Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: What is the average nodes per second for minimax?

Author: blass uri

Date: 10:01:46 06/19/00

Go up one level in this thread


On June 19, 2000 at 12:47:05, blass uri wrote:

>On June 19, 2000 at 11:40:26, leonid wrote:
>
>>On June 19, 2000 at 06:41:52, blass uri wrote:
>>
>>>On June 19, 2000 at 00:32:26, leonid wrote:
>>>
>>>>On June 18, 2000 at 19:18:10, Tom Kerrigan wrote:
>>>>
>>>>>lenoid, do not call material evaluation "positional." They [basically] have
>>>>>nothing to do with each other. Also, do not refer to evaluations as "partial" or
>>>>
>>>>
>>>>Logic that deel with material echange is for me 100% positional logic. Reason is
>>>>simple - slightest change in position on the chess board lead sooner or later to
>>>>some material lost or gain. If you look into material echange, you look into
>>>>position on it as well.
>>>
>>>People do not use the word positional for material.
>>>
>>>positional evaluation is the part of the evaluation function that is not about
>>>counting material.
>>>
>>>When somebody wins material in game it is called tactics.
>>>Positional understanding is about the value of the not material factors.
>>
>>
>>And why somebody would like to even care about "positional understanding"? Not
>>for the sake to reach somewhere material gain?
>>
>>Don't accuse me to be too materialistic. I am not true capitalist. I hardly have
>>some money on my bank acount.
>>
>>
>>>Positional understanding is important in chess because there are position when
>>>one side has less material but better position.
>>
>>
>>Better position that lead later to some material advantage? If you will look
>>into this through the logic of material echange, you will reach "positional
>>understanding" of this position in a very "natural way".
>>
>>
>>>If you search deep enough you can see that the side with material advantage is
>>>losing but there are cases when you need to see 50 plies forward to see that the
>>>side with material advantage is losing and computers cannot see it by search.
>>
>>If material echange deep 50 plies will reveal something about position, fact is
>>still there - material echange say everything about this position.
>>
>>
>>>A good program must have positional evaluation that tells it in part of the
>>>cases even without search that the side with material advantage is probably
>>>losing.
>>
>>Agree! But this is only because material echange can't see that deep in most
>>places. We must, at least, have some general guide line when we don't know where
>>to go through perfect calculation of material. Those guide lines, that say where
>>to go when we don't know for sure what to do, are called tactics.
>
>No
>The guide lines that say where to go when we cannot calculate are called the
>evaluation function.
>
>Tactics is about search and not about evaluation function except material.
>
>If we can see by search that we win material it is tactics but in most of the
>cases we cannot see by search that we win material and only material evaluation
>will give the value 0.00 after search.
>
>I believe that even if we assume that programs are 100000 times faster they
>cannot search deep enough to see winning material in most of the positions.
>
> We are still
>>very much in this domain because our computers are very weak. Before it was even
>>more so. I even presume that "tactic" 20 years ago was even more important that
>>today. Pobably it will become less relevent after the next 20 years as well.
>
>I believe that the not material part of the evaluation function will become more
>important in the future.
>
>If you see 2 plies deeper than your opponent tactics is important if both sides
>cannot search deep and the not material evaluation is less important because
>you can destroy a good position by one blunder.
>
>If both sides can search deep the probability for tactical mistakes(losing
>material) is smaller and the quality of the evaluation function is more
>important.
>
>Having a good evaluation function can push the program with the better
>evaluation function to sacrifice material for initiative and to win the game.
>
>I saw too many cases when one side sacrificed material when the win is too deep
>to see by search to tell you that having big not material evaluations of more
>than a pawn and in some cases more than 2 or 3 pawns is important for the
>future.
>
>>Reason for this - material echange say you everything about position and tactics
>>all the time. Question is only: How far it can see?
>>
>>Leonid.
>
>I believe that only material evaluation has no chance against GM's even if we
>assume that programs are 1000000 times faster.
>
>Uri

I can add that the position that your program needed 12 plies to find Qb7 is a
good example for it.

analysis showed that black has a losing position even after Qb7 and I guess that
the problem in the game was that the program was too materialistic.

Suspecting by evaluation that black is losing before Qb7 is important
It is of course not easy to do it without other mistakes in the evaluation that
can lead to wrong sacrifices.

Humans can see by evaluation that white has an attack
They need to analyze in order to see if the attack is enough to win but no
computer think in this way.

programs should evaluate the position as unclear and do some selective search to
get a better evaluation function because it is impossible to see everything by
search(especially when this position is only position that can be in the future
of the game and not the root position).

It is of course not easy to do it but I believe that this is the right way to go
forward.


Here is the relevant position:
[D]4r1k1/p4ppp/1pq1P3/3p1PPP/3B1Q2/P4NK1/8/3n4 b - - 0 1


Uri



This page took 0.03 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.