Author: Robert Hyatt
Date: 18:03:16 11/27/98
Go up one level in this thread
On November 27, 1998 at 19:52:54, James Long wrote: >As I rewrite my engine, I'm being forced to think about >some things I haven't seen in a while. One of those >things is the aspiration windows. I know the theory, >but am a little confused with the implementation. > >Say the 6th iteration of search returns a score of >1000 (1 pawn). The 7th iteration starts with a >window of 700,1300. This search fails low, and is >restarted with a window of -CHECKMATE-1,701. > >Now the time runs out in the middle of the research, >but not before the pv is updated at the root. > >Is any pv found during that research invalid (meaning >I should use the pv from the 6th iteration)? > I'm not sure what you mean here. If you back a score all the way back to the root, then yes you can use that score. If you mean that after you run out of time, *every* move has failed low, then *no* you can't play a move without grave risk. Because you would have zero idea about which move is the best... just use more time but you have to do some- thing to figure this out before playing something bad. IE suppose you think you are winning a pawn after depth=6, but at depth 7 you realize this isn't true, that your opponent can win it back... but you don't know this, only that every move failed low on the original window. You might randomly choose amove that just gives up the pawn, or you might do something much worse. >Another scenario: > >The 7th iteration starts with the same window (700,1300). >Time runs out before all root moves are searched, but >not before the pv is updated with a move that scored >800. So the search didn't fail low, but time ran out >before it was complete. Is this pv any good? perfectly good... > >Thanks in advance. > >--- >James
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.