Page 18 of 18
Re: Komodo 12 and MCTS
Posted: Sun Jun 03, 2018 7:39 pm
by nabildanial
https://docs.google.com/forms/d/e/1FAIp ... g/viewform
This is related to Komodo? Is the Komodo Team planning to develop a NN engine?
Re: Komodo 12 and MCTS
Posted: Sun Jun 03, 2018 8:19 pm
by mjlef
Hurnavich wrote: ↑Sun Jun 03, 2018 6:45 pm
Hi, ok so box checked and running MCTS i'm thick here does the display output show me something different or just shows the best move found like normal K
Thanks
in MCTS mode, there is not iterative depth, so instead, Komodo outputs the best move, pv, eval... every 3 seconds. Note that Multi-PV is basically free in MCTS mode, so I suggest using the + in the Fritz GUI to show a few lines.
Re: Komodo 12 and MCTS
Posted: Sun Jun 03, 2018 8:19 pm
by mjlef
We want to look at using neural networks for a bunch of different things in engines.
Re: Komodo 12 and MCTS
Posted: Sun Jun 03, 2018 11:52 pm
by main line
Can somebody with MCTS version reproduce 21. Bg5 !! from AlphaZero - Stockfish game
FEN:
rn3r1k/pn1p1ppq/bpp4p/7P/4N1Q1/6P1/PP3PB1/R1B1R1K1 w - - 0 21
Full game:
https://lichess.org/study/EOddRjJ8
Re: Komodo 12 and MCTS
Posted: Mon Jun 04, 2018 1:56 am
by mjlef
main line wrote: ↑Sun Jun 03, 2018 11:52 pm
Can somebody with MCTS version reproduce 21. Bg5 !! from AlphaZero - Stockfish game
FEN:
rn3r1k/pn1p1ppq/bpp4p/7P/4N1Q1/6P1/PP3PB1/R1B1R1K1 w - - 0 21
Full game:
https://lichess.org/study/EOddRjJ8
I ran it for a minute on a 4 core laptop using the soon to be released update, and Komodo in MCTS mode prefers b4:
23/32 01:12 79k 1k +0.14 b2-b4 d7-d5 Ne4-d2 Qh7-c2 a2-a4 Nb7-d8 Qg4-f4 Kh8-g8 b4-b5 Nd8-e6 Qf4-d6 Ba6-b7 Re1xe6 f7xe6 Qd6xe6+ Rf8-f7 Qe6-e8+ Rf7-f8 Qe8-e6+
I can try it on a bigger machine when it is not running games. If anyone else has Alpha Zero positions, I would be happy to try them.
Mark
Re: Komodo 12 and MCTS
Posted: Mon Jun 04, 2018 2:26 am
by cc2150dx
main line wrote: ↑Sun Jun 03, 2018 11:52 pm
Can somebody with MCTS version reproduce 21. Bg5 !! from AlphaZero - Stockfish game
FEN:
rn3r1k/pn1p1ppq/bpp4p/7P/4N1Q1/6P1/PP3PB1/R1B1R1K1 w - - 0 21
Full game:
https://lichess.org/study/EOddRjJ8
Komodo MCTS, Infinite analysis 3 Core
FEN: rn3r1k/pn1p1ppq/bpp4p/7P/1P2N1Q1/6P1/P4PB1/R1B1R1K1 b - b3 0 21
Komodo-12-64bit:
33/55 1:05:17 8,385k 0 +0.15 b2-b4 d7-d5 Ne4-c3 Ba6-c4 Bc1-b2 Nb7-d6 a2-a4 Nb8-a6 b4-b5 Na6-b4 Qg4-f4 Rf8-d8 b5xc6 Nb4-d3 Qf4-d2 Ra8-c8 Re1-e7 Nd3xb2 Qd2xb2 Rc8xc6 Nc3xd5 Bc4xd5 Bg2xd5 Rc6-c2 Qb2-a3 Qh7-f5 Qa3-f3 Rc2-d2 Bd5-c6 Qf5-a5 Ra1-d1 Rd2xd1+ Qf3xd1
Re: Komodo 12 and MCTS
Posted: Mon Jun 04, 2018 2:36 am
by cc2150dx
mjlef wrote: ↑Mon Jun 04, 2018 1:56 am
the soon to be released update,
I can't wait
Re: Komodo 12 and MCTS
Posted: Mon Jun 04, 2018 5:12 am
by Werewolf
mjlef wrote: ↑Mon Jun 04, 2018 1:56 am
main line wrote: ↑Sun Jun 03, 2018 11:52 pm
Can somebody with MCTS version reproduce 21. Bg5 !! from AlphaZero - Stockfish game
FEN:
rn3r1k/pn1p1ppq/bpp4p/7P/4N1Q1/6P1/PP3PB1/R1B1R1K1 w - - 0 21
Full game:
https://lichess.org/study/EOddRjJ8
I ran it for a minute on a 4 core laptop using the soon to be released update, and Komodo in MCTS mode prefers b4:
23/32 01:12 79k 1k +0.14 b2-b4 d7-d5 Ne4-d2 Qh7-c2 a2-a4 Nb7-d8 Qg4-f4 Kh8-g8 b4-b5 Nd8-e6 Qf4-d6 Ba6-b7 Re1xe6 f7xe6 Qd6xe6+ Rf8-f7 Qe6-e8+ Rf7-f8 Qe8-e6+
I can try it on a bigger machine when it is not running games. If anyone else has Alpha Zero positions, I would be happy to try them.
Mark
Sounds good. What’s the new thread limit?
Re: Komodo 12 and MCTS
Posted: Mon Jun 04, 2018 8:14 am
by lkaufman
12 Threads.