opt-k-means

This code finds all possible k-clusters, which are convex, and select the minimum one with minimal inter cluster function

NEWS (0.0.2)

• There is a new file called not-so-opt.py, containing a completely implementation in python. That is, runing this program works the same but it does not contain any reference to .so.
• The implementation of Kmeans in Scipy is not the one in Scipy. The initialization of the centroids is done as described in the kmeans++ algorithm.

NEWS (0.0.1)

• The program prints only once the inter cluster measure of the optimum solution with the inter cluster measure found by k-means (implemented in SCIPY)
• The default number of procs is set to 4
• Bookkeping is done in a separate directory
• The minimum cluster is written to a file named 'minimum_cluster' with its inter cluster measure
• The kmeans minimum cluster is written to a file named 'kmeans_cluster' with its inter cluster measure

REQUIREMENTS

The following libraries are required:

• numpy
• sciPy

Also, because there are some parts of the code that use c code, it is necessary a compiler like gcc.

TO INSTALL

First take the file array.c, and issue the following command on the terminal

gcc -O3 -shared -Wl,-soname,array -o arraylib.so -fPIC array.c

TO USE

The script must be run in the terminal, so to see the help just write:

python opt-k-means -h

The data file name must be a file of float numbers, separated by commas and in each line, there must be the same number of features.