Author: Christophe Theron
Date: 08:56:42 06/01/99
Go up one level in this thread
On June 01, 1999 at 05:51:49, Bernhard Bauer wrote:
>You surely know how Crafty uses null move. From search.c you can see:
(snip)
>Bob Hyatt doesn't use null move if there are "few" pieces on the board. That
>could be viewed as low mobility. Of course other views are possible.
This rule is simple but too simple. There are ways to get much better
performances in the endgame. But of course you have to live with a few
"holes"...
>Zugzwang detection is another thing that may help.
But "zugzwang detection" does not exist! If somebody is able to write a
"zugzwang detection" routine, then the problem is solved. But there is no easy
way to detect that a side is in zugzwang, except by doing a search. And the cost
is too high.
>This "smart" choice makes you (and the users of your program) sad when your
>program fails to solve a certain position.
I keep an algorithm when it makes me more happy than sad. :)
>>>The explanation "the computer can't do it" has proven wrong in decades.
>>
>>I don't understand why you say this... Did I say that the computer can't do it?
>>My program does it for example.
>>
>No you haven't said this. I do not argue against you. I simply don't like the
>holes in the programs. I want them to disapea.
Don't you think I want the same?
But "holes" are not the only problems chess programs have to face. You look
stupid when somebody finds a hole in your program, but you look also stupid when
your program is not able to compute deep enough to find a combination that other
programs see quickly.
Don't focus too much on holes. If you do, then what you will get is a simple
alphabeta program with no selection at all. There will be no holes, but it will
takes ages to find simple things...
>>>The question remains: what is the best way to do it. There are ways that will
>>>not slow down your program and will succeed in other positions.
>>
>>I am all hears. Tell us how it is possible.
>>
>See above and have a look at Tiger. As Tiger solves the 2-mover you have
>allready implemented something thats better than in Fritz.
But Tiger fails also badly on some other positions... And you would complain if
you would find them... So it's a neverending story.
>>>So I hope you can understand my point of view.
>>
>>Not if you tell me that I'm taking poor decisions.
>>
>A decision that leads to the wrong result for the gain of speed is poor.
>Finding a result in a somewhat long time is better than finding it never.
I agree. That's why some of my selection algorithms are enabled only in the
deeper parts of the tree. I can miss something, but with longer time I will not
miss it anymore.
Christophe
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.