Author: Robert Hyatt
Date: 07:30:42 06/14/98
Go up one level in this thread
On June 14, 1998 at 01:32:22, Hristo wrote:
>On June 14, 1998 at 00:49:33, Robert Hyatt wrote:
>
>>On June 13, 1998 at 22:35:54, Hristo wrote:
>>
>>>Hello Don!
>>>
>>>>... Since I consider myself to be a scientist I will always
>>>>collect facts and definitions and work within them. ...
>>>
>>>What is the power of prediction that the total TACTICS method gives you
>>>?!
>>>It's not that good, is it !? If you are a scientist how can you stand,
>>>so firmly, behind a method which has such low level of "prediction" ...
>>>the current state of the chess programs has very little scientific
>>>value(for chess it self). It has much more comercial value !!!
>>>"My" program is stronger than "his" hence I know better !!! Bull ...
>>>nobody knows! Lets look at the facts, not with a microscope but with
>>>some understanding to what these facts mean. The facts point out that
>>>even if you manage to search 200 000 000 position per second you are
>>>still not even close to get to the bottom of the problem! Deal with this
>>>fact for a change!
>>>
>>>Hristo
>>
>>
>>then deal with this: If I can search 200,000,000 nodes per second, I am
>>*one hell* of a lot closer to the "truth" about a position than a
>>program
>>that searches 5K nodes per second, no matter how you evaluate things. I
>
>Bob ... remember this. It is very possible that you are not correct!
>Your statement is absolutely wrong ... but this is only if I didn't take
>into account what you mean. Please be carefull when talking about the
>"TRUTH" ...
>you can't handle the "TRUTH". :))
>
>
>>won't be able to see to the end of the game, obviously, but I will see
>>*every* hole in your evaluation, and I'll shove a stick in each one
>>pain-
>>fully. In chess, "rules of thumb" only take you so far. How many times
>>is an isolated pawn weak? strong? immaterial? Ditto for kingside
>>attacks
>>that fail, or succeed, or don't do either...
>>
>>I'd rather go faster, and depend less on rules of thumb that I *know*
>>are
>>imperfect... than to bet the farm on vague concepts that work here,
>>fail
>>there, and guarantee me that I can't play with someone like Kasparov.
>
>
>Bob you sound like Lord Kelvin ... perhaps you are "lord Kelvin".
>I agree that under given circumstances "pure full search" can achieve a
>perfect results in the "chess" game. While this day comes isn't it
>possible to experiment and test other ideas ... perhaps ideas that can
>give us an actual understanding of the game, not just a possible
>outcome.
>Why are you so protective of "your" ideas?! Do you, even for a moment,
>think that there might be another way of finding the correct move?!
>I do ... I'm writing my own program right now ... The program does not
>depend on search. The program is not based on any type of predetermined
>chess axioms or "rules of thumb"! It doesn't even know what is a
>doubled-pawn ... I'm still in the beggining ... it already plays chess
>(more or less) ... but the engine is not ready yet. I can already see
>that out of a single position one can get enough information to
>determine which moves are better. This doesn't come for free ... the
>evaluation function requires LOTS and LOTS of time. However, this time
>is insignificant compared to the time Crafty needs to get to 10ply(after
>1. f3).
>
No, I don't believe there is a viable alternative to "search" since we
"humans" rely on exactly the same idea. And if *we* have to search to
find tactical wins, the computers are obviously going to follow that
same
path. We've been through this for years and years now, and, without
fail,
the better the search, the better the result. And the better the
search,
the less it depends on evaluational rules-of-thumb to be successful.
For
example, early programs (mine included) *had* to evaluate pins and
forks.
Because the search wasn't good enough to understand them. Most programs
now are not even "aware" of either, because the search understands them
in passing over the positions and going deeper.
As to your "evaluation" understanding things, you are using a rule-of-
thumb because you don't understand (through provable search outcome)
that
something is likely to be good or bad. *if* search can see what is
happening
it won't ever be wrong, while rules of thumb *always* have exceptions.
And
huge chunks of code to handle such cases not only have exceptions, they
*always* have bugs as well.
>Take care Bob("Lord Kelvin") :))))
>Hristo
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.