Author: Christophe Theron
Date: 12:21:02 05/25/02
Go up one level in this thread
On May 25, 2002 at 11:58:04, Daniel Clausen wrote:
>On May 24, 2002 at 20:09:39, Christophe Theron wrote:
>
>[snip]
>
>>In this case you achieve stability by ignoring some information.
>
>Well yes, but on the other hand, it's not at all clear what exactly the
>information is here and/or how to use it further. (at least not to me)
>
>The information seems to be: if you would have used a bigger window, this move
>would probably also have this score, or maybe not.
>
>Sargon
The information extracted by failsoft is correct, to some extend.
If you return a score>=beta (instead of returning score=beta), then it really
means that the real score (the one you would get with a larger window) is
>=score.
That's why I say that failsoft extracts more information. Instead of just
knowing that real_score>=beta, you know that real_score>=score, with
score>=beta.
So if you have to re-search, you could do it with alpha=score instead of
alpha=beta. You would search a smaller window and so save some work.
Things get more complicated when you add pruning systems like null move. Because
in standard alphabeta you cannot get a fail-high followed by a fail-low on the
re-search. When you add pruning systems, you can get that. So the failsoft info
you've got is not as reliable and should be treated with more care.
Christophe
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.