Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: diminishing returns w/ increased search depth?

Author: James Swafford

Date: 18:34:15 01/27/00

Go up one level in this thread


On January 27, 2000 at 18:01:39, Peter Kappler wrote:

>On January 27, 2000 at 13:52:38, Christophe Theron wrote:
>
>>On January 26, 2000 at 20:06:04, Peter Kappler wrote:
>>
>>>On January 26, 2000 at 18:23:50, Bruce Moreland wrote:
>>>
>>>>On January 26, 2000 at 18:10:10, Dann Corbit wrote:
>>>>
>>>>>IOW, more horsepower is a tough way to make chess programs play better.  There
>>>>>is also evidence (according to some) that the increase in speed has
>>>>>*diminishing* returns.  Hence, it may take a terahertz to get there.  Don't know
>>>>>of any material that could do that, not even a Josephson Junction.
>>>>
>>>>I think it's a great way.  You just take a vacation, preferably a long one, and
>>>>when you come back you make one call to Gateway and poof, free Elo points.
>>>>
>>>>Got an article that shows that the Elo curve flattens out with increased depth?
>>>>
>>>>bruce
>>>
>>>
>>>Ironically, Dann and I recently debated this very topic.  If you use the 7-day
>>>filter, you can probably find the discussion under "DBx1000 = how strong?" or
>>>"tactical sufficiency threshold".
>>>
>>>It seems obvious to me that the curve must flatten out with increased depth, but
>>>I'd be fascinated to see evidence to the contrary.
>>>
>>>--Peter
>>
>>
>>I'd be fascinated to see evidence of what you say.
>>
>>
>>    Christophe
>
>Hi Christophe,
>
>I must admit that seeing you and Bruce on the opposite side of the fence has me
>beginning to doubt myself.  :-)
>
>I find this topic particuarly interesting, as I think it says a lot about the
>basic nature of chess.  My intuition tells me that in most chess positions, a
>certain search depth is sufficient to "understand" that position, and avoid any
>serious problems.  Searching beyond that depth would give diminishing returns.
>
>It seems like a simple experiment to conduct.  Just play lots of self-play games
>between an n-ply and n+1 ply search, and after 500 games, increase n.  Keep
>increasing 'n'  Surely these experiments have been conducted.
>
>If you can point me to any relevant articles, I'd be grateful.  I have read the
>"Crafty goes deep" article in ICCA, but the results didn't seem conclusive to
>me.
>
>--Peter

Hi Pete,

Check out http://wwwipd.ira.uka.de/Tichy/DarkThought/node46.html.

An interesting quote from Ernst's page:

For these search depths the ``Best Change'' rates of both CRAFTY and DARKTHOUGHT
stayed range-bound around 16%. As a tentative conclusion we conjecture that the
three columns of Table 1.4 taken together provide convincing empirical evidence
that the very gradual decreases of the ``Best Change'' rates at high search
depths are not only artifacts of specific implementations but rather represent
an actually general phenomenon of chess programs which rely on depth-first
alpha-beta search with iterative deepening.

--
James




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.