Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Uri's ETC

Author: Tord Romstad

Date: 14:33:23 03/24/04

Go up one level in this thread


On March 24, 2004 at 17:26:21, Dieter Buerssner wrote:

>On March 24, 2004 at 17:14:26, Tord Romstad wrote:
>
>>On March 24, 2004 at 16:52:04, Dieter Buerssner wrote:
>>
>
>>>Are there cases, where ETC on needs one depth more, to solve a tactical
>>>problem, than ETC off? I would guess, that this could be the case now and
>>>then.
>>
>>I am 100% sure it happens now and then, and I'm also 100% sure that the
>>opposite thing happens.
>
>Yes, I don't doubt, that the opposite can happen (although it would not happen
>in a plain alpha-beta search without extensions/pruning).
>
>>One of the things I consider when making pruning
>>and reduction decisions is the history of the move.  Moves which have
>>very rarely failed high in the past are more likely to be pruned or reduced.
>>Of course, it will happen many places in the tree that one particular move
>>at one particular node is pruned with ETC disabled and not pruned with
>>ETC enabled, because the history count of the move will be different in the
>>two cases.
>>
>>Even tiny changes in my move ordering often results in a tactical problem
>>being solved one ply earlier or later.
>
>No doubt about this. Without having it tried yet, I feel that ignoring extension
>decisions for ETC is more severe, however (somehow like a first order error
>term, while other things you mention are second order error terms). After all,
>we think we have a good reason to extend some moves. Getting a cutoff into the
>search tree (or better graph) by ignoring that reason seems dubios to me.

It is very possible that you are right, and I should certainly do some
experiments with this again.  Thanks for reminding me.

>>>Are you trying ETC after hash probing and before null move?
>>
>>No, after both.
>
>Can you elaborate? I don't see the sense in doing this twice.

Sorry, I wasn't clear enough.  I don't do it twice.  I do hash probing
first, then null move, and finally ETC.

Tord



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.