MNG-Package Readme                                 Mon Jul 19 14:17:09     2004


Merging Neural Gas (MNG) is described in detail in
M. Strickert and B. Hammer, Self-Organizing Context Learning. 
In: M.Verleysen (ed.), 
    European Symposium on Artificial Neural Networks (ESANN 2004),
    D-side publications, pp. 39-44, 2004


The mng.c is an exact implementation of the described algorithm
*except for an additional postsynaptic inhibition*
that prevents neuron to be selected as winner in two successive steps.


To get started, type 
> make
This should compile on cygwin/linux gcc platforms without errors.

Type 
> . tr_bauto
and wait for some minutes to perform a MNG training of
the unbiased P(0|0) = P(1|0) = P(0|1) = P(1|1) = P(0) = P(1) = 0.5
binary automaton.
A number of *bauto*dat files and a *bauto*bz2 file will be generated.


See tr_bauto.sh for further information.
It may be a good advice to use an editor with syntax highlighting
for source code studies.

Everything remaining inside the package is distributed under
the GNU GENERAL PUBLIC LICENSE given in LICENSE.txt.

Good luck.

  Marc Strickert (mstricke@uos.de)



Files in this package:

backup.sh              # to be called one level above surce directory
batomaton2baum.sh      # convert strings into a gnuplotable tree
bauto_27_57.awk        # another binary tree with 2/7 vs 5/7 probability
bauto_5_5.awk          # unbiased automaton
bautomaton.awk         # Voeglins 3/7 vs 4/7 automaton version
getfinalneurons.awk    # extract neurons after training (unused)
makefile               # see this for further details
mng.c                  # the MNG c source code
mng.h                  # the MNG c source code header 
mngclassify.awk        # a slow awk classifier to be replaced by mngclassify.c
mngclassify.c          # classify new data with given network
neighsize.c            # calculate decay parameter values
readme.txt             # guess it
rfield.awk             # extract receptive field by using the test section
tr_bauto.sh            # instructive example for the automaton task

// end of file
