Author: Uri Blass
Date: 22:11:57 10/25/05
Go up one level in this thread
On October 25, 2005 at 13:39:18, billiau wrote: >Hi, > >I just saw the results from CEGT in Blitz games (Thanks to the CEGT !) : >http://www.husvankempen.de/nunn/eloblitz.html > >Actually, based on the first results, it seems that Fruit 2.2 [HI50+KS104] is >great in test suites but not better than the standard setting in games. > >It's why, i am actually testing Fruit 2.2 [KS104]. > >G. Billiau I wonder what was wrong in your previous testing when you found Fruit 2.2 [HI50+KS104] to be better in blitz relative to the default version and I understood that it was after 2 weeks of testing that probably also included hundreds of games. possible difference between your tests and CEGT tests 1)different books when fruit is using it's default book 2)different time control when you used fischer time control and CEGT use 40 moves/4 minutes 3)pondering(CEGT use no pondering when you used pondering at least in games that you played on the ihternet) 4)learning(opponents already learned good lines against Fruit default and had problems when they needed to play against new personality). 5)luck(fruit may be lucky on your games and unlucky in CEGT games 6)different opponents Note that history=50 with no other changes was not tested at blitz but the evidence that we see also suggest that history=50 is not better for blitz. It may be interesting to see history=50 with no other changes both in blitz and in long time control for comparison but I guess that it is not going to happen and I doubt if we are going to see more games for fruit personalities after seeing results that show no significant improvement in the best case(the possible error is still too high to say no improvement but in the best case the improvement is very small). Note that for me results of personality when you change one parameter that is a search parameter are interesting to know even if I expect reduction in playing strength and it may be interesting to know also results of fruit when you disable futility pruning or disable history pruning or disable null move pruning but it is obvious that the CEGT cannot test every personality that somebody asks them to test. 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.