Author: James T. Walker
Date: 18:29:22 06/17/99
Go up one level in this thread
On June 17, 1999 at 18:33:50, Eugene Nalimov wrote: >On June 17, 1999 at 17:58:22, James T. Walker wrote: > >>I finished my first 200 games today between Fritz 5.32 and Hiarcs 7.32. The >>games were played at game/5min. They were run using auto232 operation. Fritz >>had a K6-2-400 for 100 games and then swapped and took the PII 333 for the >>second 100 games and vice versa. Fritz used the F5 opening book and Hiarcs used >>it's own book. Hiarcs had all 3/4 man tablebases plus the 5man tablebases >>supplied on the CD plus about 10 more 5 man tablebases. (I don't have them all) >>Fritz won the match 106.5-93.5 indicating about a 22 point superiority at blitz. >> Of course only 200 games leaves some doubt and I will run another set later but >>now I want to play 200 games with Hiarcs 7.32 vs Junior 5.0. By the way Hiarcs >>turned at least one win into a draw because of the tablebases. It was a KPPKP >>endgame which is supplied with the Hiarcs CD. The problem is there is no KQPKP >>tablebase so when Hiarcs got the pawn to the 2nd rank (With Black) it did not >>know how to win the game and got stuck in 3 time rep with Mate in 7 showing in >>the window?! I had this same problem with another program which prompted a >>question a few weeks ago. Bob Hyatt says that the only way to use the >>tablebases is to have them all. I believe it but apparently some programmers do >>not. At a minimum you need the promotion tablebases if you have the pawn >>tablebases. >>Jim Walker > >Actually, there is a solution for a missing promotion TBs. It was suggested >several times in the Crafty mail list, but Bob is unwilling to implement it - my >guess is that it is going against his sense of aesthetic, because it work not >100% of the time: > >(1) If you have MATE-in-M in the root position, >(2) No one of the moves lead to MATE-in-(M-1), > >Then throw away the moves that lead to TBs draws, loses, or MATE-in-(>=M), turn >off TB probing for N pieces (where N is # of the pieces at the root), and start >normal search. There is some analogue between this approach and current approach >when Crafty tries to exploit opponent's error in a draw position - if it's ahead >on the material in the draw position, it throw away losing moves and searches >remainining once with it's normal evaluation function. > >Of course Bob will say that the full TBs set can reside at a $120 disk. The >problem is that not many people (let alone IMs and GMs) will spend $120 for that >dedicated disk. > >BTW, were there games at WCCC9 where TBs affected the result? If I remember it >correctly, somebody here is doing his/her private match between Hiarcs and >Fritz, and among the first 10 games 4-man TBs already helped Hiarcs. So, 0.5 >points from 10. > >Eugene Hello Eugene, Thanks for your input. I'm not sure how much help the tablebases are in practice. I know if I were a chess programmer, I would want them for one very good reason. When you reach an ending in the tablebases and it gives you "mate in 35" or something like that it is very comforting to know that your program will not blow this particular ending! You can sit back, relax and watch it work to perfection. By the way I just purchased a 8.4 Gig HD for $99 just for Tablebases ($149 - $50 rebate)so it is true. I wish I could get all the TB's but it's difficult to download them because AOL like to cut me off. I only have 256 meg Ram so can't generate them all. Maybe you need to market them on CD! Jim Walker
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.