Affinity-based clustering of vectors for partitioning the columns of a matrix

A computer program product, comprising a computer usable storage medium having a computer readable program embodied therein, said computer readable program comprising an algorithm for partitioning the columns of a matrix A, said algorithm adapted to perform the steps of:

• generating a list of clusters having n clusters such that each of the n clusters is a unique column of the matrix A, said matrix A being stored in a memory device of the computer system, said matrix A having n columns and m rows, n being an integer of at least 2, m being an integer of at least 1, each said cluster consisting of one or more columns of said matrix A;
• determining if a termination condition is satisfied and if said determining so determines that said termination condition is satisfied then terminating said algorithm else executing the following steps:
• selecting a next pair of clusters from the list of clusters, said next pair of clusters consisting of a first cluster and a second cluster, said next pair of clusters having an affinity that is not less than an affinity between any pair of clusters not yet selected from the list of clusters;
• merging the first and second clusters to form a new cluster;
• inserting the new cluster into the list of clusters while removing the first and second clusters from the list of clusters;
• and
• re-executing said determining step,
• wherein the method further comprises storing the list of clusters comprising all of said inserted new clusters in a computer-readable storage device.