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 | * MedianOfWidestDimension.java |
---|
19 | * Copyright (C) 2007 University of Waikato, Hamilton, New Zealand |
---|
20 | */ |
---|
21 | |
---|
22 | package weka.core.neighboursearch.balltrees; |
---|
23 | |
---|
24 | import weka.core.EuclideanDistance; |
---|
25 | import weka.core.Instance; |
---|
26 | import weka.core.Instances; |
---|
27 | import weka.core.Option; |
---|
28 | import weka.core.OptionHandler; |
---|
29 | import weka.core.RevisionUtils; |
---|
30 | import weka.core.TechnicalInformation; |
---|
31 | import weka.core.TechnicalInformationHandler; |
---|
32 | import weka.core.Utils; |
---|
33 | import weka.core.TechnicalInformation.Field; |
---|
34 | import weka.core.TechnicalInformation.Type; |
---|
35 | |
---|
36 | import java.util.Enumeration; |
---|
37 | import java.util.Vector; |
---|
38 | |
---|
39 | /** |
---|
40 | <!-- globalinfo-start --> |
---|
41 | * Class that splits a BallNode of a ball tree based on the median value of the widest dimension of the points in the ball. It essentially implements Omohundro's KD construction algorithm. |
---|
42 | * <p/> |
---|
43 | <!-- globalinfo-end --> |
---|
44 | * |
---|
45 | <!-- technical-bibtex-start --> |
---|
46 | * BibTeX: |
---|
47 | * <pre> |
---|
48 | * @techreport{Omohundro1989, |
---|
49 | * author = {Stephen M. Omohundro}, |
---|
50 | * institution = {International Computer Science Institute}, |
---|
51 | * month = {December}, |
---|
52 | * number = {TR-89-063}, |
---|
53 | * title = {Five Balltree Construction Algorithms}, |
---|
54 | * year = {1989} |
---|
55 | * } |
---|
56 | * </pre> |
---|
57 | * <p/> |
---|
58 | <!-- technical-bibtex-end --> |
---|
59 | * |
---|
60 | <!-- options-start --> |
---|
61 | * Valid options are: <p/> |
---|
62 | * |
---|
63 | * <pre> -N |
---|
64 | * Normalize dimensions' widths.</pre> |
---|
65 | * |
---|
66 | <!-- options-end --> |
---|
67 | * |
---|
68 | * @author Ashraf M. Kibriya (amk14[at-the-rate]cs[dot]waikato[dot]ac[dot]nz) |
---|
69 | * @version $Revision: 5953 $ |
---|
70 | */ |
---|
71 | public class MedianOfWidestDimension |
---|
72 | extends BallSplitter |
---|
73 | implements OptionHandler, TechnicalInformationHandler { |
---|
74 | |
---|
75 | /** for serialization. */ |
---|
76 | private static final long serialVersionUID = 3054842574468790421L; |
---|
77 | |
---|
78 | /** |
---|
79 | * Should we normalize the widths(ranges) of the dimensions (attributes) |
---|
80 | * before selecting the widest one. |
---|
81 | */ |
---|
82 | protected boolean m_NormalizeDimWidths = true; |
---|
83 | |
---|
84 | /** |
---|
85 | * Constructor. |
---|
86 | */ |
---|
87 | public MedianOfWidestDimension() { |
---|
88 | } |
---|
89 | |
---|
90 | /** |
---|
91 | * Constructor. |
---|
92 | * @param instList The master index array. |
---|
93 | * @param insts The instances on which the tree |
---|
94 | * is (or is to be) built. |
---|
95 | * @param e The Euclidean distance function to |
---|
96 | * use for splitting. |
---|
97 | */ |
---|
98 | public MedianOfWidestDimension(int[] instList, Instances insts, |
---|
99 | EuclideanDistance e) { |
---|
100 | super(instList, insts, e); |
---|
101 | } |
---|
102 | /** |
---|
103 | * Returns a string describing this nearest neighbour search algorithm. |
---|
104 | * |
---|
105 | * @return a description of the algorithm for displaying in the |
---|
106 | * explorer/experimenter gui |
---|
107 | */ |
---|
108 | public String globalInfo() { |
---|
109 | return |
---|
110 | "Class that splits a BallNode of a ball tree based on the " |
---|
111 | + "median value of the widest dimension of the points in the ball. " |
---|
112 | + "It essentially implements Omohundro's KD construction algorithm."; |
---|
113 | } |
---|
114 | |
---|
115 | |
---|
116 | /** |
---|
117 | * Returns an instance of a TechnicalInformation object, containing detailed |
---|
118 | * information about the technical background of this class, e.g., paper |
---|
119 | * reference or book this class is based on. |
---|
120 | * |
---|
121 | * @return the technical information about this class |
---|
122 | */ |
---|
123 | public TechnicalInformation getTechnicalInformation() { |
---|
124 | TechnicalInformation result; |
---|
125 | |
---|
126 | result = new TechnicalInformation(Type.TECHREPORT); |
---|
127 | result.setValue(Field.AUTHOR, "Stephen M. Omohundro"); |
---|
128 | result.setValue(Field.YEAR, "1989"); |
---|
129 | result.setValue(Field.TITLE, "Five Balltree Construction Algorithms"); |
---|
130 | result.setValue(Field.MONTH, "December"); |
---|
131 | result.setValue(Field.NUMBER, "TR-89-063"); |
---|
132 | result.setValue(Field.INSTITUTION, "International Computer Science Institute"); |
---|
133 | |
---|
134 | return result; |
---|
135 | } |
---|
136 | |
---|
137 | /** |
---|
138 | * Splits a ball into two. |
---|
139 | * @param node The node to split. |
---|
140 | * @param numNodesCreated The number of nodes that so far have been |
---|
141 | * created for the tree, so that the newly created nodes are |
---|
142 | * assigned correct/meaningful node numbers/ids. |
---|
143 | * @throws Exception If there is some problem in splitting the |
---|
144 | * given node. |
---|
145 | */ |
---|
146 | public void splitNode(BallNode node, int numNodesCreated) throws Exception { |
---|
147 | correctlyInitialized(); |
---|
148 | //int[] instList = getNodesInstsList(node); |
---|
149 | double[][] ranges = m_DistanceFunction.initializeRanges(m_Instlist, |
---|
150 | node.m_Start, |
---|
151 | node.m_End); |
---|
152 | |
---|
153 | int splitAttrib = widestDim(ranges, m_DistanceFunction.getRanges()); |
---|
154 | |
---|
155 | //In this case median is defined to be either the middle value (in case of |
---|
156 | //odd number of values) or the left of the two middle values (in case of |
---|
157 | //even number of values). |
---|
158 | int medianIdxIdx = node.m_Start + (node.m_End-node.m_Start)/2; |
---|
159 | //the following finds the median and also re-arranges the array so all |
---|
160 | //elements to the left are < median and those to the right are > median. |
---|
161 | int medianIdx = select(splitAttrib, m_Instlist, node.m_Start, node.m_End, (node.m_End-node.m_Start)/2+1); //Utils.select(array, indices, node.m_Start, node.m_End, (node.m_End-node.m_Start)/2+1); //(int) (node.m_NumInstances/2D+0.5D); |
---|
162 | |
---|
163 | Instance pivot; |
---|
164 | |
---|
165 | node.m_SplitAttrib = splitAttrib; |
---|
166 | node.m_SplitVal = m_Instances.instance(m_Instlist[medianIdx]) |
---|
167 | .value(splitAttrib); |
---|
168 | |
---|
169 | node.m_Left = new BallNode(node.m_Start, medianIdxIdx, numNodesCreated+1, |
---|
170 | (pivot=BallNode.calcCentroidPivot(node.m_Start, |
---|
171 | medianIdxIdx, m_Instlist, |
---|
172 | m_Instances)), |
---|
173 | BallNode.calcRadius(node.m_Start, medianIdxIdx, |
---|
174 | m_Instlist, m_Instances, |
---|
175 | pivot, m_DistanceFunction) |
---|
176 | ); |
---|
177 | node.m_Right = new BallNode(medianIdxIdx+1, node.m_End, numNodesCreated+2, |
---|
178 | (pivot=BallNode.calcCentroidPivot(medianIdxIdx+1, |
---|
179 | node.m_End, m_Instlist, |
---|
180 | m_Instances)), |
---|
181 | BallNode.calcRadius(medianIdxIdx+1, node.m_End, |
---|
182 | m_Instlist, m_Instances, |
---|
183 | pivot, m_DistanceFunction) |
---|
184 | ); |
---|
185 | } |
---|
186 | |
---|
187 | /** |
---|
188 | * Partitions the instances around a pivot. Used by quicksort and |
---|
189 | * kthSmallestValue. |
---|
190 | * |
---|
191 | * @param attIdx The attribution/dimension based on which the |
---|
192 | * instances should be partitioned. |
---|
193 | * @param index The master index array containing indices of the |
---|
194 | * instances. |
---|
195 | * @param l The begining index of the portion of master index |
---|
196 | * array that should be partitioned. |
---|
197 | * @param r The end index of the portion of master index array |
---|
198 | * that should be partitioned. |
---|
199 | * @return the index of the middle element (in the master |
---|
200 | * index array, i.e. index of the index of middle element). |
---|
201 | */ |
---|
202 | protected int partition(int attIdx, int[] index, int l, int r) { |
---|
203 | |
---|
204 | double pivot = m_Instances.instance(index[(l + r) / 2]).value(attIdx); |
---|
205 | int help; |
---|
206 | |
---|
207 | while (l < r) { |
---|
208 | while ((m_Instances.instance(index[l]).value(attIdx) < pivot) && (l < r)) { |
---|
209 | l++; |
---|
210 | } |
---|
211 | while ((m_Instances.instance(index[r]).value(attIdx) > pivot) && (l < r)) { |
---|
212 | r--; |
---|
213 | } |
---|
214 | if (l < r) { |
---|
215 | help = index[l]; |
---|
216 | index[l] = index[r]; |
---|
217 | index[r] = help; |
---|
218 | l++; |
---|
219 | r--; |
---|
220 | } |
---|
221 | } |
---|
222 | if ((l == r) && (m_Instances.instance(index[r]).value(attIdx) > pivot)) { |
---|
223 | r--; |
---|
224 | } |
---|
225 | |
---|
226 | return r; |
---|
227 | } |
---|
228 | |
---|
229 | /** |
---|
230 | * Implements computation of the kth-smallest element according |
---|
231 | * to Manber's "Introduction to Algorithms". |
---|
232 | * |
---|
233 | * @param attIdx The dimension/attribute of the instances in |
---|
234 | * which to find the kth-smallest element. |
---|
235 | * @param indices The master index array containing indices of |
---|
236 | * the instances. |
---|
237 | * @param left The begining index of the portion of the master |
---|
238 | * index array in which to find the kth-smallest element. |
---|
239 | * @param right The end index of the portion of the master index |
---|
240 | * array in which to find the kth-smallest element. |
---|
241 | * @param k The value of k |
---|
242 | * @return The index of the kth-smallest element |
---|
243 | */ |
---|
244 | public int select(int attIdx, int[] indices, |
---|
245 | int left, int right, int k) { |
---|
246 | |
---|
247 | if (left == right) { |
---|
248 | return left; |
---|
249 | } else { |
---|
250 | int middle = partition(attIdx, indices, left, right); |
---|
251 | if ((middle - left + 1) >= k) { |
---|
252 | return select(attIdx, indices, left, middle, k); |
---|
253 | } else { |
---|
254 | return select(attIdx, indices, middle + 1, right, k - (middle - left + 1)); |
---|
255 | } |
---|
256 | } |
---|
257 | } |
---|
258 | |
---|
259 | /** |
---|
260 | * Returns the widest dimension. The width of each |
---|
261 | * dimension (for the points inside the node) is |
---|
262 | * normalized, if m_NormalizeNodeWidth is set to |
---|
263 | * true. |
---|
264 | * @param nodeRanges The attributes' range of the |
---|
265 | * points inside the node that is to be split. |
---|
266 | * @param universe The attributes' range for the |
---|
267 | * whole point-space. |
---|
268 | * @return The index of the attribute/dimension |
---|
269 | * in which the points of the node have widest |
---|
270 | * spread. |
---|
271 | */ |
---|
272 | protected int widestDim(double[][] nodeRanges, double[][] universe) { |
---|
273 | final int classIdx = m_Instances.classIndex(); |
---|
274 | double widest = 0.0; |
---|
275 | int w = -1; |
---|
276 | if (m_NormalizeDimWidths) { |
---|
277 | for (int i = 0; i < nodeRanges.length; i++) { |
---|
278 | double newWidest = nodeRanges[i][m_DistanceFunction.R_WIDTH] / |
---|
279 | universe[i][m_DistanceFunction.R_WIDTH]; |
---|
280 | if (newWidest > widest) { |
---|
281 | if(i == classIdx) continue; |
---|
282 | widest = newWidest; |
---|
283 | w = i; |
---|
284 | } |
---|
285 | } |
---|
286 | } |
---|
287 | else { |
---|
288 | for (int i = 0; i < nodeRanges.length; i++) { |
---|
289 | if (nodeRanges[i][m_DistanceFunction.R_WIDTH] > widest) { |
---|
290 | if(i == classIdx) continue; |
---|
291 | widest = nodeRanges[i][m_DistanceFunction.R_WIDTH]; |
---|
292 | w = i; |
---|
293 | } |
---|
294 | } |
---|
295 | } |
---|
296 | return w; |
---|
297 | } |
---|
298 | |
---|
299 | /** |
---|
300 | * Returns the tip text for this property. |
---|
301 | * |
---|
302 | * @return tip text for this property suitable for |
---|
303 | * displaying in the explorer/experimenter gui |
---|
304 | */ |
---|
305 | public String normalizeDimWidthsTipText() { |
---|
306 | return |
---|
307 | "Whether to normalize the widths(ranges) of the dimensions " |
---|
308 | + "(attributes) before selecting the widest one."; |
---|
309 | } |
---|
310 | |
---|
311 | /** |
---|
312 | * Should we normalize the widths(ranges) of the dimensions (attributes) |
---|
313 | * before selecting the widest one. |
---|
314 | * @param normalize Should be true if the widths are to be |
---|
315 | * normalized. |
---|
316 | */ |
---|
317 | public void setNormalizeDimWidths(boolean normalize) { |
---|
318 | m_NormalizeDimWidths = normalize; |
---|
319 | } |
---|
320 | |
---|
321 | /** |
---|
322 | * Whether we are normalizing the widths(ranges) of the dimensions (attributes) |
---|
323 | * or not. |
---|
324 | * @return true if widths are being normalized. |
---|
325 | */ |
---|
326 | public boolean getNormalizeDimWidths() { |
---|
327 | return m_NormalizeDimWidths; |
---|
328 | } |
---|
329 | |
---|
330 | /** |
---|
331 | * Returns an enumeration describing the available options. |
---|
332 | * |
---|
333 | * @return an enumeration of all the available options. |
---|
334 | */ |
---|
335 | public Enumeration listOptions() { |
---|
336 | Vector<Option> newVector = new Vector<Option>(); |
---|
337 | |
---|
338 | newVector.addElement(new Option( |
---|
339 | "\tNormalize dimensions' widths.", |
---|
340 | "N", 0, "-N")); |
---|
341 | |
---|
342 | return newVector.elements(); |
---|
343 | } |
---|
344 | |
---|
345 | /** |
---|
346 | * Parses a given list of options. |
---|
347 | * |
---|
348 | <!-- options-start --> |
---|
349 | * Valid options are: <p/> |
---|
350 | * |
---|
351 | * <pre> -N |
---|
352 | * Normalize dimensions' widths.</pre> |
---|
353 | * |
---|
354 | <!-- options-end --> |
---|
355 | * |
---|
356 | * @param options the list of options as an array of strings |
---|
357 | * @throws Exception if an option is not supported |
---|
358 | */ |
---|
359 | public void setOptions(String[] options) |
---|
360 | throws Exception { |
---|
361 | |
---|
362 | setNormalizeDimWidths(Utils.getFlag('N', options)); |
---|
363 | } |
---|
364 | |
---|
365 | /** |
---|
366 | * Gets the current settings. |
---|
367 | * @return An array of strings suitable for passing to |
---|
368 | * setOptions or to be displayed by a |
---|
369 | * GenericObjectEditor. |
---|
370 | */ |
---|
371 | public String[] getOptions() { |
---|
372 | Vector<String> result; |
---|
373 | |
---|
374 | result = new Vector<String>(); |
---|
375 | |
---|
376 | if (getNormalizeDimWidths()) |
---|
377 | result.add("-N"); |
---|
378 | |
---|
379 | return result.toArray(new String[result.size()]); |
---|
380 | } |
---|
381 | |
---|
382 | /** |
---|
383 | * Returns the revision string. |
---|
384 | * |
---|
385 | * @return the revision |
---|
386 | */ |
---|
387 | public String getRevision() { |
---|
388 | return RevisionUtils.extract("$Revision: 5953 $"); |
---|
389 | } |
---|
390 | } |
---|