Author: Andrew Dados
Date: 10:17:20 02/18/00
Go up one level in this thread
On February 18, 2000 at 12:20:09, Claudio Bollini wrote: >Hi, chess friends! > >I've just finished an endgame test for three "knowledgeable" progs: >Junior 6, Hiarcs 7.32 and Shredder 4. Initially, my intention >was to find out which one performs better. Additionally, I want to >check the (presumable) boost with TBs. > >I used a huge set of 1258 assorted endgame positions. I took it (I >believe) from a Rebel epd database (mes.epd). Anyway, you can >download it from Dann Corbit's FTP site: ftp://38.168.214.175/pub > >Confirming my assumptions, H732 got the best score, although with >a very narrow margin. *But* incidentally I saw that TB didn't improve >that much their performances. > >(BTW, I had to deactivate manually TBs use in "chssbase.ini". Engine >option from J6 menu seemed to ignore my request and kept using them). > >As shown below, differences are negligible, if any (Shredder 4 case). >I still believe that TBs *should* increase endgame performance, but I >cannot explain what happened here. Any idea? > >Best regards, >Claudio Bollini > > >==================================================================== > >File: "mes.epd": 1258 endgame positions. >PII 266, 64Mb RAM, 1' per move, 8Mb Hash. >TB: from ChessBase (150Mb), files for 3 and 4 men. Most important: >KPK,KPPK,KPKP,KNKP,KNPK,KBKP,KRKP,KRKN,KRKB,KQKP,KQKN,KQKB,KQKR > > *With* TableBase *Without* TableBase > > hits % average hits % average > time time > >H732 1044 (82.99%) 3" 1047 (83.23%) 4" >J6 1040 (82.67%) 4" 1021 (81.16%) 4" >S4 1024 (81.40%) 4" 1022 (81.24%) 3" > >==================================================================== It is no surprise to me. Most of 4 piece TB knowledge can be (is) put into programs evaluation function anyway, so they would not benefit from 4 men TBs. On rare occasions they may probably misevaluate KPKP or KRKP in leaves and then those TBs are advantageous. 5 men TBs are different story... most important being KRKRP for practical play. (One more thing: depending on TB probing code implementation slowdown with TBs is often offset a bit by noticeably deeper search.) -Andrew-
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.