Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Will next version of Junior have underpromotion code?

Author: blass uri

Date: 00:23:30 06/08/99

Go up one level in this thread



On June 08, 1999 at 00:56:53, Dave Gomboc wrote:

>On June 07, 1999 at 20:29:58, Dann Corbit wrote:
>
>>On June 07, 1999 at 20:20:25, Dave Gomboc wrote:
>>[snip]
>>>At least for the present, it is only an opinion.
>>>
>>>Minimax has a way of removing the horrible error in end-point evaluations.  It's
>>>pretty rare that missing an underpromotion will actually hurt you.
>>In the sample I examined, about 7% of the time it scores lower than always
>>queening.  I can send a list of such EPD's to anyone who would like a copy.
>
>Your sample includes only games in which promotion actually occurs, and the 7%
>includes many positions where promoting to a rook might be slightly better, but
>if the person playing had overlooked the possibility to underpromote, they could
>have still won the game my promoting to a queen, either with the same move
>sequence, or a different one.  The important figure would represent how often
>the rough evaluation of a position (e.g. clearly winning) depends upon a
>promotion, and how many of these cases there are in which a queen will never
>work.
>
>When you look for this, now you are counting practically every winning ending --
>and many winning middlegames -- with pawns on the board that gets resigned.  I
>do not have a statistic to back up my thoughts on this matter, but I am
>extremely confident that the rate of "necessary underpromotions" will be far
>less than 7%.  0.0007% would be more like it.
>
>Dave

I am sure that it happens more than 0.0007%

I did not do statistics  about it but I think that Junior lose points because of
not seeing underpromotions  in  more than 1 game out of 1000 but clearly less
than 1 game out of 100.

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.