Computer Chess Club Archives


Search

Terms

Messages

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

Author: leonid

Date: 10:58:26 06/24/00

Go up one level in this thread


On June 24, 2000 at 02:00:49, blass uri wrote:

>On June 23, 2000 at 20:28:28, leonid wrote:
>
>>On June 23, 2000 at 19:22:07, Tom Kerrigan wrote:
>>
>>>On June 22, 2000 at 17:28:07, leonid wrote:
>>>
>>>>On June 22, 2000 at 13:27:03, Tom Kerrigan wrote:
>>>>
>>>>>On June 21, 2000 at 21:18:07, leonid wrote:
>>>>>
>>>>>>On June 21, 2000 at 19:03:40, Tom Kerrigan wrote:
>>>>>>
>>>>>>>On June 21, 2000 at 17:07:06, leonid wrote:
>>>>>>>
>>>>>>>>On June 21, 2000 at 13:38:41, Tom Kerrigan wrote:
>>>>>>>>
>>>>>>>>>If you think that material-only evaluation programs are good for anything,
>>>>>>>>>you're sadly mistaken.
>>>>>>>>
>>>>>>>>
>>>>>>>>I said only that material evaluation is evaluation about everything in principe.
>>>>>>>>About tactics... or just say it. I agree that in actual state of hardware it is
>>>>>>>>not enough have only material evaluation, but its importance  will grow as
>>>>>>>>rapidly as hardware capacity will improve. Very soon program that have in its
>>>>>>>
>>>>>>>Only evaluating material has zero importance. Why would you do it when you can
>>>>>>>evaluate material AND positional terms with no penalty? Besides, material is
>>>>>>>just a rule of thumb, just like any positional term. Thinking that you can make
>>>>>>>a good program by only considering material is absurd, no matter how fast your
>>>>>>>computer is.
>>>>>>>-Tom
>>>>>>
>>>>>>Do we speak about my program or about general idea? If we speak about my program
>>>>>>it is not that interesting, since we will talk only about one program in
>>>>>>particular. When we speak about general idea, yes, material echange can say
>>>>>>everything. Only through the material echange you can find mate or draw. By the
>>>>>>same mean you can find all other move in the game, name it positional, tactical
>>>>>>or otherwise. We can talk how much computer power we need for the best program
>>>>>>right now to find this or other kind of move, but this is something else. Idea
>>>>>>is simple - material echange do everything and everywhere. In chess game logic
>>>>>>is enough to see everything in it from beginning up to the end.
>>>>>
>>>>>No, you're not listening. "Material exchange" (why can't you call it
>>>>>material-only search or something?) is not the only way to find mates or draws.
>>>>>In fact, you don't need to know a damn thing about material to find mates or
>>>>>draws. That's my point. As soon as you start evaluating material, why wouldn't
>>>>>you also evaluate e.g. doubled pawns at the same time?
>>>>
>>>>
>>>>Two times tried to respond on the question of doubled pawns but lost two times
>>>>my message. I hope now it will work.
>>>>
>>>>Evaluation of doubled pawns should be regarded (just as example) as two stage
>>>>search.
>>>
>>>What's the point of doing the search in 2 stages? Why not do it in 1? Everybody
>>>in the entire world except you does a "1 stage search."
>>>
>>>There are a number of things that you can cut into 2 parts, but it doesn't mean
>>>you should cut them.
>>>
>>>-Tom
>>
>>Reason is simple, in many instances you have just one clear response. When it is
>>the case, you have no more need in making any choice at all. Only when you have
>>few responses (few because they have the same value), now you must chose between
>>them. And you do this choice on the second stage.
>
>I am not sure if I understand you.

Ura, think about initial position on the chess board, as the most simple
example. Final result if too far away. So, if you will do your calculation 5
moves ahead, you will have 20 moves of the same value as final result. Second
stage (when you don't use openings) will help you to make the chose between
them.

>Do you say that you first evaluate material and if the evaluation is equal you
>start to check between the equal moves which move is best by positional
>evaluation.
>
>If you do it you do not use the alpha-beta because the alpha-beta gives you no
>idea about the second best move[my opinion is that using alpha beta in the first

Alpha-beta can give you very often as much as 15 moves of equal value in the
first half of usual game. I speak all the time about what I call "positional
logic". As I mentioned it before, multiplicity of the response is the main
problem when you want to know efficency of your positional logic. More judicial
choice between those moves can give the impression that one logic is better that
the next one. All difference could be only in second stage, if other program
work in the same way. Surprises are all the time possible. I did all my
programming alone. Very often when I spoke with other programmers I found that
they can hardly understand what I am been talking about. Before I expected that
all chess programs end sooner or later by the same main structure. But it is not
what I see in reality.

>plies is not a good idea because it is important to have an idea about the exact
>evaluation to know better which lines to analyze(I see cases when programs
>analyze moves that lead to mate in 1 because they do not reject the moves that
>they analyze for the right reason) but you should use alpha beta most of the
>time].

I don't speak about bugs in the logic. I speak how it must work in general
situation.

Leonid.

>You can use the only material evaluation in part of the cases but not always
>(if you do search(alpha,beta,depth) and the only material evaluation tells you
>that the value>beta+maxpositional score you can know that there is no point to
>calculate the positional score because the score will be above beta even after
>adding the maximal positional score)
>
>Uri



This page took 0.01 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.