Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: repetition detection & hashing now with text

Author: Robert Hyatt

Date: 06:24:17 01/16/02

Go up one level in this thread


On January 16, 2002 at 03:31:21, Tony Werten wrote:

>On January 16, 2002 at 00:08:19, martin fierz wrote:
>
>>On January 15, 2002 at 23:38:54, Robert Hyatt wrote:
>>
>>>On January 15, 2002 at 21:50:52, martin fierz wrote:
>>>
>>>>aloha,
>>>>
>>>>i have a question about repetitions, and it's only about the search itself, not
>>>>about the game history: in my checkers program, to find repetition draws i keep
>>>>track of the hash keys of the current variation, and if i find that the key for
>>>>the current position has already ocurred before in the search, i return zero
>>>>without searching further. i think this is the standard implementation?!
>>>>anyway, i am wondering about the following problem: imagine you are searching a
>>>>variation, and in the end you make a repetition. the program will stop searching
>>>>there in the tree and assign the position a value of 0. now imagine that you
>>>>have a different path of moves without a repetition which ends up at that node,
>>>>probes the hashtable and returns 0. is this just my imagination, or can this
>>>>happen and can it be bad?
>>>>
>>>>cheers
>>>>  martin
>>>
>>>
>>>It can happen.  As can the inverse, where you retrieve a real score from the
>>>hash table but you can never reach the position because there is a three-fold
>>>repetition between where you are and where you are going...
>>
>>so... is there a solution to this problem?
>
>Sort of. Make 2 changes.
>
>1) In your eval, if the score is 0 make it 1 for side to move.
>2) Never store nodes with score=0 in your hashtable. (Now you're sure it's a
>repetition.)
>
>Tony

Not quite.  Now you miss repetitions that you would find by hashing, because
hashing often lets you see deeper (due to branch grafting) than the basic
search can see by itself.  You lose some either way.  Best to fully do hashing
and tolerate the few problems there...




>
>>or is it just tacitly assumed that it
>>"never happens"?
>>
>>cheers
>>  martin



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.