Author: Robert Hyatt
Date: 17:21:10 08/02/99
Go up one level in this thread
On August 02, 1999 at 12:39:33, J. Wesley Cleveland wrote: >On August 02, 1999 at 09:19:29, Robert Hyatt wrote: > >>On August 02, 1999 at 06:25:19, José Carlos wrote: >> >>> Maybe discarding "bad enought moves" (considering a fixed minimum difference >>>beteewn first and second move), or the mate-test mentioned could help much in >>>quick games. >>> It could be interesting using such idea in games faster than 5min/game. >>> >>> José C. >> >> >>There are two obvious problems with this: (1) discarding 'really bad' moves >>doesn't help... because most use a time-limit for their search. And discarding >>a few moves at the root doesn't help other than we might barely get one ply >>deeper than normal; (2) Rememeber that in the position I posted, Bxh6 looks >>"really bad" at ply 1, 2, etc.. because Qxb6 wins a piece, Bxh6 loses a piece. >>A difference of 6, roughly... So discarding this move because it looks bad >>would be a big mistake... > >My idea about "forced moves", is to set the search time to 1/3 (more or less) >the normal, window to (very bad, -infinity), search all moves except the forced >one, if any moves returns a score better than "very bad", start over with a >normal search, or else make the forced move without searching it. The idea being >if all other moves lose, make the one that looks good, and if it is bad, you >haven't lost much. This is exactly the sort of approach that killed me in 1981. Because all of the other moves look bad when compared to the 'best' move. Because the best move wins a piece instantly. None of the other moves appear to do so. And they _never_ do so. But if you search this 'best' move long enough you realize it loses, and then the other moves are better...
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.