The Matlab codes in this directory are written for the "basis
segmentation" algorithms described in

Ella Bingham, Aristides Gionis, Niina Haiminen, Heli Hiisilä, Heikki
Mannila, Evimaria Terzi, "Segmentation and dimensionality reduction". 2006
SIAM Conference on Data Mining, April 20-22, 2006, Bethesda, Maryland, USA.

The Matlab codes are written by Heli Hiisilä and Ella Bingham, 2005,
with contributions from the co-authors of the above paper.

PURPOSE OF THE SOFTWARE:

The purpose of the software is to segment a multidimensional time
series. In particular, we find a small set of latent variables and a
segmentation of the series such that the data in each segment can be
explained well by some linear combination of the latent variables.
This linear combination of latent variables constitutes a basis, and
we are essentially reducing the dimension of the input sequence by
representing it using the basis. The number of latent variables (that
is, the size of the basis) is m and the number of segments is k.

HOW TO USE THE SOFTWARE:	

Installing
----------

Unpack the codes by saying
"tar -cf basis_segmentation_codes.tar"

The Matlab codes need not be compiled but can instead be run inside
Matlab by invoking Matlab in the same directory as where the codes are
stored.

Running
-------

BasisSegmentation.m is the main program which invokes other programs.
It is called in Matlab as 

[output] = BasisSegmentation(X,k,m,p)

where the input variables are:
X - input data, has rows as dimensions and columns as observations
k - amount of segments
m - amount of basis vectors
p - whether to represent a segment by its mean (p=1) or median (p=2)

and the output is a structure containing the reconstruction errors,
segment boundaries and segment means given by the algorithms described
in the paper.

File tao_data85.mat gives an example of input data, and it was used to
generate the results in Table 6 and Figure 8 of the above paper. The
data is originally from the UCI KDD archive [S. Hettich and S.D. Bay,
The UCI KDD Archive, 1999, UC, Irvine, http://kdd.ics.uci.edu ]

January 31, 2006, Ella Bingham, ella@iki.fi
(Updated May 11, 2007)