Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Killer / History

Author: Robert Hyatt

Date: 18:29:34 04/16/00

Go up one level in this thread


On April 16, 2000 at 20:51:56, Georg v. Zimmermann wrote:

>>> idea of Killermoves is outdated and almost everyone only uses History today. >>Is that correct ?
>>
>>No it isn't.  Technically history moves provide everything killer moves
>>provide, and then some.  But a plus for killers is that you can try them
>>before generating any moves, and if you get a cutoff due to the killer, you
>>avoid the move generation completely.  History doesn't let you do this as
>>the data structure doesn't contain moves...
>>
>
>This is interesting! I didn't think about that. The program I'm working on right
>now adds the Killers in orderMoves() when the moves are generated already,
>loosing time. So I should maybe rewrite it to look at killers first and then do
>the move generation. Does that mean that you search killers before promising
>captures ?
>
>>>


No. In crafty, I generate moves in two chunks because it is easy with
bitmaps. I generate captures first, then search the promising ones, then
I try the killers before generating the non-captures and trying them.



>>> What is the best way to see how good my move ordering is ? Should I simply
>>> count how often the move "ordered" as number 1 turns out to be the best
>>> move ?
>
>>
>>That is a good approach.  I do this in Crafty, and average 92%.  That is,
>>92% of the time, when I get a fail high, it is on the first move searched.
>>
>
>I was just typing another question here (half a page with 2-finger system) when
>I found the answer myself. :-)
>
>Thanks for the help,
>
>Georg



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.