Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: an idea to test positional understanding of chess programs.

Author: Chuck

Date: 15:03:02 05/24/99

Go up one level in this thread


On May 24, 1999 at 17:54:52, blass uri wrote:

>
>On May 24, 1999 at 17:38:52, KarinsDad wrote:
>
>>On May 24, 1999 at 17:05:04, blass uri wrote:
>>
>>>
>>>I agree that a better implementation can also help, but the main point is that
>>>the take back can help in cases of tactical mistakes that you understand after a
>>>small number of moves that you are in trouble.
>>>
>>>If the position of one program (with better positional understanding) is going
>>>to improve slowly then the opponent will have no idea when it went wrong so the
>>>take back is not going to help it.
>>>
>>>Uri
>>
>>Does this mean that the takeback will be limited to one ply at a given time?
>>
>>If so, then that may not be enough to undo a tactical problem (i.e. it takes 14
>>ply to see the flaw, but the program only looked 10 ply).
>>
>>If not, then it may be possible to take back to the point of minimizing the
>>positional advantage (i.e. the program searches for any way to avoid the trap it
>>walked into, even going back 4 moves).
>
>
>It is not going to be linited to 1 ply but it is going to be limited because the
>program is losing 5 minutes for every move that it take back to the opponent so
>if it take back 40 moves then it is going to lose 5*40=200 minutes and this say
>losing on time.
>
>If a program does many tactical mistakes then it is going to lose but if a
>program is a better positional player and does one  tactical mistake for 40
>moves then the take back is going to help it.
>
>Uri
>
>>
>>This is hard to evaluate.
>>
>>KarinsDad :)

In addition to take back, it would be important for the program to note any
conditions related to it, so you could analyze it later and see why. Also,
I think it would generally have to pull back 5 moves or more, and punishing the
program by taking away 5 minutes for each move would mean that it would likely
just step right into another hole. So I don't think time should be taken from
the program doing the takeback, giving the opponent 5 minutes would suffice.

Yes, I could see this as a valid method for improving a computers play, but
then again you go back to a question of whether the knowledge needed is overall
worth the time cost. The likely idea is to use just enough knowledge to
score more than 50% against any opponent (thus you could win a match) since
the search speed seems to be so critical to computers.

Chuck



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.