mloss.org MPIKmeans

MPIKmeans 1.5

Peter Gehler — Fri, 16 Jan 2009 15:48:47 -0000

A K-means clustering implementation for command-line, Python, Matlab and C. This algorithm yields the very same solution as standard Kmeans, even after each iteration. However it uses some triangle inequalities and stores some distances. Therefore it is MUCH faster than standard Kmeans but uses more memory. See the corresponding paper for more details if you are interested.

Comment by Dmitrey Kroshko on 2008-09-02 20:20

Dmitrey Kroshko — Tue, 02 Sep 2008 20:20:15 -0000

I try to run makefile (from linux KUBUNTU) but it seems like it requires MATLAB installed. Is it true?

Comment by Peter Gehler on 2008-09-03 09:12

Peter Gehler — Wed, 03 Sep 2008 09:12:40 -0000

You can als run "make shared" and use the python version or bind your own C code.

Comment by Jerry on 2008-10-16 14:18

Jerry — Thu, 16 Oct 2008 14:18:22 -0000

"help mpikmeansmex" does not work! I think there is no mpikmeansmex.m file in the Archive:)

Comment by wang haoxue on 2008-11-25 02:43

wang haoxue — Tue, 25 Nov 2008 02:43:11 -0000

this means is very good!

Comment by zzinwhu on 2009-02-15 19:49

zzinwhu — Sun, 15 Feb 2009 19:49:23 -0000

How can I download it?

Comment by Bill on 2009-03-21 04:04

Bill — Sat, 21 Mar 2009 04:04:41 -0000

Hello! I need to use this function for my research experiment in matlab. But the archive was missing file "mpikmeansmex", so I couldn't run "help mpikmeansmex". please help me! thx!

Comment by Peter Gehler on 2009-04-07 16:04

Peter Gehler — Tue, 07 Apr 2009 16:04:58 -0000

There was a typo in the README. To get help in matlab simply type "help mpi_kmeans"

Comment by Janosch Peters on 2010-06-24 14:42

Janosch Peters — Thu, 24 Jun 2010 14:42:57 -0000

mpi kmeans is causing a segfault when the number of clusters is too close to the number of features. See the gdb output below:

Program received signal SIGSEGV, Segmentation fault. 0x00007fffef49e465 in add_point_to_cluster (cluster_ind=71796994, CX=0x8f2eb70, px=0xb90beb0, nr_points=0x6c235a0, dim=128)
at /home/zaphod/Code/Recognosco/src/mpi_kmeans-1.5/mpi_kmeans.cxx:100
100     if (nr_points[cluster_ind]==0)
(gdb) backtrace
#0  0x00007fffef49e465 in add_point_to_cluster (cluster_ind=71796994, CX=0x8f2eb70, px=0xb90beb0, nr_points=0x6c235a0, dim=128)
at /home/zaphod/Code/Recognosco/src/mpi_kmeans-1.5/mpi_kmeans.cxx:100
#1  0x00007fffef49e68b in remove_identical_clusters (CX=0x8f2eb70, cluster_distance=0xbad9ac0, X=0xa8706b0, cluster_count=0x6c235a0, c=0x6b1ea80, 
dim=128, nclus=2000, npts=18853) at /home/zaphod/Code/Recognosco/src/mpi_kmeans-1.5/mpi_kmeans.cxx:138
#2  0x00007fffef49f12f in kmeans_run (CX=0x8f2eb70, X=0xa8706b0, c=0x6b1ea80, dim=128, npts=18853, nclus=2000, maxiter=0)
at /home/zaphod/Code/Recognosco/src/mpi_kmeans-1.5/mpi_kmeans.cxx:373
#3  0x00007fffef49fc78 in kmeans (CX=0x8f2eb70, X=0xa8706b0, assignment=0x6b1ea80, dim=128, npts=18853, nclus=2000, maxiter=0, restarts=20)
at /home/zaphod/Code/Recognosco/src/mpi_kmeans-1.5/mpi_kmeans.cxx:584