Chess Neural Network: ANOTHER VICTORY FOR OCTAVIUS!
Luke (bigpond)
around with neural networks on such a machine. I researched the internet
back then and found nothing on Chess and ANNs, so I took it up as a "hobby".
Within my first six months of creating and training Octavius, I recieved
interest from Sao Paul University, University of California, and the Karpov
Chess Institute.
I now have a Pentium 2.4GHz with 1Gb RAM - delicious...
So, over these 4 and a half years I have been training several different
types of neural networks (all using the Octavius program) and the other
day... for the first time EVER... he beat me using ONLY the neural network
(NO material analysis).
I admit - I blundered... but generally, that is how one loses a game of
chess. Nevertheless, Octavius was the victor. I am used to play testing him,
and I am a reasonable player, so my thinking time is about 1 or 2 seconds on
average. Octavius's thinking time is about 5 seconds (it's the training that
takes a long time!).
Here are the current ANN specs for those interesed:
Input Nodes: 768
Hidden Nodes [Layer 1]: 1024
Hidden Nodes [Layer 2]: 512
Output Nodes: 1
Total Nodes: 2305
Total Connections: 1,311,232
Total Backpropogations to Date: 284,552
Unfortunately, I am still yet to set up a web site for Octavius as my plan
has always been to make him available for free public download. Though you
might still find him using search engines on some older sites.
I am currently working on Version 2.0 with some nice visual additions for
analysis, as well as a "tutor" mode.
Luke vs Octavius (P4Oct4.ann)
26 April 2004
1. e2e4 g8f6
2. b1c3 e7e5
3. f1b5 c7c6
4. b5a4 f8c5
5. g1f3 e8g8
6. e1g1 d7d6
7. d2d3 c8g4
8. a4b3 f6h5
9. a2a3 f8e8
10. b3a2 d8f6
11. c1g5 f6g6
12. b2b4 c5b6
13. g5h4 h5f4
14. g1h1 b8d7
15. g2g3 f4h3
16. d1e2 h7h5
17. a1e1 d6d5
18. e4d5 c6d5
19. c3d5 e5e4
20. d3e4 e8e4
21. e2e4 f7f5
22. e4e7 g4f3
Cheers,
Luke Pellen
lpe...@netspace.net.au
lukep...@yahoo.com.au
Stephen Waits
>
> So, over these 4 and a half years I have been training several different
> types of neural networks (all using the Octavius program) and the other
> day... for the first time EVER... he beat me using ONLY the neural network
> (NO material analysis).
Congrats!
> Unfortunately, I am still yet to set up a web site for Octavius as my plan
> has always been to make him available for free public download. Though you
> might still find him using search engines on some older sites.
>
> I am currently working on Version 2.0 with some nice visual additions for
> analysis, as well as a "tutor" mode.
Have you considered some kind of network client so that he (it is a boy
right?) can play other people on the net? Or, maybe he could play on
FICS (is the protocol open?)??? Either way, this would get him lots
more real-world training than you can provide.
--Steve
st...@tropheus.demon.co.uk
<lpe...@bigpond.com> wrote:
>Unfortunately, I am still yet to set up a web site for Octavius as my plan
>has always been to make him available for free public download. Though you
>might still find him using search engines on some older sites.
>
Luke,
Do you want some free web space? I have lots of space to spare for
neural network applications. Send me an e-mail if you are interested.
Steve
--
EasyNN-plus. The easy way to build neural networks.
Build networks from numeric, text and image files.
http://www.easynn.com