Engine results: a surprise!

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Posts: 22896
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller

Re: Engine results: a surprise!

Post by hgm » Fri Oct 18, 2013 12:18 pm

PK wrote:once You count pseudo-legal moves in Your evaluation function, it makes sense to weigh them differently. Sungorus uses 4*knight mobility, 3 * bishop, 2* rook and 1* queen. It is sufficiently good for start.
To me that has always suggested that this eval term is more about board control than about number of moves. You would not expect Knight moves to be on the average more dangerous than Queen moves; on the contrary. But a square attacked by a Knight is interdicted to a lot more enemy pieces than a square attacked only by a Queen.

Of course it could also have something to do with the urgency to have moves for withdrawing the piece, should it get into trouble. But also then, trouble is a relative notion, and what scares a Queen, could be laughed at by a protected Knight. But if this is a factor, then it suggests the first eight moves of the Queen should also be worth 4 (or even more), like the first (and only) eight moves of the Knight, and it is only the moves in excess of 16 (that even a Rook doesn't have) that count for 1. But of course the Queen would almost always have those first 8 moves, so the mobility bonus associated with them can be incorporated in the base value without much loss of accuracy.

It would be interesting to know if it would be better to have an eval term that gives points for every attacked square, 1 point if the lowest attacker is a Queen, 2 points if it is a Rook, 4 points if it is a Knight... Nearly the same as the weighted mobility, but not quite, as attacking squares that are already attacked by Knight with a Queen would be recognized as rather pointless.

So two 'saturation effects' are to be expected:
* Extra moves on a piece that already has many aren't that important
* Extra attacks on a square that already has many (or better, i.e. by less-valuable piece) aren't that important

The latter might have to be considered in relation to the number (or quality) of attacks on it by the opponent. I guess this goes a lot towards 'safe mobility'; an obvious measure for the effect achieved by all those attacks would be "what can you or the opponent move there safely, now?".

Posts: 3743
Joined: Thu May 15, 2008 7:57 pm
Location: Berlin, Germany
Full name: Sven Schüle

Re: Engine results: a surprise!

Post by Sven » Fri Oct 18, 2013 3:08 pm

Richard Allbert wrote:Hi Sven

Spike is only too strong if the TC's are equal. I Test my engine vs Spike (and others > 2700) using 1s / 40 moves for them, and a LOT more for mine.

This has the benefit of saving a bit of time. :)
Not if you want to play many games and the Elo difference is some hundreds of points. E.g. if the engine to be tested is around Elo 2000, thus 700 points weaker, and doubling time gives roughly 70 Elo points then you probably need 1000s / 40 moves for an almost equal strength.


Post Reply