Author: Nathan Thom
Date: 21:09:04 01/15/03
Go up one level in this thread
On January 15, 2003 at 22:59:57, Uri Blass wrote: >If you use hash tables but not for pruning you can always get the entire pv. >Movei still does not use hash tables for pruning but only for order of moves >because the last time that I tried using hash tables for pruning test positions >showed that the result is not better. > >One possible reason is the fact that I do not store partial extensions in the >hash tables(I thought that it still may be better to avoid search and use the >result at depth k to prune expensive search at depth k+partial but tests showed >that it cause more problems). > I dont understand how using hash tables for pruning would not speed the search? >Another possible reason is my different pruning methods that mean that I do not >always get the same result when I search the same depth from the same position >even when the partial extensions are the same(it is the same for everybody who >use repetition detection but I have more reasons). I also dont understand why you dont get the same result when searching the same position to the same depth each time. What does repetition detection do to affect this? > >When I decided to prune only when the depth is bigger and not equal to the >reported depth I also got slightly worse results(usually I was slightly faster >but there were still few cases when I got the solution one ply later). > >Maybe the fact that I got slightly worse result was bad luck but I prefered >simply not to use hash tables for pruning because I have no time to check if >this is a small improvement or not a small improvement. > >I am not going to change it for the cct5 because it is not something that I can >change in the last minute and it is more important for me to add better >evaluation of king safety and maybe better eval of pawn structure(not sure if I >will do it). > >Uri
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.