Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: DB and Singular Extensions

Author: Peter W. Gillgasch

Date: 19:06:00 11/22/98

Go up one level in this thread



On November 22, 1998 at 19:36:30, Robert Hyatt wrote:

>On November 22, 1998 at 18:04:04, jonathan Baxter wrote:
>
>>Can anyone send me an electronic version of the ICCA article by Hsu on
>>Singular Extensions used in DB?
...
>pv-singular move:  you search the first move at a node with the normal
>window, and the remainder of the moves at that depty with a lowered (offset)
>window (alpha-w, beta-w) where w is the singular margin.  If all those moves
>fail low, the first move is "singular" since it is clearly better.  If any
>one move fails high, you have to do the original test on the first move to
>see if this new best is singular or not...  messy but understandable.

It is very messy indeed. 1st you do your usual full window search,
then you do the closed window pv singularity probe searches, then
you have to make sure that the move that failed high is (not)
becoming part of the PV (simple PVS), then you have possibly check
all the moves versus the PV singularity margins as Bob said and
some other minor implementation details.

Then you will see the hash table problems and oscillating trees
(non terminating searches :) if you break that one...

After you did that, you probably lost interest in implementing FH
singularity probe searches. At least that is what happened to me :)


>fh-singular move:  you search the first move at a node with the normal
>window and get a fail high.  You first search the *remainder* of the moves
>with the offset window, but to reduced depth (say D-2) to see if all of them
>still fail low with the offset window.  If so, extend.  If not, do the same
>check as in pv-singular.
>
>The idea is that at a PV or fail-high node, you try to prove that the best
>move is better than all other moves at that ply, by a significant margin.  No
>"fail-low" node test has been formulated...

Well the problem is that no one can come op with any problem that needs
FL singularity as an answer ;)

>If that isn't clear, feel free to ask more...
>
>Bob



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.