************************************************************************
Example: art2_tetra.xxx		ART2 tetrahedron network
************************************************************************

Problem description:
====================

The ART2 tetrahedron network shows the self-organized classification
of real valued input pattern vectors by an ART2 network. The input
patterns are noisy real valued coordinates of the verteces of a
tetrahedron in 3D space. They should automatically be classified into
four different clusters. There exist variations of the input patterns
with different amount of noise added to the input patterns.

See the SNNS user manual for a more detailed description of the ART2
implementation in SNNS.


Pattern-Files:	art2_tetra.pat	
==============	art2_tetra_low.pat
		art2_tetra_med.pat
		art2_tetra_high.pat

All art2_tetra pattern files contain 40 input patterns with 3 real
values each, describing a noisy coordinate of a vertex of the
tetrahedron. The files differ by the amount of noise added to the
verteces as indicated by the suffix 'low' 'med'(ium) and 'high'.


Network-Files:	art2_tetra.net
==============

This network file contains a trained ART2 network for the tetrahedron
vertex classification task described above. The standard configuration file
for this network is art2_tetra.cfg

You may generate your own ART2 network with the BIGNET tool from the
Info-Panel of SNNS. This automatically generates all units and the
necessary connections.

Because the unit types and link structure are highly specialized in
ART2 we strongly urge you only to use this tool to generate ART2
networks in SNNS.


Config-Files:	art2_tetra.cfg	(one 2D display only)
=============	

The drawing of the 3D display is relatively slow for this network, so
you probably want to work with the 2D display once you know how the
units are connected.
The 3D display is a nice example for a moderately complicated 3D
network layout.


Result-Files:	(none)
=============


Hints:
======

Read the chapter about ART2 in the SNNS manual very carefully!

Note that ART2 needs a special network initialization function 
	(REMOTE panel: OPTIONS select init function: ART2_Weights).
Note that there exist two different ART2 update functions: 
	(REMOTE panel: OPTIONS select update function: ART2_Synchronous
						or     ART2_Stable)
Note that ART2 needs a special learning function:
	(REMOTE panel: OPTIONS select learning function: ART2)
These should already be set when loading the example ART2 network.

You may use the ART2 learning parameters as given in the figure
'Setting the ART2 learning parameters $\rho$, a, b, c, and $\theta$.

There exists additional documentation in form of the diploma
thesis of Kai-Uwe Herrmann (in German), available via anon. ftp
from our public ftp server  ftp.informatik.uni-stuttgart.de  as file
/pub/SNNS/NN-papers-german/herrmann_kaiuwe_DA.ps.Z

************************************************************************