TonyJH and I were both watching Eric Rosen a few weeks ago on Twitch, playing against DuckChessZero:
Facing the World's STRONGEST Duck Chess Engine
https://www.twitch.tv/imrosen/video/1711412379
Here is the web interface to play against DuckChessZero:
http://peter.website/duck-chess/analysis
"This is an engine + analysis interface for duck chess that runs entirely in the browser, with the networks trained with tabula rasa self-play, like AlphaZero"
https://github.com/petersn/duckchess
Duck Chess
Moderator: Ras
-
- Posts: 121
- Joined: Sat Dec 03, 2022 9:28 pm
- Full name: Rick Groszkiewicz
Re: Duck Chess
Retired actuary and software developer. I love chess, coffee, wine and food
-
- Posts: 28353
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Duck Chess
I managed to let the diagram in the Duck Chess article at chessvariants.com play in such a way that it also uses the Duck to reserve its own next move, rather than only attempting to block the best move of the opponent. This should cover the essence of Duck Chess; it is not really perfet, because it only attempts blocking the opponent's best move on an adjacent square. But the errors it would make because of this are probably insignificant to the errors it would make because it does not search deep enough.
To implement this I replaced the AlphaBeta function of the diagram's innate AI by one that tracks 3 scores, rather than just the best (and only takes a beta cutoff if the lowest of these three exceeds beta): one for when the best move is blocked on the adjacent square, one for when the Duck is placed to prevent the opponent's best Duck move, and one for all other Duck locations. It then returns the lowest of these scores as best score, together with the Duck location required to achieve it. And it returns the second-lowest as the 'exception score' for when the Duck was already there, so that it could not stay in the best position (the 'duckzwang score').
It should make an entertaining opponent.
As a spin-off I also made diagrams that can play Blue-Chip Chess and Refusal Chess.
To implement this I replaced the AlphaBeta function of the diagram's innate AI by one that tracks 3 scores, rather than just the best (and only takes a beta cutoff if the lowest of these three exceeds beta): one for when the best move is blocked on the adjacent square, one for when the Duck is placed to prevent the opponent's best Duck move, and one for all other Duck locations. It then returns the lowest of these scores as best score, together with the Duck location required to achieve it. And it returns the second-lowest as the 'exception score' for when the Duck was already there, so that it could not stay in the best position (the 'duckzwang score').
It should make an entertaining opponent.
As a spin-off I also made diagrams that can play Blue-Chip Chess and Refusal Chess.
-
- Posts: 183
- Joined: Tue Jun 20, 2006 4:41 am
- Location: USA
Re: Duck Chess
I don't remember that. I am interested to know more about DuckChessZero though. I don't think I heard of it before now.rgrosz789 wrote: ↑Mon Jan 30, 2023 4:37 am TonyJH and I were both watching Eric Rosen a few weeks ago on Twitch, playing against DuckChessZero:
Facing the World's STRONGEST Duck Chess Engine
https://www.twitch.tv/imrosen/video/1711412379
Looks like that twitch video is only available to subscribers of his channel.
-
- Posts: 121
- Joined: Sat Dec 03, 2022 9:28 pm
- Full name: Rick Groszkiewicz
Re: Duck Chess
Sorry, my mistake Tony.
I saw a comment on that twitch stream from @fifdspence about testing TJDuckChess against DuckChessZero. Eric Rosen found a few bugs, actually won two out of three games.
I saw a comment on that twitch stream from @fifdspence about testing TJDuckChess against DuckChessZero. Eric Rosen found a few bugs, actually won two out of three games.
Retired actuary and software developer. I love chess, coffee, wine and food
-
- Posts: 121
- Joined: Sat Dec 03, 2022 9:28 pm
- Full name: Rick Groszkiewicz
Re: Duck Chess
Yes, it does - I lost a few games playing against it (silly mistakes).hgm wrote: ↑Mon Jan 30, 2023 11:19 am I managed to let the diagram in the Duck Chess article at chessvariants.com play in such a way that it also uses the Duck to reserve its own next move, rather than only attempting to block the best move of the opponent.
...
It should make an entertaining opponent.
Retired actuary and software developer. I love chess, coffee, wine and food
-
- Posts: 5
- Joined: Wed Dec 28, 2016 12:46 pm
- Location: Italy
Re: Duck Chess
Thanks!MTaktikos wrote: ↑Tue Jan 24, 2023 8:13 pm Good news, Belzedar and I have published today our first official duck nnue net, can be downloaded from
https://fairy-stockfish.github.io/nnue/
Have updated here the complete winboard suite with the strongest FSFDuck so far (avx2 compile), combined with this net, and the other duck engines:
https://filehorst.de/d/etcBsCHx
Enjoy!
Nice improvement in strength.
I compiled myself the exe because avx2 doesn't run on my system, maybe it is a bit slow, anyway these are my results:
Code: Select all
Rank Name Elo + - games score oppo. draws
1 TJchessDuck 0.2 502 149 149 70 97% -108 3%
2 FSFduck nnue 130 109 109 70 57% 52 6%
3 FSFduck -256 121 121 70 13% 217 3%
4 MayhemDuck 1.9 (UCI2WB) -376 169 169 30 10% 125 7%
-
- Posts: 5287
- Joined: Thu Mar 09, 2006 9:40 am
- Full name: Vincent Lejeune
Re: Duck Chess
What was the score ?rgrosz789 wrote: ↑Mon Jan 30, 2023 4:37 am TonyJH and I were both watching Eric Rosen a few weeks ago on Twitch, playing against DuckChessZero:
Facing the World's STRONGEST Duck Chess Engine
https://www.twitch.tv/imrosen/video/1711412379
-
- Posts: 121
- Joined: Sat Dec 03, 2022 9:28 pm
- Full name: Rick Groszkiewicz
Re: Duck Chess
Eric only played three games. He won two games, but should have lost all three.
In the 2nd game, DuckChessZero let Eric take its King, which was clearly a programming bug (it had a mate in 2). In the 3rd game, DuckChessZero allowed Eric to checkmate with a Knight, also clearly a programming bug (Eric's position was very bad).
In the 2nd game, DuckChessZero let Eric take its King, which was clearly a programming bug (it had a mate in 2). In the 3rd game, DuckChessZero allowed Eric to checkmate with a Knight, also clearly a programming bug (Eric's position was very bad).
Retired actuary and software developer. I love chess, coffee, wine and food
-
- Posts: 121
- Joined: Sat Dec 03, 2022 9:28 pm
- Full name: Rick Groszkiewicz
Re: Duck Chess
I have been using a Windows Virtual Machine (using VMWare) to run engine tests. I can run this VM on a Windows PC, and everything works fine.MTaktikos wrote: ↑Tue Jan 24, 2023 8:13 pm Good news, Belzedar and I have published today our first official duck nnue net, can be downloaded from
https://fairy-stockfish.github.io/nnue/
Have updated here the complete winboard suite with the strongest FSFDuck so far (avx2 compile), combined with this net, and the other duck engines:
https://filehorst.de/d/etcBsCHx
Enjoy!
Now I'm trying to set up this VM so it can run 24x7 on a separate PC. But FSFDuck does not work when I run the exact same VM on a Linux PC.
Any idea why that might happen?
http://www.softwarepolish.com/images/FSFDuck-VM.png
I can't add the image ("Sorry, the board attachment quota has been reached")
Retired actuary and software developer. I love chess, coffee, wine and food
-
- Posts: 59
- Joined: Fri Oct 25, 2019 7:58 pm
- Full name: Michael Taktikos
Re: Duck Chess
First thing that in the new VM can be tried is to reduce the hardware requirements: in the common options of Winboard, try to set instead 20 cores single core for a first test.rgrosz789 wrote: ↑Sat Feb 11, 2023 6:17 pmI have been using a Windows Virtual Machine (using VMWare) to run engine tests. I can run this VM on a Windows PC, and everything works fine.MTaktikos wrote: ↑Tue Jan 24, 2023 8:13 pm Good news, Belzedar and I have published today our first official duck nnue net, can be downloaded from
https://fairy-stockfish.github.io/nnue/
Have updated here the complete winboard suite with the strongest FSFDuck so far (avx2 compile), combined with this net, and the other duck engines:
https://filehorst.de/d/etcBsCHx
Enjoy!
Now I'm trying to set up this VM so it can run 24x7 on a separate PC. But FSFDuck does not work when I run the exact same VM on a Linux PC.
Any idea why that might happen?
http://www.softwarepolish.com/images/FSFDuck-VM.png
I can't add the image ("Sorry, the board attachment quota has been reached")
Additionally, there may be compatibility issues with the virtualization software or the operating system on the Linux machine. For this case, here is a Linux native compile of the engine, again avx2: https://filehorst.de/d/eDqcvBeF
You have to unzip it in the same folder where the nnue is, i.e. in the same folder where the Windows engine FSFDuck.exe is. May be you want also to install the Linux native xboard GUI with sudo apt install xboard,
but for a first test try to run it from the bash with ./FSFDuck to confirm if it works
And last not least, the issue may be related to the CPU architecture of the Linux machine's processor. If the Linux machine's CPU architecture is different from the Windows PC's CPU architecture, the engine may not be able to run on the Linux VM. Then perhaps a bmi2 compile, a modern (popcnt) compile or an old simple x64 compile will work
_____________________
https://github.com/mtaktikos?tab=repositories
https://github.com/mtaktikos?tab=repositories