1 /* =============================================================================
5 * =============================================================================
9 * Takes as input a file:
10 * ascii file: containing 1 data point per line
11 * binary file: first int is the number of objects
12 * 2nd int is the no. of features of each object
14 * This example performs a fuzzy c-means clustering on the data. Fuzzy clustering
15 * is performed using min to max clusters and the clustering that gets the best
16 * score according to a compactness and separation criterion are returned.
22 * ECE Department Northwestern University
23 * email: wkliao@ece.northwestern.edu
29 * Northwestern University
34 * Port to Java version
36 * University of California, Irvine
38 * =============================================================================
40 * ------------------------------------------------------------------------
42 * For the license of kmeans, please see kmeans/LICENSE.kmeans
44 * ------------------------------------------------------------------------
46 * Unless otherwise noted, the following license applies to STAMP files:
48 * Copyright (c) 2007, Stanford University
49 * All rights reserved.
51 * Redistribution and use in source and binary forms, with or without
52 * modification, are permitted provided that the following conditions are
55 * * Redistributions of source code must retain the above copyright
56 * notice, this list of conditions and the following disclaimer.
58 * * Redistributions in binary form must reproduce the above copyright
59 * notice, this list of conditions and the following disclaimer in
60 * the documentation and/or other materials provided with the
63 * * Neither the name of Stanford University nor the names of its
64 * contributors may be used to endorse or promote products derived
65 * from this software without specific prior written permission.
67 * THIS SOFTWARE IS PROVIDED BY STANFORD UNIVERSITY ``AS IS'' AND ANY
68 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
69 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
70 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL STANFORD UNIVERSITY BE LIABLE
71 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
72 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
73 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
74 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
75 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
76 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
77 * THE POSSIBILITY OF SUCH DAMAGE.
79 * =============================================================================
82 public class KMeans extends Thread {
84 * User input for max clusters
89 * User input for min clusters
94 * Check for Binary file
99 * Using zscore transformation for cluster center
100 * deviating from distribution's mean
102 int use_zscore_transform;
105 * Input file name used for clustering
110 * Total number of threads
115 * threshold until which kmeans cluster continues
125 * Global arguments for threads
130 * Output: Number of best clusters
135 * Output: Cluster centers
137 float[][] cluster_centres;
142 float[][] attributes;
152 use_zscore_transform = 1;
153 threshold = (float) 0.001;
157 public KMeans(int threadid, GlobalArgs g_args, int nthreads, int use_zscore_transform,
158 int max_nclusters, int min_nclusters, float threshold,
159 float[][] attributes, int numObjects, int numAttributes) {
160 this.threadid = threadid;
161 this.g_args = g_args;
166 Normal.work(threadid, g_args);
170 /* =============================================================================
172 * =============================================================================
174 public static void main(String[] args) {
176 int MAX_LINE_LENGTH = 1000000; /* max input is 400000 one digit input + spaces */
179 * Read options fron the command prompt
183 KMeans.parseCmdLine(args, kms);
184 nthreads = kms.nthreads;
185 System.out.println("nthreads= " + kms.nthreads);
187 if (kms.max_nclusters < kms.min_nclusters) {
188 System.out.println("Error: max_clusters must be >= min_clusters\n");
192 int numAttributes = 0;
196 * From the input file, get the numAttributes (columns in txt file) and numObjects (rows in txt file)
198 if (kms.isBinaryFile == 1) {
199 System.out.println("TODO: Unimplemented Binary file option\n");
203 FileInputStream inputFile = new FileInputStream(kms.filename);
204 byte b[] = new byte[MAX_LINE_LENGTH];
206 while ((n = inputFile.read(b)) != 0) {
207 for (int i = 0; i < n; i++) {
213 inputFile = new FileInputStream(kms.filename);
215 if((line = inputFile.readLine()) != null) {
217 boolean prevWhiteSpace = true;
218 while(index < line.length()) {
219 char c = line.charAt(index++);
220 boolean currWhiteSpace = Character.isWhitespace(c);
221 if(prevWhiteSpace && !currWhiteSpace){
224 prevWhiteSpace = currWhiteSpace;
229 /* Ignore the first attribute: numAttributes = 1; */
230 numAttributes = numAttributes - 1;
232 /* Allocate objects and read attributes of all objects */
233 float[][] buf = new float[numObjects][numAttributes];
234 float[][] attributes = new float[numObjects][numAttributes];
235 KMeans.readFromFile(inputFile, kms.filename, buf, MAX_LINE_LENGTH);
236 System.out.println("Finished Reading from file ......");
239 * The core of the clustering
242 int len = kms.max_nclusters - kms.min_nclusters + 1;
246 KMeans[] km = new KMeans[nthreads];
247 GlobalArgs g_args = new GlobalArgs();
248 g_args.nthreads = nthreads;
249 for(int x = 0; x < numObjects; x++) {
250 for(int y = 0; y < numAttributes; y++) {
251 attributes[x][y] = buf[x][y];
255 /* Create and Start Threads */
256 for(int i = 1; i<nthreads; i++) {
257 km[i] = new KMeans(i, g_args, nthreads, kms.use_zscore_transform,
258 kms.max_nclusters, kms.min_nclusters, kms.threshold, attributes, numObjects, numAttributes);
262 for(int i = 1; i<nthreads; i++) {
267 System.out.println("Finished Starting threads......");
269 for (int i = 0; i < nloops; i++) {
271 // Since zscore transform may perform in cluster() which modifies the
272 // contents of attributes[][], we need to re-store the originals
274 for(int x = 0; x < numObjects; x++) {
275 for(int y = 0; y < numAttributes; y++) {
276 attributes[x][y] = buf[x][y];
281 Cluster.cluster_exec(nthreads,
284 attributes, // [numObjects][numAttributes]
285 kms, //main class that holds users inputs from command prompt and output arrays that need to be filled
286 g_args); // Global arguments common to all threads
289 System.out.println("Printing output......");
290 System.out.println("Best_nclusters= " + kms.best_nclusters);
292 /* Output: the coordinates of the cluster centres */
295 for (int i = 0; i < kms.best_nclusters; i++) {
296 System.out.print(i + " ");
297 for (int j = 0; j < numAttributes; j++) {
298 System.out.print(kms.cluster_centres[i][j] + " ");
300 System.out.println("\n");
304 System.out.println("Finished......\n");
308 public static void parseCmdLine(String args[], KMeans km) {
311 while (i < args.length && args[i].startsWith("-")) {
314 if(arg.equals("-m")) {
315 if(i < args.length) {
316 km.max_nclusters = new Integer(args[i++]).intValue();
318 } else if(arg.equals("-n")) {
319 if(i < args.length) {
320 km.min_nclusters = new Integer(args[i++]).intValue();
322 } else if(arg.equals("-t")) {
323 if(i < args.length) {
324 km.threshold = new Integer(args[i++]).intValue();
326 } else if(arg.equals("-i")) {
327 if(i < args.length) {
328 km.filename = args[i++];
330 } else if(arg.equals("-b")) {
331 if(i < args.length) {
332 km.isBinaryFile = new Integer(args[i++]).intValue();
334 } else if(arg.equals("-z")) {
335 km.use_zscore_transform=0;
336 } else if(arg.equals("-nthreads")) {
337 if(i < args.length) {
338 km.nthreads = new Integer(args[i++]).intValue();
340 } else if(arg.equals("-h")) {
344 if(km.nthreads == 0 || km.filename == null) {
350 * The usage routine which describes the program options.
352 public void usage() {
353 System.out.println("usage: ./kmeans -m <max_clusters> -n <min_clusters> -t <threshold> -i <filename> -nthreads <threads>\n");
354 System.out.println( " -i filename: file containing data to be clustered\n");
355 System.out.println( " -b input file is in binary format\n");
356 System.out.println( " -m max_clusters: maximum number of clusters allowed\n");
357 System.out.println( " -n min_clusters: minimum number of clusters allowed\n");
358 System.out.println( " -z : don't zscore transform data\n");
359 System.out.println( " -t threshold : threshold value\n");
360 System.out.println( " -nthreads : number of threads\n");
365 * Read attributes from the input file into an array
367 public static void readFromFile(FileInputStream inputFile, String filename, float[][] buf, int MAX_LINE_LENGTH) {
368 inputFile = new FileInputStream(filename);
372 byte b[] = new byte[MAX_LINE_LENGTH];
374 byte oldbytes[]=null;
377 // transaction will never abort because it is only executed
378 // on master machine and therefore the fileread native call is
379 //allowed as a warning
381 while ((n = inputFile.read(b)) != 0) {
384 if (oldbytes!=null) {
395 byte newbytes[]= new byte[x+oldbytes.length];
396 boolean isnumber=false;
397 for(int ii=0;ii<oldbytes.length;ii++) {
398 if (oldbytes[ii]>='0'&&oldbytes[ii]<='9')
400 newbytes[ii]=oldbytes[ii];
402 for(int ii=0;ii<x;ii++) {
403 if (b[ii]>='0'&&b[ii]<='9')
405 newbytes[ii+oldbytes.length]=b[ii];
408 x++; //skip past space or cr
411 buf[i][j]=(float)Double.parseDouble(new String(newbytes, 0, newbytes.length));
425 boolean isnumber=false;
427 if ((b[y]>='0')&&(b[y]<='9'))
437 //need to continue for another read
438 oldbytes=new byte[y-x];
439 for(int ii=0;ii<(y-x);ii++)
440 oldbytes[ii]=b[ii+x];
444 //otherwise x is beginning of character string, y is end
447 buf[i][j]=(float)Double.parseDouble(new String(b,x,y-x));
452 i++;//skip to next line
453 j = -1;//don't store line number
454 x=y;//skip to end of number
455 x++;//skip past return
457 x=y;//skip to end of number
458 x++;//skip past space
466 /* =============================================================================
470 * =============================================================================