Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Couple of chess programming questions: another MTD drawback

Author: Jeremiah Penery

Date: 10:53:17 09/11/02

Go up one level in this thread


On September 11, 2002 at 13:20:55, Vincent Diepeveen wrote:

>On September 11, 2002 at 12:31:44, Robert Hyatt wrote:
>
>>On September 10, 2002 at 20:45:43, Vincent Diepeveen wrote:
>>
>>>On September 10, 2002 at 18:06:01, Gian-Carlo Pascutto wrote:
>>>
>>>>You want to store two actual values, not flags that indicate what
>>>>kind of bound it is.
>>>
>>>did i implement it smarter then or what?
>>>i used 2 bits in total. 'upperbound, lowerbound, truebound'.
>>>the search result is based upon a single bound. So it IS the same,
>>>it IS higher or it IS lower.
>>>
>>>What am i missing here?
>>>
>>
>>
>>A _lot_.
>>
>>This is a known issue with mtd(f) and fail-soft.  If you bounce over the
>>true score, to the "other side" then suddenly where you were storing
>>upper bounds you are now storing lower bounds, and vice-versa.  Now if
>>you bounce back over the true score again, you have no useful upper bounds
>>where you need them, you only have lower bounds.  And you have no useful lower
>>bounds where you need them you have only upper bounds.
>
>I do not see *anyhow* how you can jump over the score.

Because it's stupid to _always_ increase the window by .001 (or .01).  It will
make you do far too many researches.



This page took 0.04 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.