Author: Tim Foden
Date: 06:53:49 08/16/00
Go up one level in this thread
On August 16, 2000 at 06:49:01, Inmann Werner wrote:
>On August 16, 2000 at 05:32:23, Tim Foden wrote:
>>I guess that it therefore begs this question: does this mean that you should
>>never store a capture move in a killer (as it is a total waste of time)?
>>
>>Cheers, Tim.
>
>Yes, it is a waste of time. Never store captures as killers!!!
>Most positions in search are "idiotic", and one capture refuses them. Therefore
>first captures is always good. (maybe not at an mate in 1, but in deeper
>evaluation always!!!)
>There is also the term "winning captures". Only these should be tried at first.
>Beating a pawn with a queen "most time" is no good idea!
>
>Best thing for move ordering (but slow) would be, to look at each capturing
>move, if it is "promising", and only trie the "promising" at first. Your tree
>will get really small, but your NPS also would decrease. It is a exchange, but
>worth testing..
>
>And do not forget!, the first move to try (before move generation) is the
>Hashmove!!!
In GreenLight the move ordering is currently:
1) try hash move.
2) generate captures, sort so that "winning" captures tried first,
only use the "winning" capture moves at this time.
3) try 2 killer moves.
4) generate the rest of the moves, tacked on the end of the "losing"
captures, and select from them using the history heuristic table for
the next 5 moves.
5) use the rest of the moves in the order they were generated.
This is basically using the ideas from Crafty 11.17. I haven't looked at a
recent Crafty, but I suspect it is similar.
But I guess it is never worth putting the "winning" captures in the killers.
Cheers, Tim.
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.