Author: Bruce Moreland
Date: 09:05:28 06/23/00
Go up one level in this thread
On June 23, 2000 at 05:36:36, Adrien Regimbald wrote: >Hi, > >I have recently noticed a problem with Faile while performing play testing >before releasing the current version - > >It is properly detecting 3-fold repetition, but sometimes it misses it in the >following case: >A position has been seen before, and is searched to depth x. >The position comes up again. If Faile plays the same move, it allows the >opponent to play a move which makes it a draw by 3-fold repetition. >Faile's search to depth x goes immediately due to hash table matches. It begins >to search depth x+1, and suddenly realizes that its intended move will result in >a draw, because it has finally had to actually search now. Faile runs out of >allocated time after searching the PV move. Faile will then play the move >(since this is the move suggested at previous depths and there is no time to >search other root moves), and the opponent, of course, plays the repeat and >draws the game. > >This is very frustrating, and seems to be the only significant problem with >Faile's play so far in testing (other than some problems with book coverage, >which should be fixed when I take some time to create a nice book - the current >book is only being used so it has _SOME_ book ;P). > >I have been trying to figure out how to solve this problem, and I've realized >that I have to somehow "age" the hash table entries .. but I'm not sure as to >when I should do this in order to get it working both to full effect in >benefiting the search, but also not causing these problems with 3-fold >repetition. Looking at other sources has so far been fairly futile - some of >them don't do anything to combat this issue, some use much more complex hashing >schemes than I do, and I can't seem to find a clear answer to this problem. > >Does anyone have any suggestions for me? I am using a single table, with >replacements done if the current depth is greater than the depth in the hash >table. Don't allow hash cutoffs so close to the root of the tree. bruce
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.