Author: blass uri
Date: 04:59:23 10/06/99
Go up one level in this thread
On October 06, 1999 at 03:14:29, Jouni Uski wrote: >I have run Nunn test *) many times between different engines and my conclusion >is that >it works quite good in predicting differences in playing strength. And >definitely it works >better than ANY test position suite. >E.g. I tried to test if lg2000 is really better than Crafty 16.6. I run test in >AMD450 Mhz >4 times (5'+5'). Here are results: > >1 lg2000 - crafty 9 - 11 >2 8 - 12 >3 9 - 11 >4 8,5 - 11,5 > >So You can reproduce result quite reliable! Interesting I didn't found ANY games >exactly reproduced in these 80 games, but still crafty always wins. > >I also run 80 games with Fritz general book. Then result was lg2000 - crafty >40,5 - 39,5. > >Jouni Uski > > > >*) Nunn test contains 10 opening positions selected by GM Nunn. In test You play >them >between two programs without books with white and black pieces. I read in the past that Fritz5.32 beats Fritz5(16 bit) in results like 15:5 in the nunn match but there is only a small difference in practical games so you cannot learn from the nunn test about the ability of chess engines. I believe that it is possible to teach programs to play better in the nunn test without improving the level in practical 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.