Author: Dana Turnmire
Date: 19:08:49 09/30/02
Go up one level in this thread
On September 30, 2002 at 17:35:38, John Merlino wrote: >On September 30, 2002 at 17:07:38, Dana Turnmire wrote: > >>It has been pointed out that in a long match such as SSDF holds CM9000 would be >>at a substantial disadvantage because of programs using their learning features >>to repeat wins over CM. >> >>Assuming CM has a top notch opening book wouldn't using the randomizer have >>basically the same effect of the learning feature? In the "setup personality" >>option why not set the random setting to something higher than 0? Would setting >>it to 100 prevent any repeat games without hurting the playing strength? > >Using any setting other than 0 can, theoretically, lower the overall playing >strength. Randomness basically adds a random positive or negative amount to >every legal move at the root. This can result in the non-optimal move being >played. The higher the randomness value, the worse the overall playing strength, >all else being equal. > >Also, the randomness value of a personality has nothing to do with the opening >book. It only affects the search. > >jm I see. The manual said it "adjusts the personality's diversity of play, while maintaining its strength of play." Could there be a feature that randomizes the opening book so it will not repeat the same opening for a set number of games? Might this not solve the problem of repeat losses against learning programs? After all there are literally thousands of opening variations with an equal score. Seems that wouldn't be a problem and would be simpler than using a built in learning feature though I'm not a programmer.
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.