Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: To Bob: Crafty, move ordering with SEE

Author: Robert Hyatt

Date: 10:05:23 09/09/99

Go up one level in this thread


On September 09, 1999 at 12:17:56, Steve Maughan wrote:

>Bob
>
>A couple of questions.  Firstly, I've seen in a couple of posts you have give a
>measure of Move Order Quality as the percentage of cutoffs that are as a result
>of the first move played.  Surely this is really a measure of Hash Table
>Strategy since it is the hash table move that is played first (Crafty's hashing
>strategy being sophisticated and including internal iterative deepening). Is it
>not the case that if you swapped Killer and History moves, the statistic would
>not change? Am I missing something?

Yes.. You are assuming that nearly _every_ position is a hash hit.  This is
wrong.  IE only about 10% or so of typical middlegame positions result in
hash hits.  Which means that 90% of the time I am relying on _other_ move
ordering to pick the best move.  Most of the time this is a winning capture,
but killers and history do help.


>
>Secondly, when talking about Killers, is this the last (two) moves that created
>a cutoff at this level or do you store refutational killers ie a killer based
>upon the last move?


I store the last two moves that either failed high or backed up a true
score.  Note that this doesn't gain much over history moves, if anything,
unless you do as I do and try the killer moves without generating any moves
first.  Suddenly when a killer causes a cutoff you avoided generating the non-
captures totally, a time saver...


>
>Thanks
>
>Steve Maughan
>
>>I do it like this:
>>
>>1. transposition move
>>2. positive captures sorted by SEE score (best capture first)
>>3. killer 1
>>4. killer 2
>>5. history moves (4)
>>6. rest of moves, which just happens to do remaining captures first
>>   since they were generated first.
>>



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.