documents */ |
---|
291 | private boolean m_uniformPrior = true; |
---|
292 | |
---|
293 | /** Max number of iterations during each restart */ |
---|
294 | private int m_maxLoop = 100; |
---|
295 | |
---|
296 | /** Minimum number of changes */ |
---|
297 | private int m_minChange = 0; |
---|
298 | |
---|
299 | /** Globally replace missing values */ |
---|
300 | private ReplaceMissingValues m_replaceMissing; |
---|
301 | |
---|
302 | /** Number of instances */ |
---|
303 | private int m_numInstances; |
---|
304 | |
---|
305 | /** Number of attributes */ |
---|
306 | private int m_numAttributes; |
---|
307 | |
---|
308 | /** Randomly generate initial partition */ |
---|
309 | private Random random; |
---|
310 | |
---|
311 | /** Holds the best partition built */ |
---|
312 | private Partition bestT; |
---|
313 | |
---|
314 | /** Holds the statistics about the input dataset */ |
---|
315 | private Input input; |
---|
316 | |
---|
317 | /** |
---|
318 | * Generates a clusterer. |
---|
319 | * |
---|
320 | * @param data the training instances |
---|
321 | * @throws Exception if something goes wrong |
---|
322 | */ |
---|
323 | public void buildClusterer(Instances data) throws Exception { |
---|
324 | // can clusterer handle the data ? |
---|
325 | getCapabilities().testWithFail(data); |
---|
326 | |
---|
327 | m_replaceMissing = new ReplaceMissingValues(); |
---|
328 | Instances instances = new Instances(data); |
---|
329 | instances.setClassIndex(-1); |
---|
330 | m_replaceMissing.setInputFormat(instances); |
---|
331 | data = weka.filters.Filter.useFilter(instances, m_replaceMissing); |
---|
332 | instances = null; |
---|
333 | |
---|
334 | // initialize all fields that are not being set via options |
---|
335 | m_data = data; |
---|
336 | m_numInstances = m_data.numInstances(); |
---|
337 | m_numAttributes = m_data.numAttributes(); |
---|
338 | random = new Random(getSeed()); |
---|
339 | |
---|
340 | // initialize the statistics of the input training data |
---|
341 | input = sIB_ProcessInput(); |
---|
342 | |
---|
343 | // object to hold the best partition |
---|
344 | bestT = new Partition(); |
---|
345 | |
---|
346 | // the real clustering |
---|
347 | double bestL = Double.NEGATIVE_INFINITY; |
---|
348 | for (int k = 0; k < m_numRestarts; k++) { |
---|
349 | if(m_verbose) { |
---|
350 | System.out.format("restart number %s...\n", k); |
---|
351 | } |
---|
352 | |
---|
353 | // initialize the partition and optimize it |
---|
354 | Partition tmpT = sIB_InitT(input); |
---|
355 | tmpT = sIB_OptimizeT(tmpT, input); |
---|
356 | |
---|
357 | // if a better partition is found, save it |
---|
358 | if (tmpT.L > bestL) { |
---|
359 | tmpT.copy(bestT); |
---|
360 | bestL = bestT.L; |
---|
361 | } |
---|
362 | |
---|
363 | if(m_verbose) { |
---|
364 | System.out.println("\nPartition status : "); |
---|
365 | System.out.println("------------------"); |
---|
366 | System.out.println(tmpT.toString()+"\n"); |
---|
367 | } |
---|
368 | } |
---|
369 | |
---|
370 | if(m_verbose){ |
---|
371 | System.out.println("\nBest Partition"); |
---|
372 | System.out.println("==============="); |
---|
373 | System.out.println(bestT.toString()); |
---|
374 | } |
---|
375 | |
---|
376 | // save memory |
---|
377 | m_data = new Instances(m_data, 0); |
---|
378 | } |
---|
379 | |
---|
380 | /** |
---|
381 | * Cluster a given instance, this is the method defined in Clusterer |
---|
382 | * interface do nothing but just return the cluster assigned to it |
---|
383 | */ |
---|
384 | public int clusterInstance(Instance instance) throws Exception { |
---|
385 | double prior = (double) 1 / input.sumVals; |
---|
386 | double[] distances = new double[m_numCluster]; |
---|
387 | for(int i = 0; i < m_numCluster; i++){ |
---|
388 | double Pnew = bestT.Pt[i] + prior; |
---|
389 | double pi1 = prior / Pnew; |
---|
390 | double pi2 = bestT.Pt[i] / Pnew; |
---|
391 | distances[i] = Pnew * JS(instance, i, pi1, pi2); |
---|
392 | } |
---|
393 | return Utils.minIndex(distances); |
---|
394 | } |
---|
395 | |
---|
396 | /** |
---|
397 | * Process the input and compute the statistics of the training data |
---|
398 | * @return an Input object which holds the statistics about the training data |
---|
399 | */ |
---|
400 | private Input sIB_ProcessInput() { |
---|
401 | double valSum = 0.0; |
---|
402 | for (int i = 0; i < m_numInstances; i++) { |
---|
403 | valSum = 0.0; |
---|
404 | for (int v = 0; v < m_data.instance(i).numValues(); v++) { |
---|
405 | valSum += m_data.instance(i).valueSparse(v); |
---|
406 | } |
---|
407 | if (valSum <= 0) { |
---|
408 | if(m_verbose){ |
---|
409 | System.out.format("Instance %s sum of value = %s <= 0, removed.\n", i, valSum); |
---|
410 | } |
---|
411 | m_data.delete(i); |
---|
412 | m_numInstances--; |
---|
413 | } |
---|
414 | } |
---|
415 | |
---|
416 | // get the term-document matrix |
---|
417 | Input input = new Input(); |
---|
418 | input.Py_x = getTransposedNormedMatrix(m_data); |
---|
419 | if (m_uniformPrior) { |
---|
420 | input.Pyx = input.Py_x.copy(); |
---|
421 | normalizePrior(m_data); |
---|
422 | } |
---|
423 | else { |
---|
424 | input.Pyx = getTransposedMatrix(m_data); |
---|
425 | } |
---|
426 | input.sumVals = getTotalSum(m_data); |
---|
427 | input.Pyx.timesEquals((double) 1 / input.sumVals); |
---|
428 | |
---|
429 | // prior probability of documents, ie. sum the columns from the Pyx matrix |
---|
430 | input.Px = new double[m_numInstances]; |
---|
431 | for (int i = 0; i < m_numInstances; i++) { |
---|
432 | for (int j = 0; j < m_numAttributes; j++) { |
---|
433 | input.Px[i] += input.Pyx.get(j, i); |
---|
434 | } |
---|
435 | } |
---|
436 | |
---|
437 | // prior probability of terms, ie. sum the rows from the Pyx matrix |
---|
438 | input.Py = new double[m_numAttributes]; |
---|
439 | for (int i = 0; i < input.Pyx.getRowDimension(); i++) { |
---|
440 | for (int j = 0; j < input.Pyx.getColumnDimension(); j++) { |
---|
441 | input.Py[i] += input.Pyx.get(i, j); |
---|
442 | } |
---|
443 | } |
---|
444 | |
---|
445 | MI(input.Pyx, input); |
---|
446 | return input; |
---|
447 | } |
---|
448 | |
---|
449 | /** |
---|
450 | * Initialize the partition |
---|
451 | * @param input object holding the statistics of the training data |
---|
452 | * @return the initialized partition |
---|
453 | */ |
---|
454 | private Partition sIB_InitT(Input input) { |
---|
455 | Partition T = new Partition(); |
---|
456 | int avgSize = (int) Math.ceil((double) m_numInstances / m_numCluster); |
---|
457 | |
---|
458 | ArrayList<Integer> permInstsIdx = new ArrayList<Integer>(); |
---|
459 | ArrayList<Integer> unassigned = new ArrayList<Integer>(); |
---|
460 | for (int i = 0; i < m_numInstances; i++) { |
---|
461 | unassigned.add(i); |
---|
462 | } |
---|
463 | while (unassigned.size() != 0) { |
---|
464 | int t = random.nextInt(unassigned.size()); |
---|
465 | permInstsIdx.add(unassigned.get(t)); |
---|
466 | unassigned.remove(t); |
---|
467 | } |
---|
468 | |
---|
469 | for (int i = 0; i < m_numCluster; i++) { |
---|
470 | int r2 = avgSize > permInstsIdx.size() ? permInstsIdx.size() : avgSize; |
---|
471 | for (int j = 0; j < r2; j++) { |
---|
472 | T.Pt_x[permInstsIdx.get(j)] = i; |
---|
473 | } |
---|
474 | for (int j = 0; j < r2; j++) { |
---|
475 | permInstsIdx.remove(0); |
---|
476 | } |
---|
477 | } |
---|
478 | |
---|
479 | // initialize the prior prob of each cluster, and the probability |
---|
480 | // for each attribute within the cluster |
---|
481 | for (int i = 0; i < m_numCluster; i++) { |
---|
482 | ArrayList<Integer> indices = T.find(i); |
---|
483 | for (int j = 0; j < indices.size(); j++) { |
---|
484 | T.Pt[i] += input.Px[indices.get(j)]; |
---|
485 | } |
---|
486 | double[][] mArray = input.Pyx.getArray(); |
---|
487 | for (int j = 0; j < m_numAttributes; j++) { |
---|
488 | double sum = 0.0; |
---|
489 | for (int k = 0; k < indices.size(); k++) { |
---|
490 | sum += mArray[j][indices.get(k)]; |
---|
491 | } |
---|
492 | sum /= T.Pt[i]; |
---|
493 | T.Py_t.set(j, i, sum); |
---|
494 | } |
---|
495 | } |
---|
496 | |
---|
497 | if(m_verbose) { |
---|
498 | System.out.println("Initializing..."); |
---|
499 | } |
---|
500 | return T; |
---|
501 | } |
---|
502 | |
---|
503 | /** |
---|
504 | * Optimize the partition |
---|
505 | * @param tmpT partition to be optimized |
---|
506 | * @param input object describing the statistics of the training dataset |
---|
507 | * @return the optimized partition |
---|
508 | */ |
---|
509 | private Partition sIB_OptimizeT(Partition tmpT, Input input) { |
---|
510 | boolean done = false; |
---|
511 | int change = 0, loopCounter = 0; |
---|
512 | if(m_verbose) { |
---|
513 | System.out.println("Optimizing..."); |
---|
514 | System.out.println("-------------"); |
---|
515 | } |
---|
516 | while (!done) { |
---|
517 | change = 0; |
---|
518 | for (int i = 0; i < m_numInstances; i++) { |
---|
519 | int old_t = tmpT.Pt_x[i]; |
---|
520 | // If the current cluster only has one instance left, leave it. |
---|
521 | if (tmpT.size(old_t) == 1) { |
---|
522 | if(m_verbose){ |
---|
523 | System.out.format("cluster %s has only 1 doc remain\n", old_t); |
---|
524 | } |
---|
525 | continue; |
---|
526 | } |
---|
527 | // draw the instance out from its previous cluster |
---|
528 | reduce_x(i, old_t, tmpT, input); |
---|
529 | |
---|
530 | // re-cluster the instance |
---|
531 | int new_t = clusterInstance(i, input, tmpT); |
---|
532 | if (new_t != old_t) { |
---|
533 | change++; |
---|
534 | updateAssignment(i, new_t, tmpT, input.Px[i], input.Py_x); |
---|
535 | } |
---|
536 | } |
---|
537 | |
---|
538 | tmpT.counter += change; |
---|
539 | if(m_verbose){ |
---|
540 | System.out.format("iteration %s , changes : %s\n", loopCounter, change); |
---|
541 | } |
---|
542 | done = checkConvergence(change, loopCounter); |
---|
543 | loopCounter++; |
---|
544 | } |
---|
545 | |
---|
546 | // compute the sIB score |
---|
547 | tmpT.L = sIB_local_MI(tmpT.Py_t, tmpT.Pt); |
---|
548 | if(m_verbose){ |
---|
549 | System.out.format("score (L) : %s \n", Utils.doubleToString(tmpT.L, 4)); |
---|
550 | } |
---|
551 | return tmpT; |
---|
552 | } |
---|
553 | |
---|
554 | /** |
---|
555 | * Draw a instance out from a cluster. |
---|
556 | * @param instIdx index of the instance to be drawn out |
---|
557 | * @param t index of the cluster which the instance previously belong to |
---|
558 | * @param T the current working partition |
---|
559 | * @param input the input statistics |
---|
560 | */ |
---|
561 | private void reduce_x(int instIdx, int t, Partition T, Input input) { |
---|
562 | // Update the prior probability of the cluster |
---|
563 | ArrayList<Integer> indices = T.find(t); |
---|
564 | double sum = 0.0; |
---|
565 | for (int i = 0; i < indices.size(); i++) { |
---|
566 | if (indices.get(i) == instIdx) |
---|
567 | continue; |
---|
568 | sum += input.Px[indices.get(i)]; |
---|
569 | } |
---|
570 | T.Pt[t] = sum; |
---|
571 | |
---|
572 | if (T.Pt[t] < 0) { |
---|
573 | System.out.format("Warning: probability < 0 (%s)\n", T.Pt[t]); |
---|
574 | T.Pt[t] = 0; |
---|
575 | } |
---|
576 | |
---|
577 | // Update prob of each attribute in the cluster |
---|
578 | double[][] mArray = input.Pyx.getArray(); |
---|
579 | for (int i = 0; i < m_numAttributes; i++) { |
---|
580 | sum = 0.0; |
---|
581 | for (int j = 0; j < indices.size(); j++) { |
---|
582 | if (indices.get(j) == instIdx) |
---|
583 | continue; |
---|
584 | sum += mArray[i][indices.get(j)]; |
---|
585 | } |
---|
586 | T.Py_t.set(i, t, sum / T.Pt[t]); |
---|
587 | } |
---|
588 | } |
---|
589 | |
---|
590 | /** |
---|
591 | * Put an instance into a new cluster and update. |
---|
592 | * @param instIdx instance to be updated |
---|
593 | * @param newt index of the new cluster this instance has been assigned to |
---|
594 | * @param T the current working partition |
---|
595 | * @param Px an array of prior probabilities of the instances |
---|
596 | */ |
---|
597 | private void updateAssignment(int instIdx, int newt, Partition T, double Px, Matrix Py_x) { |
---|
598 | T.Pt_x[instIdx] = newt; |
---|
599 | |
---|
600 | // update probability of attributes in the cluster |
---|
601 | double mass = Px + T.Pt[newt]; |
---|
602 | double pi1 = Px / mass; |
---|
603 | double pi2 = T.Pt[newt] / mass; |
---|
604 | for (int i = 0; i < m_numAttributes; i++) { |
---|
605 | T.Py_t.set(i, newt, pi1 * Py_x.get(i, instIdx) + pi2 * T.Py_t.get(i, newt)); |
---|
606 | } |
---|
607 | |
---|
608 | T.Pt[newt] = mass; |
---|
609 | } |
---|
610 | |
---|
611 | /** |
---|
612 | * Check whether the current iteration is converged |
---|
613 | * @param change number of changes in current iteration |
---|
614 | * @param loops number of iterations done |
---|
615 | * @return true if the iteration is converged, false otherwise |
---|
616 | */ |
---|
617 | private boolean checkConvergence(int change, int loops) { |
---|
618 | if (change <= m_minChange || loops >= m_maxLoop) { |
---|
619 | if(m_verbose){ |
---|
620 | System.out.format("\nsIB converged after %s iterations with %s changes\n", loops, |
---|
621 | change); |
---|
622 | } |
---|
623 | return true; |
---|
624 | } |
---|
625 | return false; |
---|
626 | } |
---|
627 | |
---|
628 | /** |
---|
629 | * Cluster an instance into the nearest cluster. |
---|
630 | * @param instIdx Index of the instance to be clustered |
---|
631 | * @param input Object which describe the statistics of the training dataset |
---|
632 | * @param T Partition |
---|
633 | * @return index of the cluster that has the minimum distance to the instance |
---|
634 | */ |
---|
635 | private int clusterInstance(int instIdx, Input input, Partition T) { |
---|
636 | double[] distances = new double[m_numCluster]; |
---|
637 | for (int i = 0; i < m_numCluster; i++) { |
---|
638 | double Pnew = input.Px[instIdx] + T.Pt[i]; |
---|
639 | double pi1 = input.Px[instIdx] / Pnew; |
---|
640 | double pi2 = T.Pt[i] / Pnew; |
---|
641 | distances[i] = Pnew * JS(instIdx, input, T, i, pi1, pi2); |
---|
642 | } |
---|
643 | return Utils.minIndex(distances); |
---|
644 | } |
---|
645 | |
---|
646 | /** |
---|
647 | * Compute the JS divergence between an instance and a cluster, used for training data |
---|
648 | * @param instIdx index of the instance |
---|
649 | * @param input statistics of the input data |
---|
650 | * @param T the whole partition |
---|
651 | * @param t index of the cluster |
---|
652 | * @param pi1 |
---|
653 | * @param pi2 |
---|
654 | * @return the JS divergence |
---|
655 | */ |
---|
656 | private double JS(int instIdx, Input input, Partition T, int t, double pi1, double pi2) { |
---|
657 | if (Math.min(pi1, pi2) <= 0) { |
---|
658 | System.out.format("Warning: zero or negative weights in JS calculation! (pi1 %s, pi2 %s)\n", pi1, pi2); |
---|
659 | return 0; |
---|
660 | } |
---|
661 | Instance inst = m_data.instance(instIdx); |
---|
662 | double kl1 = 0.0, kl2 = 0.0, tmp = 0.0; |
---|
663 | for (int i = 0; i < inst.numValues(); i++) { |
---|
664 | tmp = input.Py_x.get(inst.index(i), instIdx); |
---|
665 | if(tmp != 0) { |
---|
666 | kl1 += tmp * Math.log(tmp / (tmp * pi1 + pi2 * T.Py_t.get(inst.index(i), t))); |
---|
667 | } |
---|
668 | } |
---|
669 | for (int i = 0; i < m_numAttributes; i++) { |
---|
670 | if ((tmp = T.Py_t.get(i, t)) != 0) { |
---|
671 | kl2 += tmp * Math.log(tmp / (input.Py_x.get(i, instIdx) * pi1 + pi2 * tmp)); |
---|
672 | } |
---|
673 | } |
---|
674 | return pi1 * kl1 + pi2 * kl2; |
---|
675 | } |
---|
676 | |
---|
677 | /** |
---|
678 | * Compute the JS divergence between an instance and a cluster, used for test data |
---|
679 | * @param inst instance to be clustered |
---|
680 | * @param t index of the cluster |
---|
681 | * @param pi1 |
---|
682 | * @param pi2 |
---|
683 | * @return the JS divergence |
---|
684 | */ |
---|
685 | private double JS(Instance inst, int t, double pi1, double pi2) { |
---|
686 | if (Math.min(pi1, pi2) <= 0) { |
---|
687 | System.out.format("Warning: zero or negative weights in JS calculation! (pi1 %s, pi2 %s)\n", pi1, pi2); |
---|
688 | return 0; |
---|
689 | } |
---|
690 | double sum = Utils.sum(inst.toDoubleArray()); |
---|
691 | double kl1 = 0.0, kl2 = 0.0, tmp = 0.0; |
---|
692 | for (int i = 0; i < inst.numValues(); i++) { |
---|
693 | tmp = inst.valueSparse(i) / sum; |
---|
694 | if(tmp != 0) { |
---|
695 | kl1 += tmp * Math.log(tmp / (tmp * pi1 + pi2 * bestT.Py_t.get(inst.index(i), t))); |
---|
696 | } |
---|
697 | } |
---|
698 | for (int i = 0; i < m_numAttributes; i++) { |
---|
699 | if ((tmp = bestT.Py_t.get(i, t)) != 0) { |
---|
700 | kl2 += tmp * Math.log(tmp / (inst.value(i) * pi1 / sum + pi2 * tmp)); |
---|
701 | } |
---|
702 | } |
---|
703 | return pi1 * kl1 + pi2 * kl2; |
---|
704 | } |
---|
705 | |
---|
706 | /** |
---|
707 | * Compute the sIB score |
---|
708 | * @param m a term-cluster matrix, with m[i, j] is the probability of term i given cluster j |
---|
709 | * @param Pt an array of cluster prior probabilities |
---|
710 | * @return the sIB score which indicates the quality of the partition |
---|
711 | */ |
---|
712 | private double sIB_local_MI(Matrix m, double[] Pt) { |
---|
713 | double Hy = 0.0, Ht = 0.0; |
---|
714 | for (int i = 0; i < Pt.length; i++) { |
---|
715 | Ht += Pt[i] * Math.log(Pt[i]); |
---|
716 | } |
---|
717 | Ht = -Ht; |
---|
718 | |
---|
719 | for (int i = 0; i < m_numAttributes; i++) { |
---|
720 | double Py = 0.0; |
---|
721 | for (int j = 0; j < m_numCluster; j++) { |
---|
722 | Py += m.get(i, j) * Pt[j]; |
---|
723 | } |
---|
724 | if(Py == 0) continue; |
---|
725 | Hy += Py * Math.log(Py); |
---|
726 | } |
---|
727 | Hy = -Hy; |
---|
728 | |
---|
729 | double Hyt = 0.0, tmp = 0.0; |
---|
730 | for (int i = 0; i < m.getRowDimension(); i++) { |
---|
731 | for (int j = 0; j < m.getColumnDimension(); j++) { |
---|
732 | if ((tmp = m.get(i, j)) == 0 || Pt[j] == 0) { |
---|
733 | continue; |
---|
734 | } |
---|
735 | tmp *= Pt[j]; |
---|
736 | Hyt += tmp * Math.log(tmp); |
---|
737 | } |
---|
738 | } |
---|
739 | return Hy + Ht + Hyt; |
---|
740 | } |
---|
741 | |
---|
742 | /** |
---|
743 | * Get the sum of value of the dataset |
---|
744 | * @param data set of instances to handle |
---|
745 | * @return sum of all the attribute values for all the instances in the dataset |
---|
746 | */ |
---|
747 | private double getTotalSum(Instances data) { |
---|
748 | double sum = 0.0; |
---|
749 | for (int i = 0; i < data.numInstances(); i++) { |
---|
750 | for (int v = 0; v < data.instance(i).numValues(); v++) { |
---|
751 | sum += data.instance(i).valueSparse(v); |
---|
752 | } |
---|
753 | } |
---|
754 | return sum; |
---|
755 | } |
---|
756 | |
---|
757 | /** |
---|
758 | * Transpose the document-term matrix to term-document matrix |
---|
759 | * @param data instances with document-term info |
---|
760 | * @return a term-document matrix transposed from the input dataset |
---|
761 | */ |
---|
762 | private Matrix getTransposedMatrix(Instances data) { |
---|
763 | double[][] temp = new double[data.numAttributes()][data.numInstances()]; |
---|
764 | for (int i = 0; i < data.numInstances(); i++) { |
---|
765 | Instance inst = data.instance(i); |
---|
766 | for (int v = 0; v < inst.numValues(); v++) { |
---|
767 | temp[inst.index(v)][i] = inst.valueSparse(v); |
---|
768 | } |
---|
769 | } |
---|
770 | Matrix My_x = new Matrix(temp); |
---|
771 | return My_x; |
---|
772 | } |
---|
773 | |
---|
774 | /** |
---|
775 | * Normalize the document vectors |
---|
776 | * @param data instances to be normalized |
---|
777 | */ |
---|
778 | private void normalizePrior(Instances data) { |
---|
779 | for (int i = 0; i < data.numInstances(); i++) { |
---|
780 | normalizeInstance(data.instance(i)); |
---|
781 | } |
---|
782 | } |
---|
783 | |
---|
784 | /** |
---|
785 | * Normalize the instance |
---|
786 | * @param inst instance to be normalized |
---|
787 | * @return a new Instance with normalized values |
---|
788 | */ |
---|
789 | private Instance normalizeInstance(Instance inst) { |
---|
790 | double[] vals = inst.toDoubleArray(); |
---|
791 | double sum = Utils.sum(vals); |
---|
792 | for(int i = 0; i < vals.length; i++) { |
---|
793 | vals[i] /= sum; |
---|
794 | } |
---|
795 | return new DenseInstance(inst.weight(), vals); |
---|
796 | } |
---|
797 | |
---|
798 | private Matrix getTransposedNormedMatrix(Instances data) { |
---|
799 | Matrix matrix = new Matrix(data.numAttributes(), data.numInstances()); |
---|
800 | for(int i = 0; i < data.numInstances(); i++){ |
---|
801 | double[] vals = data.instance(i).toDoubleArray(); |
---|
802 | double sum = Utils.sum(vals); |
---|
803 | for (int v = 0; v < vals.length; v++) { |
---|
804 | vals[v] /= sum; |
---|
805 | matrix.set(v, i, vals[v]); |
---|
806 | } |
---|
807 | } |
---|
808 | return matrix; |
---|
809 | } |
---|
810 | |
---|
811 | /** |
---|
812 | * Compute the MI between instances and attributes |
---|
813 | * @param m the term-document matrix |
---|
814 | * @param input object that describes the statistics about the training data |
---|
815 | */ |
---|
816 | private void MI(Matrix m, Input input){ |
---|
817 | int minDimSize = m.getColumnDimension() < m.getRowDimension() ? m.getColumnDimension() : m.getRowDimension(); |
---|
818 | if(minDimSize < 2){ |
---|
819 | System.err.println("Warning : This is not a JOINT distribution"); |
---|
820 | input.Hx = Entropy (m); |
---|
821 | input.Hy = 0; |
---|
822 | input.Ixy = 0; |
---|
823 | return; |
---|
824 | } |
---|
825 | |
---|
826 | input.Hx = Entropy(input.Px); |
---|
827 | input.Hy = Entropy(input.Py); |
---|
828 | |
---|
829 | double entropy = input.Hx + input.Hy; |
---|
830 | for (int i=0; i < m_numInstances; i++) { |
---|
831 | Instance inst = m_data.instance(i); |
---|
832 | for (int v = 0; v < inst.numValues(); v++) { |
---|
833 | double tmp = m.get(inst.index(v), i); |
---|
834 | if(tmp <= 0) continue; |
---|
835 | entropy += tmp * Math.log(tmp); |
---|
836 | } |
---|
837 | } |
---|
838 | input.Ixy = entropy; |
---|
839 | if(m_verbose) { |
---|
840 | System.out.println("Ixy = " + input.Ixy); |
---|
841 | } |
---|
842 | } |
---|
843 | |
---|
844 | /** |
---|
845 | * Compute the entropy score based on an array of probabilities |
---|
846 | * @param probs array of non-negative and normalized probabilities |
---|
847 | * @return the entropy value |
---|
848 | */ |
---|
849 | private double Entropy(double[] probs){ |
---|
850 | for (int i = 0; i < probs.length; i++){ |
---|
851 | if (probs[i] <= 0) { |
---|
852 | if(m_verbose) { |
---|
853 | System.out.println("Warning: Negative probability."); |
---|
854 | } |
---|
855 | return Double.NaN; |
---|
856 | } |
---|
857 | } |
---|
858 | // could be unormalized, when normalization is not specified |
---|
859 | if(Math.abs(Utils.sum(probs)-1) >= 1e-6) { |
---|
860 | if(m_verbose) { |
---|
861 | System.out.println("Warning: Not normalized."); |
---|
862 | } |
---|
863 | return Double.NaN; |
---|
864 | } |
---|
865 | |
---|
866 | double mi = 0.0; |
---|
867 | for (int i = 0; i < probs.length; i++) { |
---|
868 | mi += probs[i] * Math.log(probs[i]); |
---|
869 | } |
---|
870 | mi = -mi; |
---|
871 | return mi; |
---|
872 | } |
---|
873 | |
---|
874 | /** |
---|
875 | * Compute the entropy score based on a matrix |
---|
876 | * @param p a matrix with non-negative and normalized probabilities |
---|
877 | * @return the entropy value |
---|
878 | */ |
---|
879 | private double Entropy(Matrix p) { |
---|
880 | double mi = 0; |
---|
881 | for (int i = 0; i < p.getRowDimension(); i++) { |
---|
882 | for (int j = 0; j < p.getColumnDimension(); j++) { |
---|
883 | if(p.get(i, j) == 0){ |
---|
884 | continue; |
---|
885 | } |
---|
886 | mi += p.get(i, j) + Math.log(p.get(i, j)); |
---|
887 | } |
---|
888 | } |
---|
889 | mi = -mi; |
---|
890 | return mi; |
---|
891 | } |
---|
892 | |
---|
893 | /** |
---|
894 | * Parses a given list of options. <p/> |
---|
895 | * |
---|
896 | <!-- options-start --> |
---|
897 | * Valid options are: <p/> |
---|
898 | * |
---|
899 | * <pre> -I <num> |
---|
900 | * maximum number of iterations |
---|
901 | * (default 100).</pre> |
---|
902 | * |
---|
903 | * <pre> -M <num> |
---|
904 | * minimum number of changes in a single iteration |
---|
905 | * (default 0).</pre> |
---|
906 | * |
---|
907 | * <pre> -N <num> |
---|
908 | * number of clusters. |
---|
909 | * (default 2).</pre> |
---|
910 | * |
---|
911 | * <pre> -R <num> |
---|
912 | * number of restarts. |
---|
913 | * (default 5).</pre> |
---|
914 | * |
---|
915 | * <pre> -U |
---|
916 | * set not to normalize the data |
---|
917 | * (default true).</pre> |
---|
918 | * |
---|
919 | * <pre> -V |
---|
920 | * set to output debug info |
---|
921 | * (default false).</pre> |
---|
922 | * |
---|
923 | * <pre> -S <num> |
---|
924 | * Random number seed. |
---|
925 | * (default 1)</pre> |
---|
926 | * |
---|
927 | <!-- options-end --> |
---|
928 | * |
---|
929 | * @param options the list of options as an array of strings |
---|
930 | * @throws Exception if an option is not supported |
---|
931 | */ |
---|
932 | public void setOptions(String[] options) throws Exception { |
---|
933 | String optionString = Utils.getOption('I', options); |
---|
934 | if (optionString.length() != 0) { |
---|
935 | setMaxIterations(Integer.parseInt(optionString)); |
---|
936 | } |
---|
937 | optionString = Utils.getOption('M', options); |
---|
938 | if (optionString.length() != 0) { |
---|
939 | setMinChange((new Integer(optionString)).intValue()); |
---|
940 | } |
---|
941 | optionString = Utils.getOption('N', options); |
---|
942 | if (optionString.length() != 0) { |
---|
943 | setNumClusters(Integer.parseInt(optionString)); |
---|
944 | } |
---|
945 | optionString = Utils.getOption('R', options); |
---|
946 | if (optionString.length() != 0) { |
---|
947 | setNumRestarts((new Integer(optionString)).intValue()); |
---|
948 | } |
---|
949 | setNotUnifyNorm(Utils.getFlag('U', options)); |
---|
950 | setDebug(Utils.getFlag('V', options)); |
---|
951 | |
---|
952 | super.setOptions(options); |
---|
953 | } |
---|
954 | |
---|
955 | /** |
---|
956 | * Returns an enumeration describing the available options. |
---|
957 | * @return an enumeration of all the available options. |
---|
958 | */ |
---|
959 | public Enumeration listOptions() { |
---|
960 | Vector<Option> result = new Vector<Option>(); |
---|
961 | result.addElement(new Option("\tmaximum number of iterations\n" |
---|
962 | + "\t(default 100).", "I", 1, "-I <num>")); |
---|
963 | result.addElement(new Option( |
---|
964 | "\tminimum number of changes in a single iteration\n" |
---|
965 | + "\t(default 0).", "M", 1, "-M <num>")); |
---|
966 | result.addElement(new Option("\tnumber of clusters.\n" + "\t(default 2).", |
---|
967 | "N", 1, "-N <num>")); |
---|
968 | result.addElement(new Option("\tnumber of restarts.\n" |
---|
969 | + "\t(default 5).", "R", 1, "-R <num>")); |
---|
970 | result.addElement(new Option("\tset not to normalize the data\n" |
---|
971 | + "\t(default true).", "U", 0, "-U")); |
---|
972 | result.addElement(new Option("\tset to output debug info\n" |
---|
973 | + "\t(default false).", "V", 0, "-V")); |
---|
974 | |
---|
975 | Enumeration en = super.listOptions(); |
---|
976 | while (en.hasMoreElements()) |
---|
977 | result.addElement((Option) en.nextElement()); |
---|
978 | |
---|
979 | return result.elements(); |
---|
980 | } |
---|
981 | |
---|
982 | /** |
---|
983 | * Gets the current settings. |
---|
984 | * @return an array of strings suitable for passing to setOptions() |
---|
985 | */ |
---|
986 | public String[] getOptions() { |
---|
987 | Vector<String> result; |
---|
988 | result = new Vector<String>(); |
---|
989 | result.add("-I"); |
---|
990 | result.add("" + getMaxIterations()); |
---|
991 | result.add("-M"); |
---|
992 | result.add("" + getMinChange()); |
---|
993 | result.add("-N"); |
---|
994 | result.add("" + getNumClusters()); |
---|
995 | result.add("-R"); |
---|
996 | result.add("" + getNumRestarts()); |
---|
997 | if(getNotUnifyNorm()) { |
---|
998 | result.add("-U"); |
---|
999 | } |
---|
1000 | if(getDebug()) { |
---|
1001 | result.add("-V"); |
---|
1002 | } |
---|
1003 | |
---|
1004 | String[] options = super.getOptions(); |
---|
1005 | for (int i = 0; i < options.length; i++){ |
---|
1006 | result.add(options[i]); |
---|
1007 | } |
---|
1008 | return result.toArray(new String[result.size()]); |
---|
1009 | } |
---|
1010 | |
---|
1011 | /** |
---|
1012 | * Returns the tip text for this property |
---|
1013 | * @return tip text for this property suitable for |
---|
1014 | * displaying in the explorer/experimenter gui |
---|
1015 | */ |
---|
1016 | public String debugTipText() { |
---|
1017 | return "If set to true, clusterer may output additional info to " + |
---|
1018 | "the console."; |
---|
1019 | } |
---|
1020 | |
---|
1021 | /** |
---|
1022 | * Set debug mode - verbose output |
---|
1023 | * @param v true for verbose output |
---|
1024 | */ |
---|
1025 | public void setDebug (boolean v) { |
---|
1026 | m_verbose = v; |
---|
1027 | } |
---|
1028 | |
---|
1029 | /** |
---|
1030 | * Get debug mode |
---|
1031 | * @return true if debug mode is set |
---|
1032 | */ |
---|
1033 | public boolean getDebug () { |
---|
1034 | return m_verbose; |
---|
1035 | } |
---|
1036 | |
---|
1037 | /** |
---|
1038 | * Returns the tip text for this property. |
---|
1039 | * @return tip text for this property |
---|
1040 | */ |
---|
1041 | public String maxIterationsTipText() { |
---|
1042 | return "set maximum number of iterations (default 100)"; |
---|
1043 | } |
---|
1044 | |
---|
1045 | /** |
---|
1046 | * Set the max number of iterations |
---|
1047 | * @param i max number of iterations |
---|
1048 | */ |
---|
1049 | public void setMaxIterations(int i) { |
---|
1050 | m_maxLoop = i; |
---|
1051 | } |
---|
1052 | |
---|
1053 | /** |
---|
1054 | * Get the max number of iterations |
---|
1055 | * @return max number of iterations |
---|
1056 | */ |
---|
1057 | public int getMaxIterations() { |
---|
1058 | return m_maxLoop; |
---|
1059 | } |
---|
1060 | |
---|
1061 | /** |
---|
1062 | * Returns the tip text for this property. |
---|
1063 | * @return tip text for this property |
---|
1064 | */ |
---|
1065 | public String minChangeTipText() { |
---|
1066 | return "set minimum number of changes (default 0)"; |
---|
1067 | } |
---|
1068 | |
---|
1069 | /** |
---|
1070 | * set the minimum number of changes |
---|
1071 | * @param m the minimum number of changes |
---|
1072 | */ |
---|
1073 | public void setMinChange(int m) { |
---|
1074 | m_minChange = m; |
---|
1075 | } |
---|
1076 | |
---|
1077 | /** |
---|
1078 | * get the minimum number of changes |
---|
1079 | * @return the minimum number of changes |
---|
1080 | */ |
---|
1081 | public int getMinChange() { |
---|
1082 | return m_minChange; |
---|
1083 | } |
---|
1084 | |
---|
1085 | /** |
---|
1086 | * Returns the tip text for this property. |
---|
1087 | * @return tip text for this property |
---|
1088 | */ |
---|
1089 | public String numClustersTipText() { |
---|
1090 | return "set number of clusters (default 2)"; |
---|
1091 | } |
---|
1092 | |
---|
1093 | /** |
---|
1094 | * Set the number of clusters |
---|
1095 | * @param n number of clusters |
---|
1096 | */ |
---|
1097 | public void setNumClusters(int n) { |
---|
1098 | m_numCluster = n; |
---|
1099 | } |
---|
1100 | |
---|
1101 | /** |
---|
1102 | * Get the number of clusters |
---|
1103 | * @return the number of clusters |
---|
1104 | */ |
---|
1105 | public int getNumClusters() { |
---|
1106 | return m_numCluster; |
---|
1107 | } |
---|
1108 | |
---|
1109 | /** |
---|
1110 | * Get the number of clusters |
---|
1111 | * @return the number of clusters |
---|
1112 | */ |
---|
1113 | public int numberOfClusters() { |
---|
1114 | return m_numCluster; |
---|
1115 | } |
---|
1116 | |
---|
1117 | /** |
---|
1118 | * Returns the tip text for this property. |
---|
1119 | * @return tip text for this property |
---|
1120 | */ |
---|
1121 | public String numRestartsTipText() { |
---|
1122 | return "set number of restarts (default 5)"; |
---|
1123 | } |
---|
1124 | |
---|
1125 | /** |
---|
1126 | * Set the number of restarts |
---|
1127 | * @param i number of restarts |
---|
1128 | */ |
---|
1129 | public void setNumRestarts(int i) { |
---|
1130 | m_numRestarts = i; |
---|
1131 | } |
---|
1132 | |
---|
1133 | /** |
---|
1134 | * Get the number of restarts |
---|
1135 | * @return number of restarts |
---|
1136 | */ |
---|
1137 | public int getNumRestarts(){ |
---|
1138 | return m_numRestarts; |
---|
1139 | } |
---|
1140 | |
---|
1141 | /** |
---|
1142 | * Returns the tip text for this property. |
---|
1143 | * @return tip text for this property |
---|
1144 | */ |
---|
1145 | public String notUnifyNormTipText() { |
---|
1146 | return "set whether to normalize each instance to a unify prior probability (eg. 1)."; |
---|
1147 | } |
---|
1148 | |
---|
1149 | /** |
---|
1150 | * Set whether to normalize instances to unify prior probability |
---|
1151 | * before building the clusterer |
---|
1152 | * @param b true to normalize, otherwise false |
---|
1153 | */ |
---|
1154 | public void setNotUnifyNorm(boolean b){ |
---|
1155 | m_uniformPrior = !b; |
---|
1156 | } |
---|
1157 | |
---|
1158 | /** |
---|
1159 | * Get whether to normalize instances to unify prior probability |
---|
1160 | * before building the clusterer |
---|
1161 | * @return true if set to normalize, false otherwise |
---|
1162 | */ |
---|
1163 | public boolean getNotUnifyNorm() { |
---|
1164 | return !m_uniformPrior; |
---|
1165 | } |
---|
1166 | |
---|
1167 | /** |
---|
1168 | * Returns a string describing this clusterer |
---|
1169 | * @return a description of the clusterer suitable for |
---|
1170 | * displaying in the explorer/experimenter gui |
---|
1171 | */ |
---|
1172 | public String globalInfo() { |
---|
1173 | return "Cluster data using the sequential information bottleneck algorithm.\n\n" + |
---|
1174 | "Note: only hard clustering scheme is supported. sIB assign for each " + |
---|
1175 | "instance the cluster that have the minimum cost/distance to the instance. " + |
---|
1176 | "The trade-off beta is set to infinite so 1/beta is zero.\n\n" + |
---|
1177 | "For more information, see:\n\n" |
---|
1178 | +getTechnicalInformation().toString(); |
---|
1179 | } |
---|
1180 | |
---|
1181 | /** |
---|
1182 | * Returns an instance of a TechnicalInformation object, containing |
---|
1183 | * detailed information about the technical background of this class, |
---|
1184 | * e.g., paper reference or book this class is based on. |
---|
1185 | * @return the technical information about this class |
---|
1186 | */ |
---|
1187 | public TechnicalInformation getTechnicalInformation() { |
---|
1188 | TechnicalInformation result; |
---|
1189 | |
---|
1190 | result = new TechnicalInformation(Type.INPROCEEDINGS); |
---|
1191 | result.setValue(Field.AUTHOR, "Noam Slonim and Nir Friedman and Naftali Tishby"); |
---|
1192 | result.setValue(Field.YEAR, "2002"); |
---|
1193 | result.setValue(Field.TITLE, "Unsupervised document classification using sequential information maximization"); |
---|
1194 | result.setValue(Field.BOOKTITLE, "Proceedings of the 25th International ACM SIGIR Conference on Research and Development in Information Retrieval"); |
---|
1195 | result.setValue(Field.PAGES, "129-136"); |
---|
1196 | |
---|
1197 | return result; |
---|
1198 | } |
---|
1199 | |
---|
1200 | /** |
---|
1201 | * Returns default capabilities of the clusterer. |
---|
1202 | * @return the capabilities of this clusterer |
---|
1203 | */ |
---|
1204 | public Capabilities getCapabilities() { |
---|
1205 | Capabilities result = super.getCapabilities(); |
---|
1206 | result.disableAll(); |
---|
1207 | result.enable(Capability.NO_CLASS); |
---|
1208 | |
---|
1209 | // attributes |
---|
1210 | result.enable(Capability.NUMERIC_ATTRIBUTES); |
---|
1211 | return result; |
---|
1212 | } |
---|
1213 | |
---|
1214 | public String toString(){ |
---|
1215 | StringBuffer text = new StringBuffer(); |
---|
1216 | text.append("\nsIB\n===\n"); |
---|
1217 | text.append("\nNumber of clusters: " + m_numCluster + "\n"); |
---|
1218 | |
---|
1219 | for (int j = 0; j < m_numCluster; j++) { |
---|
1220 | text.append("\nCluster: " + j + " Size : " + bestT.size(j) + " Prior probability: " |
---|
1221 | + Utils.doubleToString(bestT.Pt[j], 4) + "\n\n"); |
---|
1222 | for (int i = 0; i < m_numAttributes; i++) { |
---|
1223 | text.append("Attribute: " + m_data.attribute(i).name() + "\n"); |
---|
1224 | text.append("Probability given the cluster = " |
---|
1225 | + Utils.doubleToString(bestT.Py_t.get(i, j), 4) |
---|
1226 | + "\n"); |
---|
1227 | } |
---|
1228 | } |
---|
1229 | return text.toString(); |
---|
1230 | } |
---|
1231 | |
---|
1232 | /** |
---|
1233 | * Returns the revision string. |
---|
1234 | * |
---|
1235 | * @return the revision |
---|
1236 | */ |
---|
1237 | public String getRevision() { |
---|
1238 | return RevisionUtils.extract("$Revision: 5987 $"); |
---|
1239 | } |
---|
1240 | |
---|
1241 | public static void main(String[] argv) { |
---|
1242 | runClusterer(new sIB(), argv); |
---|
1243 | } |
---|
1244 | } |
---|