Author: Robert Hyatt
Date: 14:16:16 04/08/01
Go up one level in this thread
On April 08, 2001 at 02:41:20, Aaron Tay wrote: >On April 07, 2001 at 23:59:48, Robert Hyatt wrote: > >>On April 07, 2001 at 12:14:43, Uri Blass wrote: >> >>>On April 07, 2001 at 11:46:25, Rajen Gupta wrote: >>> >>>>On April 07, 2001 at 10:46:43, Uri Blass wrote: >>>> >>>>>On April 07, 2001 at 09:04:28, Rajen Gupta wrote: >>>>> >>>>>>hi; thanks for the answers: however isn't it true that the 4cd set by chess base >>>>>>(endgame turbo)consists of nalimov EGTB's whereby the chess engine can actually >>>>>>search for a position encoded within these tablebases as opposed to the ordinary >>>>>>ones in which only when a position that is in the EGTB,is actually reached >>>>>>during game play that the EGTBs kick in? >>>>> >>>>>You are wrong about your assumption about the ordinary ones. >>>> >>>>Hi uri: >>>>by ordinary ones i meant ordinary EGTB (the ken thompson ones)as opposed to the >>>>newer nalimov EGTB.i wasn't referring to a particular chess engine having the >>>>capacity to access them in the search.it is my understanding thqt with nalimov >>>>EGTB the chess engine can actually search for a position encoded within the EGTB >>>>as opposed to the older EGTB which kick in only when a psoition has actually >>>>been reached which is within the EGTB. >>>> >>>>correct me if i'm wrong >>> >>>I do not see a reason to assume that the thompson tablebases cannot be used in >>>the search in the same way that nalimov tablebases are used. >>> >> >>There is a big reason. Ken's compression algorithm is not nearly as good >>as what Eugene uses, from a random-probe point of view. IE with Eugene's >>code, a random probe doesn't have to read huge chunks of the file first, >>which makes probing inside the search doable. If you use the Thompson >>compression approach, probing would be horribly slow if done deep in the >>search. >> >> >> >> >>>The thompson tablebases did not give accurate result for part of the positions >>>and it gave only correct results for positions when the strongest side could win >>>but I see no reason that prevent chess programs to use the information in the >>>search. >> >>Ken also ignored enpassant, which means 2 pawns were impossible. >> >> >> >>> >>>I believe that the reason that Fritz could not use the thompson tablebases in >>>the search was the fact that the programmers were too lazy to teach fritz to do >>>it(they were even too lazy to tell Fritz to use the thompson tablebases >>>correctly at the root). >>> >>>The thompson tablebases are bigger than the nalimov tablebases and nobody use >>>the thompson tablebases today so this subject is not relevant. >>> >>>Uri >> >> >>Nalimov == smaller & more efficient to access... > >How about the Edwards tablebases? Crafty used it for a time I believe.. Edward's format was much bigger. It used the 10-square domain trick for one king, while Eugene's approach squeezes far more out of the tables. Edward's tables were not compressed and there was no compression/decompression-probe code for them either. That was why I changed, not to mention Edward's didn't handle en passant either and that was a problem for kpkp and kppkp...
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.