Author: Dann Corbit
Date: 12:12:25 06/13/00
Go up one level in this thread
On June 13, 2000 at 14:35:51, Mogens Larsen wrote: >On June 13, 2000 at 14:12:21, Tom Kerrigan wrote: > >>Let's say one program has a better evaluation function than another. Or searches >>faster. Should it be handicapped in the name of equal terms? >> >>-Tom > >You're comparing something inherent to a chess program to something which isn't. >Apples and oranges IMHO. Actually, the search process also contains a large amount of data. With deep blue, weren't there something like 4000 tunable eval paramters? Data structures + Algorithms = Programs Algorithms operate on data. Opening books are not really a big table of answers. They are informed choices -- a stockpile of information that *must* be weighed. You can look at the won/loss/draw percentage. You can look at the ELO of players who make a particular choice. You can look at computer tactical analysis of the position. Using these factors (and many others) you can make decisions about what to play. But you can't just play some ideal line, because eventually some weakness will be found and you will lose every game with it if you play the same thing over and over [Surely you have seen this very effect with some of the weaker programs that do not learn and do not introduce probability]. Therefore, your program *has* to learn to be competitive at the highest level. It also must introduce some element of probability. It's not like opening a book and reading "Kasparov says to move the bishop here..." since the programmer must still weigh all the data at his disposal and make an informed choice based on many different components. The same thing is true for every other stage of the program. Data is read in and analyzed. This analysis is used to form a decision.
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.