Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Search Instabilities

Author: Robert Hyatt

Date: 21:44:53 01/21/02

Go up one level in this thread


On January 21, 2002 at 14:27:34, David Rasmussen wrote:

>On January 21, 2002 at 10:41:39, Robert Hyatt wrote:
>
>>
>>You have to do two things:
>>
>>1.  If you get a fail high at the root on a zero-width window (any move after
>>the first move should be searched with a zero-width window) you can't trust it
>>unless you re-search it with a bigger beta bound and make _sure_ that it doesn't
>>then fail low.  Such fail-high (zero window) fail-low (non-zero window) is an
>>artifact of null-move and if you play such a fail high move even if it fails low
>>on the re-search, you will die...
>>
>
>If you mean, do I count value >= alpha+1 from zero-window search as a fail high,
>then no. In that case, I research with the original alpha;beta window. Isn't
>that ok?
>


What if you run out of time?  You failed high on the null-window search.
You started a new search with a wider window and time expired.  Do you play
the fail-high move or stick with the previous best move?  I stick with the
last verified move.

Unless I fail high a second time which means the original aspiration window
was too small and I am now going to +infinity.  I trust the second fail high
but not the first.




>>2.  If you get a fail high at the root on the zero-width window, and when you
>>re-search you get another fail high on the original aspiration window, or you
>>get a true score, then you can trust it.
>>
>
>That's what I'm doing now, I think.
>
>>Case 1 is the killer although it doesn't happen every move, once in a game
>>is more than enough.
>
>I'm sure.
>
>But when I posted, I was really talking about the case where you search with an
>aspiration window, and then you get a fail high with value v. You search with a
>new aspiration windows (in my case [v;MATE]), but then the search turns out to
>find nothing. The fail high was false. How do I avoid that?
>


You just take that and go with it.  It is easy to see how it can happen,
where a hash entry says "score >= XXX" which will cause a fail high, but
now you can't use that and if you can't see it with a search, you are now
forced to fail low.  But the fail high was _right_.

It is that PVS null-window fail high that can't be trusted at all...





>/David



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.