Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How do hash tables help move ordering? (Fail Low)?

Author: Roberto Waldteufel

Date: 07:09:32 09/21/98

Go up one level in this thread



On September 20, 1998 at 20:59:55, Robert Hyatt wrote:

>On September 20, 1998 at 19:49:57, Roberto Waldteufel wrote:
>
>>
>>On September 20, 1998 at 17:49:17, Inmann Werner wrote:
>>
>>>On September 20, 1998 at 17:30:31, Robert Hyatt wrote:
>>>
>>>>On September 20, 1998 at 14:21:15, John Coffey wrote:
>>>>
>>>>>I think that the subject says it all.
>>>>>
>>>>>John Coffey
>>>>
>>>>
>>>>Yes.  You store the best move when you store a position, assuming you
>>>>have a best move (you don't if you searched all moves and failed low,
>>>>of course.)  You should try this move first each time you encounter this
>>>>position...
>>>
>>>Whats the thing with the fail low?
>>>Why can I not use the Hash table for move ordering in this case?
>>>(I do now!)
>>>
>>>Werner
>>
>>Hi Werner,
>>
>>When a fail low occurs, there is no guarantee that any of the moves that were
>>searched are worth as much as the respective scores they produced (they can't be
>>worth more, but could be worth less, maybe a lot less). Thus the move with the
>>best score is not necessarily the best move....
>>
>>Best wishes,
>>Roberto
>
>
>problem is, you don't even get "real scores" of any kind... only an indication
>that the score is <= alpha, whatever alpha is...


On odd occasions you can get score<=some value that is itself lower than alpha
if you use the fail-soft version of alpha-beta, which is useful for the upper
bound in the hash table, but this is a fairly rare occurrance. In any event, it
is not possible to infer anything about the relative merits of any of the moves
from the values they returned if a fail-low occurs at the node.

Best wishes,
Roberto



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.