Two tuning concepts

Discussion of chess software programming and technical issues.

Moderator: Ras

User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Two tuning concepts

Post by cdani »

Hello.
In my tunings now I do more games than before. It happened to me more than once that when I tune a value it ends at for example -12 in selfplay but at -10 in gauntlet against other engines, so of course I leave it at -10. Also it happened the same tuning the LMR stuff, where I can be less aggressive against other engines than in selfplay.

Another not related thing I want to comment is that I have in Andscacs a somewhat more aggressive mobility values than what I see on other engines, and when I tune space, pawn structure or pawn center control stuff I have limited results, so it's like it prefers to have better mobility than good structure. It's just an obvious consequence of the tuning, but I wanted to comment it to expose than the pawn structure may be is not that sacred as one might think viewing the play of most engines, statistically evaluating it. Andscacs of course it's not that good, but anyway I wanted to expose this.
Ferdy
Posts: 4853
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Two tuning concepts

Post by Ferdy »

cdani wrote:Hello.
In my tunings now I do more games than before. It happened to me more than once that when I tune a value it ends at for example -12 in selfplay but at -10 in gauntlet against other engines, so of course I leave it at -10. Also it happened the same tuning the LMR stuff, where I can be less aggressive against other engines than in selfplay.

Another not related thing I want to comment is that I have in Andscacs a somewhat more aggressive mobility values than what I see on other engines, and when I tune space, pawn structure or pawn center control stuff I have limited results, so it's like it prefers to have better mobility than good structure. It's just an obvious consequence of the tuning, but I wanted to comment it to expose than the pawn structure may be is not that sacred as one might think viewing the play of most engines, statistically evaluating it. Andscacs of course it's not that good, but anyway I wanted to expose this.
Piece activity via mobility is of course the priority over the pawn structure. But in the beginning it is the pawn structure that helps the pieces get the mobility that they want. Now it is up to the mobility if it can deliver winning positions, if not then we go back to pawn structure for end-game battle.