Author: Robert Hyatt
Date: 14:06:22 03/27/04
Go up one level in this thread
On March 27, 2004 at 15:44:51, Harald Lüßen wrote:
>On March 27, 2004 at 10:28:23, Robert Hyatt wrote:
>
>>On March 26, 2004 at 16:38:35, Joachim Rang wrote:
>>
>>>I'm testing with an engine which allows to define an aspiration window, what are
>>>common values?
>>>
>>>thx in advance
>>>
>>>Joachim
>>
>>
>>+/- .4 to .5 has worked best for me. I've tried even +/- .25...
>
>Does it make sense to start with a big window and narrow it?
>bounds = 100; // a pawn
>for ( iteration ... )
>{
> score = search(...)
> bounds = max( bounds * 8 / 10, 10 );
> ...
>}
>
>And to be speculative: Can I assume that chess is nearly drawn
>and I can always find good moves. If I am threatened and suffer
>a fail low at my starting window around the evaluation score or
>around 0 I need not change the window but the search depth. This
>is part of the motivation of human chess players before the game.
>"I am a stronger player than this guy and I can always draw!"
>I don't believe it myself but why is this wrong (in practice)?
>
>Harald
If you plot nodes searched vs aspiration window, you will get something like
this:
|
| * *
N | * *
O | * *
D | * *
E | * *
S | * *
| * *
| * *
| * *
| * *
| * *
| * *
| * *
| *****
________________________________________________________
0 10 20 30 40 50 60 70 ....
Aspiration Window Width
The idea is that for a given engine, there is an optimal aspiration window.
Make it too small and you do too many re-searches when the root score comes back
outside the window. make it too big and you search a larger tree since the
window doesn't help much.
You might want to handle odd/even ply differently as well as without a lot of
extensions, you can get score oscillation as you go from odd to even and back to
odd...
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.