Author: Peter Kappler
Date: 20:46:17 08/03/99
Go up one level in this thread
On August 03, 1999 at 23:08:55, Dann Corbit wrote: >On August 03, 1999 at 23:00:44, Ricardo Gibert wrote: >>On August 03, 1999 at 22:35:16, Dann Corbit wrote: >>>Has anyone tried something like this for pawn value: >>>ranks advanced = 0, value = 1.000 = 100 centipawn >>>ranks advanced = 1, value = 1.000 >>>ranks advanced = 2, value = 1.013 >>>ranks advanced = 3, value = 1.065 >>>ranks advanced = 4, value = 1.299 >>>ranks advanced = 5, value = 2.547 >>>ranks advanced = 6, value = 10.347 >>> >>>This is derived from the following formula: >>>pawn_value = 1.0 + (ranks_advanced! - 1.0) * .013; >>> >>>I believe that the value of a pawn is a factorial of the number of squares it >>>has advanced... >>> >>>The reason that I think such a scheme is reasonable is as follows: >>>A pawn gains very little value on the first two moves, except some control of >>>the forward squares. However, a pawn two squares from queening is a problem, >>>and a pawn one square from queening is a *big* problem. You would gladly tie up >>>a knight to prevent queening, I think. Hence, it's value is nearly the value of >>>the knight. And, at the moment of queening, it is worth slightly *more* than a >>>queen. The reason it is worth more than a queen is that it can become a queen, >>>or another piece -- if that is advantageous. A queen cannot do that. So, the >>>moment it lands on the promotion square it has a value of something just over >>>10. >>> >>>Thoughts? >> >>I assume you are talking about passed pawns. I would think that this would make >>your evals wildly inaccurate, counterfeiting the values of alpha & beta. Some >>pawns are on the 7th spell death while others are just material to be picked up >>by your opponent. Better to do an extension. The numbers you give, better >>represent the "attention" the pawn should be paid. >It could not upset the eval by more that 150 centipawns, since the final advance >gives you a queen (if you want). I think that the pawns may just be 'material >to be picked up' because we are not giving them the proper attention. If we >considered them to have the value listed above, they would be more carefully >guarded as they race for the back. I think that also, it would prove an >attractor to pull pawns towards promotion. Programs try to promote much too >late. On the other hand, the values stated do not accurately reflect the >destructive power of the pawn, since a pawn one square from promotion is still >just a pawn. > >What I was wondering is if anyone had actually *tried* it to see what would >happen. It seems like it might be an easy enough experiment for many programs. My program, Grok, uses this type of non-linear scoring for pawns, but only if they are passed. Also, the value of such passed pawns increases inversely with the number of enemy pieces on the board. My values are not as large as yours, however. Using such large static scores is simply too risky. It's much safer to just extend the search in these situations so you can really determine if the pawn can promote safely. --Peter
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.