A WIN FOR OCTAVIUS! - Chess ANN
Luke (bigpond)
Octavius is a Chess ANN which I wrote back in about 1998/99 using a Pentium
166MHz with 32Mb RAM...
Here was the brief blurb I wrote when I made Octavius 1.4 available (just
for a bit of background):
-------------------------
Octavius V1.4 is now available for download from:
http://home.seol.net.au/luke/Octavius/
http://dove.net.au/~lpellen/Octavius/ (new site)
Features of V1.4 include:
* The ability to dynamically graph the output nodes of the neural network
during training and play.
* The ability to plot a 3 dimensional board image graph representing the
neural network input.
* More compact and resizable chess board interface.
* Tool tips.
* Support for multiple chess fonts (many nice chess fonts are easily
available through the internet).
* The ability to save training sets for re-use.
* The ability to load and save neural network settings.
* The option of using a pre-defined opening book to enhance play.
* The ability to edit the configuration of the pieces on the chess board.
* Online documentation is available (v1.3)
This is rather terse, and I am still developing the documentation
as I fully develop the software. I am also developing more tool tips
and in-program help.
-------------------------
Well, now I have a Pentium 2.4GHz with 1Gb RAM I am quite a bit happier...
Here is Octavius's first victory against a "good" player. John Pellen (my
brother) is a good club level player, and though he may not have played
perfectly, he didn't play like a woodpusher, and he lost against Octavius.
In the following game it must be noted that we gave Octavius some material
analysis backing (selected 3 candidate moves for quick "capture" analysis,
the result being a weighted figure which is then combined with the ANN
evaluation)
[Event "Octavius P4 Test"]
[Site "LUKES COMPUTER"]
[Date "2003.08.20"]
[Round "-"]
[White "John Pellen"]
[Black "Octavius V1.3 - P4oct2.ann"]
[Result "0-1"]
1. d2d4 d7d5 2. g1f3 c8f5 3. c1g5 b8d7 4. b1c3 c7c6 5. g2g3 d8b6 6. b2b3
O-O-O 7. c3a4 b6b4+ 8. c2c3 b4b5 9. f3h4 f5e4 10. f1g2 g8h6 11. f2f3 e4f3
12. g2f3 h6g8 13. e2e4 d5e4 14. c3c4 b5g5 15. f3e4 g5e3+ 16. d1e2 e3d4 17.
a1d1 d4e5 18. O-O e7e6 19. f1f7 f8c5+ 20. a4c5 e5c5+ 21. g1h1 d7b6 22. f7g7
g8f6 23. h4f3 f6e4 24. g7g4 e4c3 25. e2e6 b6d7 26. g4g5 c5b6 27. f3e5 c3a2
28. d1d7 d8f8 29. d7h7+ c8b8 30. e6d6+ b8c8 31. e5d7 f8f1+ 32. h1g2 f1g1+
33. f2g3 b6b3+ 34. f3g4 b3c4+ 35. d6f4 c4e6+ 36. f4f5 e6e2+ 37. f5f3 e2e6+
38. f3f5 e6e2+ 39. g4h4 e2h2+ 40. h4g4 g1g3+ 41. g4f4 h2f2+ 42. f4e5 f2b2+
43. e5d6 b2d4+ 44. d6e7 h8h7+ 45. f5h7 d4d7+ 46. e7f6 d7h7 47. g5g3 b7b6 48.
f6e6 a2c1 49. g3c3 h7d7+ 50. e6e5 d7g7+ 51. e5d6 g7f6 {Black mates}
I think I'm starting to find a good configuration (after much trial and
error experimentation). Octavius can be configured in many ways and is very
flexible. I hope to release version 2.0 shortly (when I get the time! -for
anyone who knows/remembers me out there I'm currently working as an
actor/stand-up comedian/musician/artist/graphic
designer/writer/programmer/website developer... basically whatever gigs I
can get hold off... see esoteric artistic outpourings at
http://www.users.bigpond.com/gelsonsarah/lpellen/anow/ ).
Thanks to those who have supported Octavius and for providing feedback.
Cheers,
Luke Pellen.
SeeB...@seebelow.nut
>
> Hello all,
>
> Octavius is a Chess ANN which I wrote back in about 1998/99 using a Pentium
> 166MHz with 32Mb RAM...
>
> Here was the brief blurb I wrote when I made Octavius 1.4 available (just
> for a bit of background):
Your Octavius project is very interesting; you have accomplished
something significant! I hope you will answer some questions for me:
1. Does your software include a tree of possible future moves by both
players, with the NN evaluating the leaves of the tree? Or does the NN
simply take the board as input and generate a move directly?
2. Can you describe the details of the net? for example, how many
neurons, how many in each layer, how many layers, what activation
function.
3. What language is it written in?
Thank You
Mitchell Timin
--
"Many are stubborn in pursuit of the path they have chosen, few in
pursuit of the goal." - Friedrich Nietzsche
http://annevolve.sourceforge.net is what I'm into nowadays.
Humans may write to me at this address: zenguy at telus dot net
Luke (bigpond)
> something significant! I hope you will answer some questions for me:
>
> 1. Does your software include a tree of possible future moves by both
> players, with the NN evaluating the leaves of the tree? Or does the NN
> simply take the board as input and generate a move directly?
>
> 2. Can you describe the details of the net? for example, how many
> neurons, how many in each layer, how many layers, what activation
> function.
>
> 3. What language is it written in?
>
> Thank You
>
> Mitchell Timin
1) Octavius, in effect, is a positional evaluation algorithm. He searches
only 1 move ahead and evaluates the positions from there.
2) The details of the network are open to customisation. The one I am
presently using has 64 input nodes, 1024 hidden nodes, and 1 output node.
Sigmoid activation function.
3) Here's the bit where you laugh.... Visual Basic (4 originally, upgraded
now to 6). I have programmed in many languages (Assembler, C, COBOL, Forth,
blah blah blah), and being raised with Vic20s and Commodore 64s BASIC is my
"comfortable" language... the language with which I "mentally translate" all
other languages... VB offers the advantage of very quick development time,
which I love.
Greg Heath
news:<f46ob.171130$bo1....@news-server.bigpond.net.au>...
> > Your Octavius project is very interesting; you have accomplished
> > something significant! I hope you will answer some questions for me:
> >
> > 1. Does your software include a tree of possible future moves by both
> > players, with the NN evaluating the leaves of the tree? Or does the NN
> > simply take the board as input and generate a move directly?
> >
> > 2. Can you describe the details of the net? for example, how many
> > neurons, how many in each layer, how many layers, what activation
> > function.
> >
> > 3. What language is it written in?
> >
> only 1 move ahead and evaluates the positions from there.
>
> 2) The details of the network are open to customisation. The one I am
> presently using has 64 input nodes, 1024 hidden nodes, and 1 output node.
> Sigmoid activation function.
How do you tell the best move from from one output?
Hope this helps.
Greg