Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: learning evaluation weights (was Re: Genetic algorithms for chess?)

Author: Don Dailey

Date: 08:19:46 05/25/98

Go up one level in this thread



>The 1,3,3,5,9 ratio has stood the test of time ...

Has it really?   I don't think anyone believes these numbers are
correct.  They are simply understood to be as close as you can get
using integer values with the pawn being a single unit.

> with the only other
>significant one that I have seen is the one by Larry Kaufmann who put it
>at 3,10,10,15,29    3=pawn  29 = queen   with the 2 bishops =21

This is based on the same observation.  Me and Larry discussed this
stuff at length and he came up with these values as an improvement
over the traditional ones.  His reasoning was to find the lowest
possible "resolution" that made sense.  He didn't like quarters, where
a pawn is 4 points and decided  "thirds" was a pretty reasonable scale.
We were designing a program at the time and decided 96 would be the
value for a pawn, a number with lot's of divisors (3 being one of them)
and close enough to 100 that it would be easy to talk about.   Of course
we were probably being a little "anal" but we like to be precise!

>These point counts are important and I didn't question the experiment
>itself. I simply questioned the lack of search depth for a chess
>experiment that purports to come up with a different point count and
>pass this point count off as maybe more accurate than the historical
>one.

Read your post, you called the experiment flawed (not the results.)
I don't think he was trying to pass off a new system either.

Anyway, this exchange has been good, it clarifies the way you actually
see the experiment and it's results which were not obvious from
your original post.

- Don



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.