Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Limit extensions

Author: Tony Werten

Date: 23:05:18 07/06/04

Go up one level in this thread


On July 06, 2004 at 14:15:05, Robert Hyatt wrote:

>On July 06, 2004 at 14:05:24, Eric Oldre wrote:
>
>>On July 06, 2004 at 13:19:11, Robert Hyatt wrote:
>>>
>>>With fractional extensions, I believe it works best like this:
>>>
>>>extended=0;
>>>if (something) extended+=ext1;
>>>if (something_else) extended+=ext2;
>>>...
>>>etc.
>>>...
>>>
>>>extended=Max(extended,one_ply);
>>>
>>>If you extend more than one ply you run the risk of reaching a position where
>>>your search will be non-terminating.  Your code runs the risk of not extending
>>>enough.  IE if your first extension is one ply and the last above is 1/2 ply, if
>>>the 1/2 ply extension is triggered, that is all you will extend.  Add 'em up and
>>>limit them to 1 ply total should be better...
>>>
>>
>>Robert,
>>I have heard you mention before that you limit extending to one ply. What I've
>>never been clear on is do you limit the extension to one ply for the entire
>>search? or to extending only one ply per node?
>
>One ply per node.  More than that and the depth can actually increase,
>obviously, rather than getting smaller as you go deeper.  1 is potentially
>dangerous but it is not very likely that you would extend 1 ply every ply you
>search.
>
>Deep Blue used the rule "no more than two plies of extensions for every two
>plies of search" which averages out like mine, but you could extend 2 plies at
>ply=N if you extend 0 at ply=N-1.  I tried it for a while but didn't like it as
>well as what I do, as it tended to over-extend...

Actually, I think it was a bit different. The used "never more than 2 ply"
combined with "only extend on fail highs" wich basicly resulted in "never more
than 2 ply in 2 moves" (special care was taken for pv nodes )

It's the "only extend on fh" that is typical about the the Deep Blue extensions
( not the 2 ply in 2 moves), and it limits the problem of overextending.

Tony

>
>
>>
>>ie. if you are calling Search from the root with depth 10, is the maximun number
>>of plys it will look ahead 11? or could it be more if you have criteria for an
>>extension on multiply plys?
>
>In theory a 10 ply search could search to depth=infinity, but in practice
>extending at _every_ ply is not likely.  But I regularly see 30+ ply search PVs
>from a 15 ply nominal search depth.
>
>
>
>>
>>Eric



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.