Author: Uri Blass
Date: 06:23:45 04/25/03
Go up one level in this thread
On April 25, 2003 at 08:39:18, Harald Faber wrote: >On April 25, 2003 at 05:56:21, Uri Blass wrote: > >>On April 25, 2003 at 05:33:24, Harald Faber wrote: >> >>>On April 25, 2003 at 04:19:36, Uri Blass wrote: >>> >>>>On April 25, 2003 at 04:04:26, Harald Faber wrote: >>>> >>>>>On April 24, 2003 at 04:01:14, Uri Blass wrote: >>>>> >>>>>>On April 23, 2003 at 16:53:03, Helmut Conrady wrote: >>>>>> >>>>>>>On April 23, 2003 at 07:25:15, Jorge Pichard wrote: >>>>>>> >>>>>>>>Who suggested that Fritz 8.0023 be tested instead of the stronger Fritz 8.008 >>>>>>>>version? >>>>>>> >>>>>>>What results suggest that Fritz 8.008 is stronger? I haven ´t seen something >>>>>>>like that so far. >>>>>>> >>>>>>>Helmut >>>>>> >>>>>>http://www.talkchess.com/forums/1/message.html?294063 >>>>>> >>>>>>8.008 scored better than 8.0023 against shredder with the same >>>>>>conditions(drawbook). >>>>> >>>>> >>>>>I don't like tests where all programs use the same opening book. >>>> >>>>What is bad in these tests? >>> >>> >>>The programs don't play lines from their own opening books which include >>>adjusted lines, opening lines the program "understands", plays well. >>>IMO the method with a unique opening book is useful to find out which program >>>plays which openings best. It might not lead to new insights. >>>See: >>>Prog A plays King's Indian vs. prog B. 10 different lines. Prog A wins by >>>8.0-2.0. So prog A is stronger, right? Especially when prog A plays some more >>>King's Indian lines vesus other progs who also fail to defend. Nonsense I say, >>>when the other progs never play King's Indian from their own book. So what you >>>learn is that progs B, C etc. don't play the King's Indian well with black. What >>>is nothing new, because the King's Indian is already disabled in their own >>>books. >>>Cool test. >>> >>> >>>>People who use programs for analysis may prefer these tests and not tests with >>>>the original book. >>> >>> >>>Those guys should prefer test positions. >> >>The problem with test positions is that they are not taken randomely from games >>and it is hard to know what program choose better moves when humans (even with >>the help of computers) do not know which move is better(otherwise correspondence >>chess could be drawn in every game when both sides use computers). >> >>Uri > > >Test positions can be classified, so no problem at all IMO. Most positions that correspondence players give computers to analyze are not test positions but positions from their games or positions that can happen in their games. 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.