Author: Uri Blass
Date: 14:43:30 12/05/02
Go up one level in this thread
On December 05, 2002 at 16:26:44, Gian-Carlo Pascutto wrote: >On December 05, 2002 at 12:59:52, Uri Blass wrote: > >>It is something that the programmer has control over. >>A lot of customers use part of the tablebases so programs should know to play >>only with part of them. > >It's not about having part of the tablebases. It's about having >specific combinations that will not work. I'd be surprised if >any program actually ships with a set that causes problems like >this. > >>If they do not know then it is the programmer's fault. >>I agree with jouni uski here. > >Well, Shredder violently complains about it, as do most >programs. > >>I did not implement tablebases in movei but when I decide to do it I am going >to do it in a correct way. > >You will have the exact same problem then. This has nothing to do with >supporting them correctly - it's an explicit messup on part of the user >you have to specifically take into account. > >-- >GCP supporting them correctly means stopping to use them if the distance to mate is not going down. It also means not giving some mate in 100 a mate score because the program has no way to know if it is a win or a draw by the 50 move rule. I think that programs should transalte mate in more than 50 to some constant and choose move that lead to mate in 100 only if they do not find a better move. It is possible to do an array of scores based on distance to mate and only short mates are going to get the mate score in the search so if the program need to choose between winning a queen and mate in 100 it is going to choose winning the queen because mate in 100 in tablebases may be translated to only 2 pawns advantage. 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.