Page 1 of 1

Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Wed Feb 13, 2019 10:12 pm
by Graham Banks

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Wed Feb 13, 2019 10:14 pm
by Graham Banks

Code: Select all

CCRL 40/40 Rating List - Custom engine selection
981277 games played by 2362 programs, run by 22 testers
Ponder off, General books (up to 12 moves), 3-4-5 piece EGTB
Time control: Equivalent to 40 moves in 40 minutes on Athlon 64 X2 4600+ (2.4 GHz), about 15 minutes on a modern Intel CPU.
Computed on February 9, 2019 with Bayeselo based on 981'277 games
Tested by CCRL team, 2005-2019, http://computerchess.org.uk/ccrl/4040/

Rank               Engine                Elo   +    -   Score  AvOp  Games
1 Komodo 12.3 MCTS 64-bit            3234  +16  -16  52.5%  -15.3  1220
  Komodo 12.2.2 MCTS 64-bit          3213  +16  -16  62.4%  -83.3  1350
  Komodo 12.1.1 MCTS 64-bit          3079  +16  -16  46.9%  +24.3  1430

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Thu Feb 14, 2019 5:44 am
by Leo
I hope they keep making 200 Elo improvements.

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Thu Feb 14, 2019 2:03 pm
by henk2
Leo wrote: Thu Feb 14, 2019 5:44 am I hope they keep making 200 Elo improvements.
KMCTS has zero future, absolutely none

It will just be Leela from now on with others losing interest in hand-tuned engines.
Other NN-engines won't pop up because training requires too many resources for a single or few people.

This is the end of computer chess.
The same thing that happened with computer draughts (10x10) when the top 2 engines started drawing literally every game with long TC.
Except this time it will be a lone wolf at the top of the food chain. People will lose interest in fishtest and it will stagnate.
Houdart has stopped updated Houdini a long while ago.
Komodo will lose.
Andsacs has hit his ceiling
Booot and Xiphos are irrelevant.
Laser only stands a chance if we extend to time control to 40 moves in 24 hours.

Then again, Leela isn't the only one to blame.
Open-source Stockfish already killed competitiveness and commercial interest.
Leela just killed the rest of it too.

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Sat Feb 16, 2019 5:53 pm
by lkaufman
Why so pessimistic? Komodo MCTS continues to progress. Lc0 shows that the future is with GPUs, but the possibilities for utilizing them are endless. I think we are just starting on a very interesting period in computer chess. Now the draw problem is another matter; chess is too drawish under current rules, this is not specifically a computer chess problem. But there are many solutions to it, including the TCEC solution of selecting unbalanced openings. If the opening positions are chosen right on the win/draw line, half of the games will be decisive as long as chess has not been solved. Another simple solution is to give White double (or triple...) time, whatever is needed to bring the percentage of games ending in White wins close to 50%.

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Sat Feb 16, 2019 7:39 pm
by Guenther
henk2 wrote: Thu Feb 14, 2019 2:03 pm
Leo wrote: Thu Feb 14, 2019 5:44 am I hope they keep making 200 Elo improvements.
KMCTS has zero future, absolutely none
your posts are hilarious (see your post 9 days before):

henk2 » Tue Feb 05, 2019 6:27 pm wrote: Re: chess programming - predictions for the next 5 years

KMCTS is promising too and maybe it will eventually turn out stronger than regular K, maybe this will spread to other engines. Maybe we'll get hybrid CPU/GPU engines too.

Re: Komodo 12.3 MCTS 64-bit Gauntlets for CCRL 40/40

Posted: Sun Feb 17, 2019 12:46 pm
by henk2
Guenther wrote: Sat Feb 16, 2019 7:39 pm
henk2 wrote: Thu Feb 14, 2019 2:03 pm
Leo wrote: Thu Feb 14, 2019 5:44 am I hope they keep making 200 Elo improvements.
KMCTS has zero future, absolutely none
your posts are hilarious (see your post 9 days before):

henk2 » Tue Feb 05, 2019 6:27 pm wrote: Re: chess programming - predictions for the next 5 years

KMCTS is promising too and maybe it will eventually turn out stronger than regular K, maybe this will spread to other engines. Maybe we'll get hybrid CPU/GPU engines too.
I cycle my clozapine with shrooms and marijuana.