Author: Mike S.
Date: 20:53:52 06/08/03
Go up one level in this thread
On June 08, 2003 at 20:41:42, Dave Kuntzsch wrote: >Just got a new computer that is large enough to hold the five piece table bases. >Do they appreciably improve ELO if the program is already relatively strong? From all what I read about that subject, NO. Especially not if you mean an "uprade" from all 4-piece tbs. to all 5-piece tbs, because then, that difference will of course be smaller than to the situation without any tbs. >Or are they only good for tricky endings that occur infrequently? I think yes.... but most ending can be tricky (more or less), often. It's also much different, if you think of improvement in terms of Elo performance (only), or improvement of the quality of the moves too. IMO, it makes a difference if an engine plays an ending perfect, or if it misses the shortest win several times (but still wins). Also, it depends on how much knowledge an engine has for basic endgames (or if it even must rely on tbs. for that), i.e. KNN-K draws, KRN-KR, KRB-KR draws and such, or how to play KBN-K although it doesn't appear often. >Has anyone done a >same engine study with one using the four piece, and the other the five piece? No, but that has been done comparing 5-piece tbs. to no tbs., in a large experiment using the Nunn Endgame Test positions, with several engines playing against themselves. The total result was a bit ambigous... except for one tbs. position which is included in that collection, the engines performed *slightly better without* tbs. than with. It seems to me, engines - in average - are not very good in predicting if the tbs. access will be too expensive in terms of time needed for the file access, or not (although engines try to do this as I've been told). That experiment gave the impression that the slowdown effect was responsible for worse performances, at least partially (in addition to other effects, which I admit remained unclear to me). It's unclear to me if the results of 4-piece tbs. vs. 5-piece tbs. would have been better or worse, in that test. But I guess the slowdown effect is much smaller with the 4-piece tbs., so I use them as a compromise (and KRN-KR, KRB-KR in addition to that, to spot possible exceptions of these draws... or to evaluate them correctly, if the engine doesn't know these important draws). It may be woth consindering to add all KR?-KR inkl. KQR-KR to the 4-piece set, if you have enought HD space, as these are the most important in practise. Without the 4-piece tbs., engines may miss KQ-KR wins (or even KBN-K) sometimes, and can play KP-KP perfectly, so I always recommend to at least use the complete 4-piece set. And of course, Mike is right when he points to analysis in the other posting. Tbs. are a big improvement for endgame analysis and also for people who like endgame studies, to cross check solutions, correctness etc. It may be useful regarding this, to put the 5-piece tables in another directory when your program supports more than one tbs. directory to be configured. By that, you could easily switch between using all tbs. (for analysis), or all 4-piece tbs. only (i.e. for engine matches). In general, no downsides are reported from people who use all tbs. in engine matches. But it would require deep analysis, if and where points or half points have been won or lost due to the tbs. usage... Maximum isn't optimum. Regards, M.Scheidl
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.