Computer Chess Club Archives


Search

Terms

Messages

Subject: More technical(?) explanation

Author: Gian-Carlo Pascutto

Date: 14:08:47 03/31/00

Go up one level in this thread


On March 31, 2000 at 12:53:43, Stephen Ham wrote:

>Would somebody here please provide a detailed explanation regarding whether >this claim is correct and why?
>
>The chess engines are on settings recommended by ChessBase USA as their optimal
>settings for this event. My extremly limited understanding is that displaying
>the top 3 choices does indeed affect the chess engines, but it causes them to
>spend more time on what it believes to be the 3 best moves. As such, this >sounds like an enhancement to me. Given that the chess engines are allowed >about 24 hours calculation time on weekdays and are searching to 16-18 ply, I >can't imagine that this weakens them in any way.

The way current state-of-the-art programs work, the program guesses the best
move, searches through it, and establishes a score for it.

Now, instead of doing the same for all other moves, they merely try to 'prove'
that the move they guessed was best is indeed better than all other moves.

Proving a move is better or worse is far faster than computing a real score
for it.

The result is either: this move is worse by at least x, or at least better
by x. The 'x' values are not necessarily the real values of the moves.

If the move was better than the guess, the program will need to find the
real value for that move, because the initally guessed best move was wrong.
If the move was worse, the guess was right and no further work needs to be
done.

Note that the program will generally only know the true score of the best
move. It will also only know the best sequence of moves (the principal
variation) for the best move, because the 'proving' search often has
principal variations that are cut short.

So to be able to show the score and full sequence of moves for anything
more than only the best move, the program will have to use the full search
for those moves instead of the 'proving' search. This will slow down the
program, causing it to play at a lower lever.

Note that the computer does spend more time on the 3 best moves, but that
this will not have any influence in selecting the move to play...it only plays
the best move after all, so the correct score for the other moves doesn't
matter, as long as we know its less.

>Please advise. We sincerely want the chess engines to play at peak performance.
>I also think we all learn more about computers in Correspondence Chess when we
>can see their 2-3 choices and understand what it caclculates and DOESN'T
>calculate in certain positions. Often, it seems they make the correct choices
>for the wrong reasons.

It's a tough call. Most recent experiments indicate that chessprograms have
a quite constant rate of changing their mind, independent of the search
depth.(*) So even if the engine is already calculating 18 ply, 19 ply will still
help a bit.

But to see one ply deeper, most engines need to calculate about 2-3 times
as long as they had to for the last ply. Because of the very long thinking
times the hashtables of the programs fill up, which may cause them to require
another 2 times as much time for seeing one ply deeper.

The extra time required to calculate 3 scores and variations, which I guess
is about 2-3 times the normal time. To see one ply deeper, they require 4-6
times as more time.

It may be more worthwhile for you to see the 3 best moves and gain a better
understanding about the computers that it would be to have them play another,
possibly better move 10% of the time.

(*) for large enough values

Disclaimer: most of the numbers I used are estimates, so they may be a bit off

--
GCP



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.