Author: Walter Koroljow
Date: 08:46:13 10/22/99
Go up one level in this thread
Three points: 1. I am impressed. You fit something like 98 games x 35(?) moves = >3000 moves with only 17 parameters (my count for CM6000). This is no freak occurence. 2. Bob Hyatt has commented that CM6000 and Ferret have very different searches. If the two algorithms are so different but can give pretty much the same answer, does this mean that the details of the search algorithm are much less important than the evaluation function? 3. How did you ever do this logistically? Consider: if you tried just two values (independently) for each of the 17 parameters, that would be 2**17 =131,072 trials and you might evaluate them over 3000 moves each for a total of about 400,000,000 evaluations. Furthermore, for each move, you checked (manually, of course) the time CM6000 took. Did you also allow the exact time for CM6000 pondering? Of course, you must have short-circuited this procedure, but I am wondering just how you did it and how long it took you. Have a good weekend, Walter
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.