Load the 'letters' pattern file (in the SNNS- examples directory) at this stage. The network is a pattern associator that can be trained to map an input image (5x7 pixel representation of letters) into output units where each letter is represented by an output unit.
All training and testing is done via the option. I will
discuss the most important features one by one. The panel consists of
two parts. The top part controls the parameters defining the training
process, the bottom three rows are blanks that have to be filled in to
define the learning rates etc. and the range over which weights will
be randomly distributed when the network is initialised. The defaults
for the learning parameters are (0.2 0 0 0 0) while the default weight
setting is between 1 and -1 (1.0 -1.0 0 0 0).