Self-Organizing Map
3-d Data Set - Demonstration Program
What does this applet do?
This applet runs
a self-organizing map tackling a 3-dimensional dataset. The map has as
its input a number of data samples of the form
(x, y, z), so at each node the weights vector is also three-dimensional.
As the map evolves, the weights (wxij, wyij and
wzij) at node (i, j)
are adjusted in the usual way, and come increasingly to resemble the
input data.
How are the data displayed?
Each node is drawn as a square
in a colour determined by interpreting the node weights
as RGB values; thus nodes with similar weights are
displayed in similar colours. (The actual weights may be determined
by clicking the mouse on a node - the weights will be displayed
below the map.)
Adjustable parameters
Two parameters can be changed. The mapsize as one
might guess, is the size of the map. Data Complexity
is a factor which determines the degree to which samples in
the dataset resemble each other. As the data complexity increases
the map will itself grow more complex, with a greater number
of distinct coloured regions, and the development of regions
which, while of similar colour, have different underlying
node weights.
Click here to load and run the applet.
Updated June 30, 2000.
Hugh Cartwright
(Hugh.Cartwright@chem.ox.ac.uk)
http://physchem.ox.ac.uk/~hmc/