Author: Andrei P
Date: 16:16:28 12/02/04
Go up one level in this thread
On December 02, 2004 at 19:02:58, Roger D Davis wrote: >Hi Andrei, > >I suggest you try an empirical approach. Download a bunch of Winboard engines >with ratings established by some reputable rating list. Then, give each engine 5 >seconds, 10 seconds, 30 seconds to solve each problem. You'll have to play with >the time a bit... For example, no engine can solve any problem at 0 seconds, and >probably almost all engines can solve all problems at 10 hours. What you want is >a time control at which about half the engines pass a puzzle and half fail. >There is no reason why the time control has to be the same across all puzzles, >but it might things simpler. > >If you treat puzzles as opponents for engines, then you should be able compute >an ELO for each puzzle. > >Roger Roger, I was thinking about doing something like that. It would give elo strength of a puzzle for the engines. I would also like to know how these puzzles do against humans, but I can expect to find a bunch of player I could run the puzzles through day in and day out. It will be interesting to see if puzzle elo scales up the same way as engine playing strength. it will probably be very close correlation, because the engines are very tactical. I am still intrigued by the way the tactical puzzles scale up for humans. Andrei
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.