home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
HPAVC
/
HPAVC CD-ROM.iso
/
SNNSV32.ZIP
/
SNNSv3.2
/
examples
/
som_cube.README
< prev
next >
Wrap
Text File
|
1994-04-25
|
3KB
|
78 lines
=============================================================================
README file for the example files som_cube.xxx
=============================================================================
Description: Self-organizing map cube example
============
The files som_cube.xxx describe network and pattern files used to
demonstrate the use of the self-organizing map (SOM, Kohonen map) in
SNNS. The eight training patterns are the verteces of a
three-dimensional cube of size 2x2x2 centered at the origin, with
vertex coordinates from (-1, -1, -1) to (1, 1, 1).
The self-organizing map is a two-dimensional 16x16 grid of neurons.
The three-dimensional input vectors are mapped to different positions
of the two-dimensional map.
See the user manual for a more detailed description of the SOM
implementation in SNNS and its usage.
Network-Files: som_cube.net
==============
This network file contains a trained SOM network for the cube verteces
task described above. The self-organizing map is a two-dimensional
16x16 grid of neurons. The standard configuration file for this
network is som_cube.cfg (one 2D display only).
Pattern-Files: som_cube.pat
==============
The eight training patterns (only input patterns, no output pattern)
are the verteces of a three-dimensional cube of size 2x2x2 centered at
the origin, with vertex coordinates from (-1, -1, -1) to (1, 1, 1).
Miscellaneous: som_cube.cont
==============
The file som_cube.cont is a control file for the tool convert2snns
(in the SNNS tools subdirectory) to create a SOM with 3 components
and grid 16x16.
Hints:
======
Note to open the control panel before opening the special Kohonen
panel. This special panel allows you to view vector component maps of
any input dimension of the map.
Note one point of frequent confusion: The button WINNER in the Kohonen
panel tests *all* patterns of the currently active pattern file and for
each winner neuron it displays the pattern number of the input pattern
for which the neuron was winner, on top of the neuron. To see these
numbers you must have specified "units top : ON SHOW winner" in the
SETUP panel of the 2D network display window. (These numbers are
better recognized if the neuron grid is made smaller and the values
displayed at the bottem of the neurons are switched off).
If a neuron is actived twice as winner for different input patterns,
the later input pattern number overwrites the earlier number.
Therefore, it may happen that not all pattern numbers appear on top of
the neurons.
The confusion stems from the fact that upon pressing the WINNER button
in the Kohonnen panel, *all* patterns are tested and the activation of
the last input pattern is displayed in the 2D display. This is always
the same input pattern and the same Kohonen neuron, regardless of what
pattern has been specified in the remote panel before.
=============================================================================
End of README file
=============================================================================