Author: Dirk Frickenschmidt
Date: 10:32:46 09/14/98
Go up one level in this thread
On September 14, 1998 at 12:31:44, Bruce Moreland wrote: Hi Bruce, > >On September 14, 1998 at 11:41:33, Moritz Berger wrote: > >>On September 14, 1998 at 11:27:43, blass uri wrote: >> >>>I undersood that 18.Qd3 is in the powerbook and this was the reason fritz5.03 >>>played it. >> >>Thanks for pointing this out, Uri! I was too focused on checking the engine >>(analysis mode, book isn't displayed) that I didn't take into account it could >>have been in the book ... That's something that ChessBase could easily fix for >>the next releases: Don't play forced draws from book - if you're behind in eval, >>the engine will chose the draw anyway, so there's nothing to lose. > >So what is the conclusion as of this pont? I see some low-probability moves, >meaning 10-20%, but you'd expect to have these sometimes. > >There is one 0.8% move played, but I think it makes sense to give Thorsten the >benefit of the doubt on that one. > >In any case, that one move doesn't explain why Fritz did poorly in several other >games. I think if you add all factors within few games, the outcome is not that much astonishing as at first sight. 1. Fritz5 *does* play poor games from time to time, and of course it has lots of losses against nice programs in 200MMX autoplayer games as well. No if you consider how often it happens while testing that wins and losses do not happen regulararly around the assumed playing strength proportions, but hat you soemtimes see some losses from one program in a row and then suddenly a series of wins (which happenens all the time if you test enough), you have *one* factor why a program can play poorly in a short tournament. 2. It is well known that other programs profit considerably more from the K6 speed than Fritz5 (running optimal on Intel) does, leading to speed differences of around minus 25-30% for Fritz compared to an Intel pairing with the same programs. The SSDF results for example are based on using Intels, as far as I know. Here you have a second possible factor for comparably weaker Fritz5 play. This is no critique concerning Thorstens testing at all: I would never say it is no good to use a K6 for Fritz5, I just say if you do so you have to know what it means for Fritz. So you have a second factor 3. As you point out, there were some moves with quite a low probability of 10-20% played out, and one with extremely low probability of 0.8%. So you get a third factor resulting in a relatively high amount of relatively weak openings in a small series of games. If you look at all three factors and at the two or so points by Fritz in these few games, you will hardly be able to find anything sensational in the outcome. As I saw even Rebel9 or Nimzo98 were playing below expected scoring, and it seems they did not face two of the three factors mentioned above. So I just keep being curious of new games from new programs I haven't tested myself so far and don't stare too much at the Fritz outcome. The tournament is too interesting not to get much more else from replaying the games, and I see no reason to be suspicious about Thorstens testing as long as there is no very good reason to be. And the on 0.8% move, however it may have occured, is not reason enough for me for long debates, as we both agree... Kind regards from Dirk >bruce
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.