Author: Alessandro Damiani
Date: 08:32:27 09/17/99
Go up one level in this thread
On September 17, 1999 at 08:57:41, Bas Hamstra wrote:
>On September 16, 1999 at 13:50:18, Alessandro Damiani wrote:
>
>>On September 16, 1999 at 12:46:20, Peter Fendrich wrote:
>>
>>>On September 16, 1999 at 05:18:42, Bas Hamstra wrote:
>>>
>>>- snip -
>>>
>>>>History:
>>>>
>>>>(I never used it)
>>>>
>>>>A table of counters, used for movesorting. You use a table History[From][To] to
>>>>keep track of the number of beta cutoffs for each move [From][To]. And use that
>>>>to improve the probability of a cutoff (as you do with killers). If you already
>>>>use more than 1 killer I have the impression History doesn't give much extra.
>>>>
>>>I don't agree with you here. My implementation of killer table only has effect
>>>on local parts of the tree. For each new path the killer moves are cleared. The
>>>history table is covering all of the tree and really improves move ordering in
>>>some nodes. Normally there is no big overlap betwee the two.
>>>//Peter
>>
>>I agree with you, Peter. That is the idea of history heuristic. It should not
>>only be used when a cut-off occurs, but always when there is a best move.
>>
>>And since everybody knows that there is no best move at a fail-low, the history
>>table should only be updated if
>>
>> best>alpha
>>
>>where best is the score of the best move.
>>
>>Alessandro
>
>How do you update killers? I store a killer when anything but the first move
>(not being a capture) is > Alpha. That seems to work better than storing only
>when >= Beta.
>
>It works less well when:
>
>a: I also include first moves
>b: I update killers only when move >= beta in stead of > Alpha.
>
>
>Regards,
>Bas Hamstra.
I update killers if
not capture(bestmove) && best>=beta
since captures are searched before killers.
Alessandro
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.