home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Collection of Education
/
collectionofeducationcarat1997.iso
/
COMPUSCI
/
NETSTUFF.ZIP
/
READ.ME
< prev
next >
Wrap
Text File
|
1991-02-27
|
4KB
|
78 lines
This diskette contains several files which are examples of neural
networks. All examples have been placed in the public domain.
The diskette is free, with my only benefit being the opportunity
to advertise my book, A Practical Guide to Neural Networks, by
Nelson and Illingworth, published by Addison-Wesley in late 1990.
Diskette contents are:
TSP : This is the familiar Traveling Salesperson Problem,
written in Pascal, and made available by AI Expert
Magazine (along with several other AI programs from a
diskette called "AI Sampler"). Authors are Bill and Bev
Thompson of Knowledge Garden Inc.
Use the space bar to exit when tired of the iterations.
Additional documentation exists in the source code file.
HAM : A C language net example created by David Leasure. This
routine is a hamming classification network described in
IEEE ASSP April 1987 by Richard P. Lippmann, pg. 9. It
examines an input 5 x 7 pattern for closest match with
its given set of exemplars (representations of 0 - 9).
Leasure has made a couple of modifications, which he
documents in the source code, and notes some
inefficiencies. I have compiled the original source so
both the .c and .exe files appear here.
HAMX : A modification of the above by M. Nelson to allow for
greater interactivity. A data filename may be passed in
on the command line (use files ham1.dat, ham2.dat,
ham3.dat). Or, if no file is passed in, the user is
prompted through the creation of their own 5 x 7 matrix
for a possible numeral representation to be recognized by
the net, and then given an opportunity to save the pattern
which they have just created.
Although Lippmann says the net should always converge, the
representation created by Leasure is not always able to
select a winner (try ham3.dat, for example).
DELTA : This C language code appears in Appendix A of the book
Adaptive Pattern Recognition and Neural Networks, by
Yoh-Han Pao. He calls the program "A Generalized Delta
Rule (GDR) Net Program for Supervised Learning." The
program provides code in support of the following tasks:
1. Specify net architecture
2. Learn weights and thresholds with use of training
set patterns.
3. Use net to obtain output values for new patterns,
either for classification purposes or for estimation
of values of associated attributes.
I have typed the code in almost exactly as it appears,
correcting for one obvious error and occasionally changing
the printf() statements slightly. In addition, my version
decreases the size of some of the defined constants in
order to fit within the memory constraints of my machine.
This program does write to a couple of extra files, so be
sure space is available.
I compiled it under Microsoft QuickC 2.0, and both the
source and .exe file are included here. It was a lot of
typing--hope I got it all correct. The "scenario" file
gives a sample run and shows how to use the program.
(Note: Appendix B in the same book gives C source for an
unsupervised learning program based on discovery of
cluster structure.)
Marilyn M. Nelson
Associated Consultants
1046 CR 500
Bayfield, CO 81122