Author: Vasik Rajlich
Date: 04:07:06 12/17/05
Go up one level in this thread
On December 17, 2005 at 04:14:32, Joseph Ciarrochi wrote: >Test is on a pentium 4 with a sample of 22 gambits across the eco codes (engine >forced to play both sides). Time was 25 minutes with 1 sec increment > >Fritz 9 had 14 wins, 18 loses, 12 draws > >This may be a notable drop in performance for fritz 9. For example, the CEGT >blitz (40 4), using more standard test sets, has fritz 9 well in front of gambit >fruit both head to head (15 wins 8 loses 7 draws) and in terms of rating (fritz >9 has 100 points on gambit fruit). There are differences between these >tournments, other than the test set, but I think these findings do suggest that >maybe fritz 9 loses a little strength against gambits? Is it possible that >fritz 9 was tuned on the nunn test sets? > >In defense of fritz 9 (and much more data is needed anyway to make firm >conclusions), it is the only engine that is keeping pace with rybka in a gambit >tournment (same test set, 25 1 time, pentium 3 processor). Here are the interim >results at about the the half way point (except for gambit fruit which is >finished) > >Rybka versus > > win loss draw >gambit fruit 1 4bx 20 11 13 >fruit 2.2.1 10 3 8 >fritz 9 8 10 3 >toga II 1.1 10 6 5 >spike 1. 0 mainz 14 0 7 (still no losses! wow) >ruffian 2.1 13 2 5 > > >In general, it looks like fritz 9 holds its own against rybka in head to head >contests, but is not as good against fruit and some other engines (see >http://www.husvankempen.de/nunn/eloblitz.html#3). Indeed, rybka seems to be >killing fruit . > > >Any thoughts or implications? Rybka is suppose to be heavily strategic, but it >really is fantastic at Blitz and gambits. > >best joseph > >ps. If anybody wants to do the gambit test on hiarcs ten, or on whatever engine, >i'd be happy to post the test set (I am going to expand it by 8 gambits)...The >games are absolutely fascinating. > > >best >Joseph Joseph, yes, this strategic vs tactical stuff is a hard topic that pretty much nobody really understands. Even a few months ago, I would have said the following: 1) The bigger the hardware (or the longer the time control), the more imbalanced the search tree should be. 2) The bigger the hardware, the slower the evaluation function can afford to be. Now, I wonder if maybe the opposite isn't true on both counts. Re. gambits, my intuition tells me that the weakest gambit programs would be those that misevaluate the gambit positions. Computer chess programs give some of their worst evaluations when one side is up material. 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.