Author: James Robertson
Date: 13:01:32 03/23/99
Go up one level in this thread
On March 23, 1999 at 15:35:59, Dann Corbit wrote:
>I see a few problems with current opening books -- not for people but for
>computers only.
>
>Quite likely, a GM will play an opening for strategic reasons. He may even toss
>a piece to gain mobility or something of that nature. For a human GM, such a
>choice is sound. For computers, most of the time it will not be sound. The
>reason is that the computers do not see 20 moves down the road at all. So the
>sacrifice just reduces their ability to take more tactical snacks on the way to
>the accidental giant meal called checkmate. When the computer falls out of
>book, it is a poor, bewildered bunny. It has the right position on the board,
>but the payoff is so distant, it does not know how to benefit.
>
>So there are proven sound openings that are simply not good for computers.
>We'll call them [TOODEEPS] {or maybe [TOODIEPS] after Vincent. ;-)}
>
>There are also buggy openings with deeply hidden tactical snares that people do
>not tend to see. Computers are always looking for the poor baby sheep that
>wanders from the herd just one false step. We'll call them [TRAPDOORS].
>
>So we have these trap doors and too deep openings collected into our chess
>opening books. We are skipping happily along in our opening book and smiling
>about all the time we are saving, and then a door gets slammed in our face.
>
>So what to do about it?
>First we have to identify them. One way to do that is play a lot of games
>against very good opposition. Unfortunately, that way is not very systematic
>and it also means that you lose those games in the process.
>
>Another way is a systematic scanning using some project like C.A.P.
>
>I imagine that there are other ways. Stangely, I don't think having a GM look
>over the openings is a solution. The strategically sound things he lets pass
>may stumble a computer.
>
>Once we found them, what do we do about it?
>One thing we could do is just remove them.
>Another thing is to include the correct trajectory as supplied by a computer on
>long time controls or a GM.
>For the trap doors, we might search for a refutation at long time controls. If
>GM's do play them again and again, a refutation may exist.
>
>Opinions? Refutations? Flabberghasted, shrieks of annoyance?
This reminds me of the poetry on your ftp site (I enjoyed both). :)
I am on the lookout for pgn files between top GM's to make my program's book out
of. It can't be too big (I have to send it over email to people) so I can't
download say, 300,000 Fried Liver Attack games. I downloaded Hotties.pgn, and it
was nice. Do you have any other pgn files like it?
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.