Author: Chris Carson
Date: 13:02:43 10/21/99
Go up one level in this thread
On October 21, 1999 at 15:55:40, Robert Hyatt wrote: >On October 21, 1999 at 14:33:39, Dave Gomboc wrote: > >>On October 21, 1999 at 10:03:22, Chris Carson wrote: >> >>>On October 21, 1999 at 09:03:34, Robert Hyatt wrote: >>>> >>>>I would ignore it. It simply won't work. Ferret and CM are totally different >>>>in their searches. It would be impossible to tune the eval and make the two >>>>programs play the same. Match for a specific set of games? Of course. Match >>>>in real games? not a chance. >>> >>>I agree with Bob. I have CM6K and you can create a personality, but >>>it will not be an exact or close to prediction of future move selection >>>for a given person or program. This is a type of regression technique >>>(known data to predict future events), the error of measure for this would >>>be very large (ie: not very accurate). Just my opinion. :) >>> >>>Best Regards, >>>Chris Carson >> >>The original poster might find another 70 games that weren't in the initial set, >>and compare to see the matching rate with those. >> >>Dave > > >This "approach" is just like tuning a program's eval so it will match the >moves made by a GM. Without understanding why the GM made the moves in the >first place. For a finite set of moves, you can get very accurate matches >using a least-squares approach to fitting. But when you play a _new_ game, >you find just how badly this does. > >Ferret and CM are so different in their search approach and evaluations, that >trying to adjust one to act like the other would be hopeless... An interesting >exercise? probably. But accurate? nope. Well said. :) Best Regards, Chris Carson
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.