Author: James T. Walker
Date: 12:17:31 03/13/00
Go up one level in this thread
On March 13, 2000 at 12:21:24, Imran Hendley wrote: >On March 13, 2000 at 10:02:59, James T. Walker wrote: > >>Hello, >>I have revisited an old test I ran over a year ago trying to determine the value >>of tablebases. Last time I ran Crafty with tablebases vs Crafty without >>tablebases and the result showed only about 5-10 points value for tablebases. >>That test was terribly flawed in that I only had a few 5 man tablebases and did >>not have the important KRPKR. This time I had about 4 Gig of >>tablebases(Compressed) available for Hiarcs 7.32. I ran 200 games with >>tablebases and 200 games without tablebases for Hiarcs 7.32. All games vs Chess >>Tiger 12.0e. Hiarcs 7.32 ran on a PII-333 with 16 M for hash tables. Chess >>Tiger ran on a K6-3-450 with 8 M for hash tables. All games were game/5 >>minutes. The results were that without tablebases Chess Tiger showed a 75 point >>advantage over Hiarcs 7.32 using the hardware mentioned. With no change in >>Hardware but Hiarcs 7.32 using the 4 Gig of tablebases available, Chess Tiger's >>advantage dropped to 35 points(200 games). This leaves a net improvement of >>about 40 points for the tablebases. I also want to mention that the hardware >>advantage appears to be about 35-40 points and that would make Hiarcs 7.32 and >>Chess Tiger 12.0e about even at G/5 min with Hiarcs 7.32 using tablebases. Of >>course only 400 games is not enough to be very accurate but it is enough to get >>a rough idea under the circumstances mentioned. I would very much like to know >>of other test done by anyone in this area. If you know of any please point me >>to the info/site. >>Jim Walker > >Excellent test. But wouldn't tablebases increase a program's rating much more at >short time controls than at long? At longer time controls a program would have >the time to calculate endgame wins without tablebases - even if not the most >efficient lines, while at short time controls it may not. Hello, You may be right! I don't know. When I did the test the first time because of the small difference someone suggested that because blitz games include a lot of luck that tablebases might not help much! I honestly don't think it matters. With longer time controls the programs will reach the tablebases earlier in the game with say 10-12 pieces on the board so take your pick. In any case I'm convinced they are worth more than the 10-20 points some people have suggested. Of course with programmers relying more on tablebases, maybe some endgame knowledge will come out of the programs making the more dependent than ever. The Knight & Bishop mate is a nice example. Also the Queen vs Rook. Some commercial programs cannot do some of these now without tablebases. 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.