Author: Gian-Carlo Pascutto
Date: 01:17:04 05/26/02
Go up one level in this thread
On May 26, 2002 at 04:02:32, George Sobala wrote: >I decided to experiment with the settings for Hiarcs8 to see if I could improve >its performance. I have some interesting though not conclusive results. This is >a work in progress. > >All games against Gambit Tiger 2.0 (default settings, using its own book under >ChessPartner GUI). > >All games played using Auto232 on single computer running Win/NT: a slow machine >(K6-2 450MHz), 6Mb hash each. > >Cumulative "historical" data at 5+2 and 10+0 time settings: > >Hiarcs8 using default settings: +27 =30 -54 (37.8%; ELO -86) > >I then changed "selectivity" to 5 from 7, and ran a match at 3+2 (yes I know >that is inconsistent but I was making this up as I went along): > >Hiarcs8 using selectivity=5: +35 =37 -38 (48.6%; ELO -9) > >A substantial improvement! Probably it was due to the change in selectivity, and >not to cumulative book learning or the slight change in time settings - but I >encourage others to experiment further. You should eliminate all other factors to be able to make a conclusion. After 100 games, the error margin will be about +- 50 elo points, which given your result, (IMHO) indicates it might be worthwile to test this more throughly. Be aware though, I'm sure Mark has tested these settings himself and picked what he believe to be strongest. >I have not tried any other selectivity settings, but quick-and-dirty trials with >"hypermodern", "aggressive-defensive" and "combination search" did not produce >any immediate radical improvements. I have no idea what "contempt" is so I have >not altered it (yet). Contempt is disrespect for the opponent. The setting determines how the program will value a draw. -- GCP
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.