Author: Uri Blass
Date: 09:54:15 04/21/02
Go up one level in this thread
On April 21, 2002 at 10:13:03, Kevin Strickland wrote: >On April 21, 2002 at 04:54:58, Uri Blass wrote: > >>On April 21, 2002 at 00:50:20, Kevin Strickland wrote: >> >>>On April 21, 2002 at 00:24:46, Vincent Diepeveen wrote: >>> >>>>On April 21, 2002 at 00:06:06, Eugene Nalimov wrote: >>>> >>>>>First, some time ago *you* wrote that you have to use 2 copies of EGTB cache and >>>>>decompression tables because you uses processes, not threads. I would not call >>>>>such the design "very professional done". >>>> >>>>You are not nice here. I understand why. You try to distract the people. >>>>the EGTB code has 0% influence onto the running time of the speed >>>>test people. In openings position you don't need EGTBs!!!!!! >>> >>> >>>He is simply stating again that the programming of Diep is not professionally >>>done. Not that his egtb code is the best. >>> >>> >>>> >>>>99% of the chessgame the EGTBs are unimportant. A few years ago >>>>when most programs sucked ass in endgames, they were very important. >>>> >>>>Now they aren't!!!! >>> >>> >>>Not true. There are still many endgames where programs are simply clueless about >>>the position, and the egtbs help enormously. And moron can see that. >> >>egtbs do not help enormously. >>There are more important things for a lot of chess programs >> >>Null move pruning,hash tables,pondering and book and internal iterative >>deepening are examples. >> >>The version of movei that is playing in the 5th division of the winboard program >>does not use none of these things and has 5.5 out of 6(all the opponents used >>book but it did not help them,part of the opponents even used hash tables but it >>also did not help them to win). >> >>The latest version use null move but it is not stable so I cannot send it for >>tournaments today. >> >>At blitz it can beat the previous version but my tests show that it has a bug >>and in my last game against the previous version it simply crushed at depth 15 >>after getting an winning endgame in the following position: >> >>[D]8/8/8/1p1K4/1P6/P2bB3/6p1/5k2 b - - 0 58 >> >>Not that even without the bug I am not sure if it could win the game >> >>it planned to play the blunder g1Q instead of Bc2 that is winning but it crushed >>at depth 16 and it is possible that it could change it's mind at tht depth. >> >>Note that it does not know that KB vs KP is a draw and it only knows that KB vs >>K is a draw and it is not going to help it to avoid the blunder. >> >>The fact that the only way that black can win the game after Bc2 is by zunzwang >>is also not going to help it to find the win even if it finds the right move(I >>plan to do a program that is not blind to zunzwang but before doing it I need to >>check that a program that is not blind to zunzwangs has no bugs). >> >><snipped> >>>Most programs use 20 megabytes of ram to use them. How in today's systems that >>>almost always have more than 256 megabytes of ram is 20 megabytes even really >>>significant? >> >> >>Not true. >> >>Most programs do not use tablebases. >> >>I looked at the list of the winboard engines and less than 1/3 of them use >>tablebases. >> >>I also know that a lot of programs do not use pondering but I did not count to >>find if they are a majority of the chess programs. >> >>Uri > >I said "Most programs use 20 megabytes of ram to use them". Not _most_ programs >use them. Of the ones that use them when I see how much ram is used for >decompression and indices it usually rounds to 20M. > >I have seen some games where even Gambit Tiger has a clear win. I could make the >moves to win, yet it shows a draw score until the egtbs kick in. > >I said there are still some positions that programs do not understand, thus >egtbs are still very important. To test this my account on ICC will play for a >month without tbs, and a month with tbs. Using Crafty I guarantee that the >rating difference will be high. If you use constant hardware then I expect the rating to go down because part of the opponents may improve their hardware. ICC is not good to test the importance of egtb. The only way that you can do it is by comp-comp games with fixed hardware in the way that the ssdf test. I also do not think that crafty is a good program to test the importance of egtb. Crafty is using them for many years so it was not important for the programmer to implementing knowedge about them. It is better to take a program like tiger that did not use tablebases 2 years ago when christophe still does not use it in palm tiger in order to find the value of tablebases. Uri
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.