Author: Vasik Rajlich
Date: 02:47:40 01/13/05
Go up one level in this thread
On January 12, 2005 at 11:30:16, Uri Blass wrote: >On January 12, 2005 at 10:54:26, Anthony Cozzie wrote: > >>On January 12, 2005 at 02:33:38, Jouni Uski wrote: >> >>>In my (private) endgame testsuite Fruit scored better than some programs >>>with tablebase support (e.g. Junior8 and Crafty). Quite stunning - it seems, >>>that excellent search depth compensates TBs! And my suite has some 5/6 piece >>>positions were TB access is definitely advantage. >>> >>>Jouni >> >>IMO the 5-piece tablebases are just not that interesting and really not worth >>that much in terms of elo. A little endgame knowledge can cover most of the >>positions and be a lot faster too. The tables still help, but its not a killer >>advantage, say 10-20 elo. Every 10 games or so the side with the tables picks >>up a half point. >> >>However, the 6-man tables are a COMPLETELY different story. Think about all the >>interesting 6-man tables: >> >>KXPPKX >>KXKPPP >>K[BN]PPKR >>KRBPKQ >> >>and the list goes on. I suspect a full 6-man set is worth 100 elo at least. > >I do not believe in it. >I do not believe that even full 7-Woman set(piece-set but I simply respond by >woman because as a feminist I object to "man set") is not worth 100 elo. > >>This is why I keep pestering Skinner to let me download some of them :) >> >>And when the 7-man tables are finished? (2010 or so :)) Computers will become >>invincible in the ending, as they win all "simple" endings like KQPPKQP with >>ease. > >There are still a lot of endgames like KRPPP vs KRPPP that are not covered by >the tablebases. > >The practical facts are: >1)A lot of games are decided in the middle game and not in the endgame. >2)A lot of endgames are not so simple that 7 piece tablebases can help in them. >3)A lot of endgames when 7 pieces or less is relevant are simple enough that >programs can find the correct moves without tablebases. > >If we assume 50% of the games are decided in the middle game and in 50% of the >rest of the games the endgame is not simple we are left only with 25% of the >games and if in 50% of them using tablebases does not change the result because >programs are good enough to find the right move without them then we have >only 12.5 of the games when the result is changed and changing the result of >12.5% of the game (in most cases by half point) is not enough for 100 elo. > >You will get something like 57/100 instead of 50/100 that is nice improvement >but only something like 50 elo. > >The effect may be bigger in blitz when programs more often cannot find the >correct move without tablebases but when the 7 piece tablebases will be ready >computers are going to be significantly faster. > >Uri Uri - how about a similar calculation for the famous Junior endgame mating "bug". Vas
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.