Author: Uri Blass
Date: 06:08:34 08/30/05
Go up one level in this thread
On August 30, 2005 at 08:29:51, Joachim Rang wrote: >On August 30, 2005 at 08:06:00, Uri Blass wrote: > >>On August 30, 2005 at 05:11:07, Joachim Rang wrote: >> >>> >>>>Or you can store 2 tables -- one with Win/No Win, and other with Lost/No Lost. >>>>Presumable they will compress even better, and in lot of cases during the search >>>>you will probe only one of them, because that is all you need to know. >>>> >>> >>>Just a crazy idea from a non-techie: Couldn't you store only win/draw in a table >>>and simply skip all lost positions and tell the chess program that in case there >>>is no entry for a position it is lost? >>> >>>regards Joachim >> >>Maybe better experts than me will answer but I will share my thoughts about it. >> >>How do you do it? >> >>basically every position is translated to a number. >> >>The way that chess programs find if position is a draw or win or a loss is that >>they translate position to a number and the number tell them where to look in a >>big array to tell them the result when 1 is win and 0 is not win. >> >>How do you expect a program to find that there is no entry for a position? >>If you do it by another number then you need to use more than 1 bit. >> > >Which array is used for EGTBs? A ordinary arrays which is indexed by a single >integer or a Associative array like for Hashtables? It is not stored exactly like hash tables. hash tables may have the same key for different positions. tablebases have different numbers for different positions. > >I think EGTBs are stored like hashtables and a lookup is done via a >(zobrist)-key. I though for hashtables sometimes you don't get a hashtable hit >if the position was not searched yet. I think one could handle missing TBs-Hits >accordingly. > >regards Joachim You can get position that was not searched in hash tables but you basically have a lot of memory in hash tables so it is not a problem. The fact that the position is not in the hash table does not mean that you remember nothing for that position. You clearly remember zobrist key of position(or at least most of the bits of it) and you know that the position is not in the hash tables based on the fact that the zobrist key of the position in the board is different. Doing a similiar thing for tablebases means that you need to spend significantly more memory on them because today the only information that you remember in tablebases is result of the game. 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.