Author: Roberto Waldteufel
Date: 10:10:05 10/27/98
Go up one level in this thread
On October 27, 1998 at 00:10:21, James B. Shearer wrote: > If you do not have square of the pawn code you will make occasional >horrible blunders. Consider the below position in which I had white against a >strong computer (bountyhunter) on ICC. Black plays qc4 converting an easy win >to a embarassing loss as white's lone pawn queens after the queens come off. >This was a blitz game but I tested the position later with rebel 8 and fritz 4 >and both like qc4 for a long time. Crafty on the other hand with its square >of the pawn code is never tempted by qc4 as shown by the log file below. > By the way this sort of position illustrates a pet idea of mine. I >think equal time (ignoring alpha beta for the moment) should be spent on >alternative moves (instead of exploring them to the same depth). This can be >done by increasing the depth by the log of the number of alternatives instead of >by a constant amount as you go down the search tree. So for example for forced >moves this is the same as extending by one ply (since log(1) = 0). However this >is more general since in positions like the below, lines after queens are traded >will have fewer choices and thus be explored to greater depth allowing the >search to figure out that qc4 is bad much quicker. >.............................................................................. (snip) >............................................................................. > James B. Shearer Hi James, A long while back I used a similar idea to yours. At the root I did a quick static test to see if the enemy king was exposed. If the program concluded that it was, then a special mate-finder search was called to try and find a forcing combination. This was structured similar to the main search, but it only looked at checks, captures and moves that threaten to mate or win material, and it reduced the depth in proportion to the log of the number of opponent's legal moves at each ply where the opponent was to move, but did not reduce the depth at all when the computer's side was to move. This way I followed in depth those lines that kept the opponent short of replies. I worked reasonably well, and found combinations faster than the normal search would (not just mates, but winning of material due to mating threats as well). I dropped it in the end because it was wasting too much time when there was no combination, and my static analysis at the root was not good enough at detecting when it was worth calling the mate-finder. Best wishes, Robert
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.