Author: Ulrich Tuerke
Date: 09:39:03 07/18/01
Go up one level in this thread
On July 18, 2001 at 12:18:13, Rafael Andrist wrote:
>>>The problem happens at the root of the search. I get a hash-cutoff, so no move
>>>is investigated and the hash move is returned. This is the code I use:
>>>
>>>//there's a appropriate hash entry, read the bounds/value:
>>>if (iHashPtr->SearchDepth >= RemainingDepth)
>>>{
>>> if (iHashPtr->Flag == HashFlagExakt)
>>> return iHashPtr->HashValue; //the returned move is the hash-move
>>> else if (iHashPtr->Flag == HashFlagAlpha && iHashPtr->HashValue > Alpha)
>>> Alpha = iHashPtr->HashValue; //adjust Alpha-Bound
>>> else if (iHashPtr->Flag == HashFlagBeta && iHashPtr->HashValue < Beta)
>>> Beta = iHashPtr->HashValue; //adjust Beta-Bound
>>>
>>> if (Alpha >= Beta)
>>> return iHashPtr->HashValue; //the returned move is the hash-move
>>>}
>>>
>>>Please, tell me what is wrong in this code.
>>>
>>>Rafael B. Andrist
>>
>>I don't hash at the root because I can't see much reason for it.
>>
>>I guess, you are doing this in order to get a best move which you try first in
>>mover ordering and perhaps you do this to get a reasonable start-window for the
>>score.
>>This should be no problem. But in any case you should then start your search
>>(with iteration nr.1), ignoring depth and score retrieved by the hash probe.
>>
>>Uli
>
>I'm going to fix it. But ignoring of hash score from previous search only at
>iteration 1 at the root or allways?
No, you may use it in the search. You should however have some mechanism to "age
out" entries from previous moves. Otherwise your hash table will sooner or later
be blocked by many old and unneeded entries.
Comet's hash entry has a component "move_nr", which is stored in addition to
depth, score, bound-flag & move. When the current move nr is larger than the
stored one, the old entries are still used when probing but they may be
overridden by any current search info.
BTW, your code above looks fine to me.
Uli
>
>Thanks
>Rafael B. Andrist
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.