Author: Dieter Buerssner
Date: 12:57:00 09/22/00
Go up one level in this thread
On September 22, 2000 at 07:42:11, Mogens Larsen wrote: >On September 22, 2000 at 01:46:42, Jason Williamson wrote: > >>Are you using any differnt values in the .ini for stuff such as null_low 4 or >>nval 1 (I use those on the ICC atm.) > >I made a few 30 games tests with and without null_low 4 and found it to be >inferior to the default setting in all of them. Just out of curiosity, how does >your .ini file look like? Mogens, the result you sent to me really is almost the only result, where Yace is doing coniderably worse with null_low 4. One other result from myself was against The Crazy Bishop at fast time controls, where it was worse. I got sent quite many test results against very strong engines with Game in 30 minutes. null_l 4 seemed to significantly better. In the boring tests Yace - Yace at different time controls, null_low 4 allways was better. Against Crafty at different time controls null_low 4 allways seemed to be better. I am still puzzled. BTW. With null_low 4, a similar null move strategy is used, to the one described by Ernst A. Heinz in the paper he cited recently under the null move thread. (I wish I had known about this excellent paper earlier). The difference is, that the threshold for selecting the depth reduction for the null move in Ernst's paper is fixed (he suggests 6), and I use a threshold, that more or less depends on the initial search depth (with some unfortunate consequences for hash table consistency). Also in the endgame, a different strategy is used. For all the people, that use Yace 0.22. To enable a very similar null move strategy, to the one Ernst suggested, you have to set null_high 4 null_split 6 This changed in 0.23 (Please don't aks why ;) and may change again soon. The null_split is currently ignored. About the nval Jason mentioned. With nval 1 Yace uses piece values, similar to those suggested to me by Uri in the Winboard forum. This really seems to improive the play against Crafty, which is very clever to avoid bad trades or to give a trap to the opponent to accept a bad trade. Against other engines, using nval 1 seems to hurt. Regards, Dieter
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.