Stockfish's contempt value

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

FlavusSnow
Posts: 89
Joined: Thu Apr 01, 2010 5:28 am
Location: Omaha, NE

Stockfish's contempt value

Post by FlavusSnow »

Can someone explain how Stockfish determines if it should take a 3-fold rep draw or not?

I run Stockfish on FICS as continuously as possible. Something that really gets my goat is the ability of low rated humans to obtain an early 3-fold rep draw all too often. I don't know how frequently it occurs really, but I'd say in a given week I'd have about 1,000 games played and there is usually something like 2 or 3 draws like this. So 1 in 300 to 1 in 500 chance.

I use draws like this to improve the opening book, but I feel like its more a weakness of the engine than a flaw in the opening book. Regardless, it lowers Stockfish's rating because 95% of its games are played versus low rated players to the level where a victory results in no points gained, but a draw results in a loss of 8 points.

Can an engine read its opponents rating during an ICS game? If so, could this be used to improve a 'contempt' calculation?
rai.rajesh

Re: Stockfish's contempt value

Post by rai.rajesh »

The way I decide this in my engine Abaddon is I check if Abaddon thinks it is down, if it is, then Abaddon will simply take the 3-fold, else if it thinks its up, then it will first look into the 'next best' move after the 3-fold accepting move, If this is within the 'margin' Abaddon will not take the 3-fold. If this next best move scores much worse then draw move, draw is accepted.

But maybe Stockfish does it differently, I am sorry I can not comment on that, as I have not gone through Stockfish code, I have a lot more of basic stuff to get hold of first!
zamar
Posts: 613
Joined: Sun Jan 18, 2009 7:03 am

Re: Stockfish's contempt value

Post by zamar »

FlavusSnow wrote:Can someone explain how Stockfish determines if it should take a 3-fold rep draw or not?
Sure. Draw is evaluated as 0.00. If Stockfish finds a move that leads to a better result, it will avoid the draw. If not it'll take the draw.

Against very weak players SF already scores >99%. Our current goal is not to improve this. Anti-human chess is very different field compared to objectively strong chess.
Joona Kiiski
User avatar
Olivier Deville
Posts: 937
Joined: Wed Mar 08, 2006 9:13 pm
Location: Aurec, France

Re: Stockfish's contempt value

Post by Olivier Deville »

Hi Rajesh, where have you been ?

No recent upgrades for OpenWar...

Olivier