Compile: cc -O -o cluster cluster.c alloc.c -lm Use: cluster [vectors] [names] where 'vectors' is the file containing vectors to be clustered, and 'names' is a file containing the name of each vector. Vectors should be organized one per line, => with a trailing blank before the carriage return. Names should be organized similarly; the order of vectors/names should be the same in both files. 'names' is a sample names file; 'vector' is a sample vector file. Note that whereas names must not have any embedded white space, elements in vectors must be separated by white space. ---- The program has been enhanced with numerous options to make it more flexible in interfacing with other applications. See the the man page for details. To generate the program, edit the Makefile according to your preferences and type `make'. It is know to compile and run os Sun3 and Sun4 systems under SunOS 3.5 and 4.1. Comments and bug fixes requested. Andreas Stolcke (stolcke@icsi.berkeley.edu) 10/12/90 ----- The second enhanced release has even more options and supports principal component analysis. RTFM page !!! A. Stolcke 4/20/91 ----- The first release after beta testing (nobody found the glaring bug in the memory allocation macro ...) has curses support added. A. Stolcke 7/14/91 ----- New in release 2.3: * Eigenvectors are now stored one-per-line, so as to be compatible with input/output format. Old eigenbases will have to recomputed, or the contents of files used with -e can be transposed. * A new option, -E, allows printing the eigenvalues determined in PCA. A. Stolcke 2/11/92 ---- New in release 2.4: * Saved some unnecessary memory allocation for patterns. * Distance matrix needs only 1/2 the space it used before (this is the bulk of total memory size, so big win here). * 20% runtime improvement by making a trivial change in the way the distance matrix is scanned. A. Stolcke 1/15/93 --- New in release 2.5: * Cluster implementation changed to use O(n^2) time and O(n) memory (was O(n^3) and O(n^2), respectively). * Added BENCHMARKS file and test/ subdirectory. A. Stolcke 1/20/93 * Added some info about the cluster algorithm, to be found in the ALGORITHM file. A. Stolcke 1/21/93 --- New in release 2.6: * Cluster output as binary code vectors (-B option). A. Stolcke 2/2/93 ---- New in release 2.7: * Use IEEE NaN instead of infinity for D/C value representation * Include test program to verify that it works (make testm). This version has been tested on Sun4/SunOS4.1.x, DECstation/Ultrix4.3 NeXT/Release 2.1 HP 9000s700/HP-UX 8.07 A. Stolcke 3/1/93 --- New in release 2.8: * Included MAIL from fmurtagh@eso.org with more background info about clustering algorithms and other software packages. A. Stolcke 4/28/93 New in release 2.9: * Fixed a memory (de)allocation bug in pca, found by Nici Schraudolph. A. Stolcke 1/10/96
