Page 3 of 4

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Mon Nov 04, 2019 8:37 pm
by Chessqueen
Ovyron wrote:
Mon Nov 04, 2019 4:16 pm
I play three kinds of chess:

-"Just me" chess: Where I bring only my brain to the board, and match it against someone else's brain.

-"Centaur" chess: Where I play the absolute best moves that I can find, with my brain, engines, databases, and whatever I have access to. Unfortunately this takes much more effort and time than anything else, so games are played in time controls that make them take years, and by the time they're over I might have already forgotten what happened in them or why I played some moves (sometimes it happens that I transpose into a new position I hadn't seen before, but that it has plenty of analysis done by myself, that I don't remember doing. Games just take too long.)

-"No me" chess: In this one my only influence is the book that tells the engine what to play in the opening, after it's over the engine is on its own. My challenge is to reach positions where my engine will perform well. The fun part is realizing that at this level, if I was playing as Centaur with 1 hour per move I'd be winning all my games against other engines, so I'm not doing it because my book doesn't include the moves I'd play as Centaur. It means to be the best I just need to put the winning moves in my book, because the opponents only starts with 5 minutes, if I include in my book 5 minute moves I'm already playing each move at the level of the entire game.

Bla-------- Bla-------- Bla------- Bla-------- Bla etc........................................................................................Bla -------Bla---------etc.......

What would happen if before the game started players in secret picked one of the 960 positions and then they revealed simultaneously and the game started from an Asymmetric Chess position? If both would pick the opening position from chess it'd speak volumes about how most of the other opening positions are garbage.

To be interested in a game you have to get engaged with it, I could never get engaged with FRC, I would much rather play normal chess or even Atomic chess, it's because of some of these reasons, that's why I have zero interest in FRC, and why I won't be following its TCEC games.
Very well for that long explanation :mrgreen:

WOW Take a look at the audience or viewers 30 minutes ago there were 320 people watching these FRChess960 games , Now it went down to 270 still great amount of people interested in Fischer Random, which means that TCEC Does NOT need you to be interested in FRChess960. They can still make it more interesting by including the same Openings played between GM Carlsen Vs GM So, and it will give us a different view of how the engines manage to play the same positions which Carlsen played soo badly
https://tcec-chess.com/

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Mon Nov 04, 2019 11:35 pm
by Ovyron
I never said they needed me.

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Tue Nov 05, 2019 12:49 pm
by Chessqueen
Ovyron wrote:
Mon Nov 04, 2019 11:35 pm
I never said they needed me.
Sorry for that :mrgreen:

I noticed that Old Houdini, Fire, and Ethereal are doing much much better than KomodoMCTS https://tcec-chess.com/

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Tue Nov 05, 2019 4:46 pm
by lkaufman
Chessqueen wrote:
Tue Nov 05, 2019 12:49 pm
Ovyron wrote:
Mon Nov 04, 2019 11:35 pm
I never said they needed me.
Sorry for that :mrgreen:

I noticed that Old Houdini, Fire, and Ethereal are doing much much better than KomodoMCTS https://tcec-chess.com/
I think that KomodoMCTS is doing poorly not because it is FRC but because they are using 256 threads. Komodo MCTS is quite good at using 8 or 16 or 32 threads, ok with 64, but beyond that at some point they probably do more harm than good, whereas for normal engines extra threads (up to core count anyway) always do at least a tiny bit of good.

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Wed Nov 06, 2019 4:13 am
by Chessqueen
lkaufman wrote:
Tue Nov 05, 2019 4:46 pm
Chessqueen wrote:
Tue Nov 05, 2019 12:49 pm
Ovyron wrote:
Mon Nov 04, 2019 11:35 pm
I never said they needed me.
Sorry for that :mrgreen:

I noticed that Old Houdini, Fire, and Ethereal are doing much much better than KomodoMCTS https://tcec-chess.com/
I think that KomodoMCTS is doing poorly not because it is FRC but because they are using 256 threads. Komodo MCTS is quite good at using 8 or 16 or 32 threads, ok with 64, but beyond that at some point they probably do more harm than good, whereas for normal engines extra threads (up to core count anyway) always do at least a tiny bit of good.
Does that Apply to the other version of Komodo too or NOT, since Komodo is doing GREAT. What makes Komodo perform much better than Komodo MCTS in all kind of Chess variants ?

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Wed Nov 06, 2019 4:54 am
by lkaufman
Chessqueen wrote:
Wed Nov 06, 2019 4:13 am
lkaufman wrote:
Tue Nov 05, 2019 4:46 pm
Chessqueen wrote:
Tue Nov 05, 2019 12:49 pm
Ovyron wrote:
Mon Nov 04, 2019 11:35 pm
I never said they needed me.
Sorry for that :mrgreen:

I noticed that Old Houdini, Fire, and Ethereal are doing much much better than KomodoMCTS https://tcec-chess.com/
I think that KomodoMCTS is doing poorly not because it is FRC but because they are using 256 threads. Komodo MCTS is quite good at using 8 or 16 or 32 threads, ok with 64, but beyond that at some point they probably do more harm than good, whereas for normal engines extra threads (up to core count anyway) always do at least a tiny bit of good.
Does that Apply to the other version of Komodo too or NOT, since Komodo is doing GREAT. What makes Komodo perform much better than Komodo MCTS in all kind of Chess variants ?
I don't think that the gap between regular Komodo and Komodo MCTS is any larger in FRC (what other variant does it play?) than in normal chess. The gap is large in this tournament because they are using 256 threads, which is more than Komodo MCTS is designed to use or has ever been tested on. The way they use many threads is completely different; the MCTS method just doesn't work properly on this many threads.

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Thu Nov 07, 2019 1:46 pm
by Chessqueen
lkaufman wrote:
Wed Nov 06, 2019 4:54 am
Chessqueen wrote:
Wed Nov 06, 2019 4:13 am
lkaufman wrote:
Tue Nov 05, 2019 4:46 pm
Chessqueen wrote:
Tue Nov 05, 2019 12:49 pm
Ovyron wrote:
Mon Nov 04, 2019 11:35 pm
I never said they needed me.
Sorry for that :mrgreen:

I noticed that Old Houdini, Fire, and Ethereal are doing much much better than KomodoMCTS
I think that KomodoMCTS is doing poorly not because it is FRC but because they are using 256 threads. Komodo MCTS is quite good at using 8 or 16 or 32 threads, ok with 64, but beyond that at some point they probably do more harm than good, whereas for normal engines extra threads (up to core count anyway) always do at least a tiny bit of good.
Does that Apply to the other version of Komodo too or NOT, since Komodo is doing GREAT. What makes Komodo perform much better than Komodo MCTS in all kind of Chess variants ?
I don't think that the gap between regular Komodo and Komodo MCTS is any larger in FRC (what other variant does it play?) than in normal chess. The gap is large in this tournament because they are using 256 threads, which is more than Komodo MCTS is designed to use or has ever been tested on. The way they use many threads is completely different; the MCTS method just doesn't work properly on this many threads.
Alliestein is doing very good without having any NET training of the 960 FRC positions as far as I know and against the all mighty Stockfish the King of FRC Chess960 :roll: https://tcec-chess.com/

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Thu Nov 07, 2019 2:19 pm
by dannyb
sorry for the off-topic question but any news about season 17?

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Thu Nov 07, 2019 6:59 pm
by Modern Times
Chessqueen wrote:
Thu Nov 07, 2019 1:46 pm

Alliestein is doing very good without having any NET training of the 960 FRC positions as far as I know and against the all mighty Stockfish the King of FRC Chess960 :roll: https://tcec-chess.com/
I think it might be up to 100 Elo weaker at FRC without having the special training and net. But at least it plays, which is more than can be said of Lc0.

Re: Finally TCEC is matching FRChess960 Engines......

Posted: Thu Nov 07, 2019 9:12 pm
by Ovyron
Solution: When Komodo is on MCTS mode disallow it from using more than 64 threads.