Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Computer Endgame Advice Requested

Author: Ralf Elvsén

Date: 15:49:15 10/05/00

Go up one level in this thread


On October 05, 2000 at 16:47:03, Stephen Ham wrote:

>Dear Ralf,
>
>I think this is meantioned at the website, but agree that some things are not
>easy to find when you want them. So I'll give you the answer from my point of
>view.
>
>My answer is this: I want the computer chess engines to be able to perform at
>their optimal levels. Franklin Campbell and Don Maddox (ChessBase USA) found
>that the computer chess engines teamed with fast Pentium III computers would
>reach a certain ply depth after 17-20 hours. To reach the next ply depth may
>require nearly twice that time and would thus tie up the computer operator's
>computer for 100% of the day, so they would never be able to use their computer.
>
>The solution was that Franklin Campbell of the ICCF agreed that he would start
>the chess engine when he was done with his computer in the evening. He would
>leave it running over-night and also while he was at work the next day. He would
>only stop the computer when he got home from work and then needed to use it to
>read/post e-mails etc. The primary choice of the chess engine was the move
>played. Then the process cycle would begin again for the next move.
>
>The result means two things to me: 1) the chess engine thus had 17-20 hours of
>calculation time, and 2) I had a new move waiting for me every day, so I had to
>respond quicly myself, in order not to get too far behind.
>
>Ralf, I hope this answers you question to your satisfaction. Thanks for your
>interest.
>
>Stephen

As I said, I have a follow-up question, or a comment rather, that I have
been thinking about in connection to your games. Sometimes
a program "fails low". What it basically means is that it realizes
that the move it has considered to play for a long time now suddenly
doesn't look so good anymore. Say, e.g. it has searched for 17 hours
and finished depth 20 and found that 1. Bxh7 is a great move. Then
it starts at depth 21 and soon realizes that the move isn't as good
as it thought. What a program does in this situation in a normal game
is to "resolve" the situation: just how bad is the move, and if it
stinks, it allocates more time than normal to find a move that is
good enough (just like a human would do). But if it in analysis
mode and used for a correspondence type game and is interupted,
it is out of control.

The ideal situation would be that a red warning light would be on:
"I'm failing low! I wouldn't play this as yet unresolved move."
So it is possible that an unknowing operator could force a program
to play a move which would not have been played in a normal game.

I have been thinking about this for a while and have a suggestion for
a solution, but I would like to get some more imput from others:
is this a problem or just my imagination?

Ralf



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.