Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: quiescent search and bad values

Author: Dan Honeycutt

Date: 10:28:02 08/29/04

Go up one level in this thread


On August 29, 2004 at 12:47:31, Dan Honeycutt wrote:

>On August 29, 2004 at 12:13:33, Rick Bischoff wrote:
>
>>Hello,
>>
>>I think I am missing some core piece of knoweldge here-- Let me explain the
>>situation.
>>
>>Let HASH_BAD = INFINITY = 16384
>>Let MATE = 8000
>>
>>Stack Trace:
>>
>>0: call alphabeta(depth = 8, -HASH_BAD, HASH_BAD)
>>1: alphabeta makes some move "m"
>>2: alphabeta(depth = 7, -HASH_BAD, HASH_BAD) makes the null move
>>2: alphabeta(depth = 4, -HASH_BAD, -HASH_BAD + 1)
>>3: alphabeta continues drilling down the left most path of the tree.. so we
>>eventually reach...
>>
>>
>>x: alphabeta(depth = 0, -HASH_BAD, -HASH_BAD + 1) calls quiescent(-HASH_BAD,
>>-HASH_BAD+1)
>>
>>Now quiescent does a static evaluation, finds that it is better than
>>beta=-HASH_BAD+1 (how could it not be!) and returns beta...  So we return to
>>frame "x" of the stack trace and it stores -HASH_BAD+1 into the hash table.
>>
>>I assume this is a bad thing-- I have some debugging flags set so that the
>>program halts if i try to store a value outside of (-MATE, MATE).
>>
>>What I am doing wrong?
>
>
>No need make a null move if beta = INFINITY but that doesn't really matter.  The
>null move simply fails to give a cutoff (and nothing goes in the hash table).
>The first position you score on the leftmost branch with -INFINITY, +INFINITY
>must return a value between alpha and beta.  So the first entry in the hash
>table is an exact score.  From then on it's scores or limits depending on how
>things go.
>
>Dan H.

Or another possibility if your problem is not related to what you store in the
hash.  If you do a PVS or scout search with limits of alpha, alpha+1 that search
must return alpha (or less).  If you get back alpha+1 (or higher) search has
failed and you have to redo it with alpha, beta.

Dan H.



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.