This is a copy of the neural network simulator used to run the
simulations in the Amateur Scientist article in the September 1992
issue of Scientific American.

The files are:

sim.c - the source code for the simulator
sim.h - header file for the simulator
4x4.train - training data for a 4-x-4 encoder
4x4.test  - testing data for a 4-x-4 encoder
Makefile  - a simple makefile for building the simulator
Imakefile - a simple Imakefile (if you have imake)

The way it's written at present it runs a 4-3-4 encoder. You can
change the number of units in each layer by editing the file sim.h.

Once the simulator is running, it understands 6 commands:

init	    - (re)initialize the network.
file <file> - set the pattern file
train <n>   - train on <n> patterns from the file
test <n>    - test on <n> patterns from the file
dump	    - dump network values (weights, activations etc)
quit	    - quit the simulator

This isn't necessarily the best neural net simulator I've ever seen,
but it does the job in a simple, straight-forward manner. It should
compile on most any platform with an ANSI C compiler.

Drew van Camp

/--- Drew van Camp --------------- drew@cs.toronto.edu-----------\
| Dept. of Computer Science, University of Toronto               |
| 6 Kings College Road, Toronto, Ontario   Voice: (416) 978-7403 |
| CANADA M5S 1A4                           Fax:   (416) 978-1455 |
\----------------------------------------------------------------/
