Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A New Self-Play Experiment -- Diminishing Returns Shown with 95% Conf.

Author: blass uri

Date: 02:19:00 05/25/00

Go up one level in this thread


On May 25, 2000 at 04:19:42, Michael Neish wrote:

>vidence is inconclusive, IMHO.  And I see _far_ too many positions in games
>>>I go over where one more ply would have made a difference.
>
>>
>>You will always see cases when one ply would make the difference.
>>The only claim about diminishing return is that you will see it more often at
>>small depthes.
>
>Excuse me, but my understanding may be a little skewed.
>
>I can see arguments in favour and against there being diminishing returns for
>deeper searches (which is maybe why no one can say one way or the other).
>
>On the one hand I don't see why there should be diminishing returns if, say,
>the Knight fork that wins your Queen could always be one more move down the
>line.  If you're at move 30 and the deadly fork is 15 ply later, and you can
>just see it, then if you put the position back a couple of ply you might not be
>able to see it anymore.
>
>On the other hand, deadly tactics that arise later on are probably due to errors
>you make earlier that in a way set it off.  If you don't see any deadly tactics
>coming within 15 ply then chances are you're doing something right, so perhaps
>you are reducing the chances of them appearing later (I admit this argument is
>far weaker than the previous one.  I'm not even sure it's correct.).
>
>Well one of the arguments has to be wrong.
>
>Comments?

I think that in most of the games between computers at long time control the
reason for the win is positional errors and not tactical errors and when one
side can see the loss it is too late(even one move before seeing the loss).

Searching one ply deeper may cause programs to do less positional errors so the
first argument is not relevant in most of the cases.

The main argument that I see for diminishing return is the fact that part of the
opening lead to draw unless the sides do mistakes that programs avoid at long
time control so at small depth is it not relevant but at big depth the game will
be drawn.

Uri



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.