Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Books with no GM input ..vs.. Books with GM input.

Author: Dann Corbit

Date: 21:24:20 07/01/04

Go up one level in this thread


On July 01, 2004 at 11:27:08, Dana Turnmire wrote:

>On July 01, 2004 at 09:09:55, francis wrote:
>
>>if you get computers to generate their own opening books - without using the
>>human GM's games as inspiration, then how well do they do?
>>
>>do they find most of the main lines?
>
>Try your engine with this test suite.
>
>BOOK OPENINGS TEST (Larry Kaufman)
>
>There are many ways to evaluate chess programs. They can be played against each
>other, played against human opponents, or tested on problem sets. In this
>article I will introduce yet another way to evaluate chess programs. The basic
>idea is to compare the program's choice of move to the choices made by masters
>over the years.
>Here is my procedure: I first selected eight of the most popular openings
>from master play. Then, using a large database of master games, I followed each
>opening down the path seen most frequently at each move in the database. The
>line was terminated when the sample of remaining games dropped below one
>hundred. Next, each move in each line which was played at least 75% of the time
>but was not an obviously forced recapture was presumed to be the best move at
>that point and was marked as a problem. This produced a set of exactly one
>hundred problems from opening play in which there exists a best move in the
>opinion of at least 75% of masters reaching those exact positions in their own
>games.
>The method of testing a program is simple. First, turn off its opening book
>and put it on analyze mode, or else on infinite level and monitor mode, which is
>pretty much the same thing. Then, play through the opening line until a problem
>position is reached, let the program think two minutes and record whether or not
>the correct move is played, enter the correct move and continue to enter moves
>until the next problem in the line is encountered, and repeat the procedure.
>When each opening line is completed, reset for new game and go to the next
>opening line. The score is simply the number of correct moves out of the one
>hundred total.
>The eight openings I selected were the Ruy Lopez, the Sicilian with 2 Nf3 d6,
>the Winawer French, the Caro-Kann with 3 Nc3, the Slav defense to the Queen's
>Gambit, the Nimzoindian defense, the King's Indian defense, and the Gruenfeld
>defense.
>Of course it must be remembered that this test deals only with how well a
>program plays the opening without benefit of an opening book. It may be argued
>that since programs do in fact play with opening books this ability is not
>crucial in actual practice. On the other hand, the ability to find the best
>opening moves without a book is an excelent indication of the overall strategic
>and tactical abilities of the program.
>The lines tested were as follows (moves in parenthesis are positions to be
>tested by the program).
>
>In the Ruy Lopez: 1 e4 e5 2Nf3 Nc6 3 Bb5 (a6) 4 (Ba4) (Nf6) 5 (0-0) Be7 6
>(Re1) (b5) 7 Bb3 0-0 8 (c3) (d6) 9 (h3) Na5 10 (Bc2) (c5) 11 (d4) Qc7 12 (Nbd2)
>cd 13 cd Nc6 14 Nb3 (a5) 15 (Be3) (a4) 16 (Nbd2).
>
>In the Sicilian: 1 e4 c5 2 Nf3 d6 3 (d4) (cd) 4 (Nd4) (Nf6) 5 (Nc3) a6 6 Bg5
>(e6) 7 (f4) Be7 8 (Qf3) (Qc7) 9 (0-0-0) (Nbd7) 10 g4 (b5) 11 (Bf6) (Nf6) 12 (g5)
>Nd7 13 f5 Nc5 14 (f6).
>
>In the Caro Kann: 1 e4 c6 2 d4 d5 3 Nc3 (de4) 4 Ne4 Bf5 5 (Ng3) (Bg6) 6 h4
>(h6) 7 (Nf3) (Nd7) 8 (h5) Bh7 9 (Bd3) (Bxd3) 10 (Qxd3) Qc7 11 (Bd2) e6 12
>(0-0-0) Ngf6 13 Ne4 0-0-0 14 (g3) Ne4.
>
>In the French: 1 e4 e6 2 d4 d5 3 Nc3 Bb4 4 e5 c5 5 (a3) (Bc3) 6 bc3 (Ne7) 7
>Qg4 Qc7 8 (Qg7) Rg8 9 Qh7 cd 10 (Ne2) (Nbc6) 11 (f4) (Bd7) 12 (Qd3) dc 13 Nc3
>(a6).
>
>In the Slav: 1 d4 d5 2 c4 c6 3 Nf3 (Nf6) 4 Nc3 dc4 5 (a4) (Bf5) 6 e3 (e6) 7
>Bc4 (Bb4) 8 (0-0) Nbd7 9 Qe2 0-0 10 e4 (Bg6) 11 (Bd3).
>
>In the King's Indian: 1 d4 Nf6 2 c4 g6 3 Nc3 Bg7 4 (e4) (d6) 5 Nf3 (0-0) 6
>(Be2) (e5) 7 0-0 Nc6 8 (d5) (Ne7) 9 Ne1 (Nd7) 10 Nd3 (f5) 11 (Bd2) Nf6 12 (f3)
>f4 13 c5 (g5) 14 Rc1.
>
>In the Nimzo-Indian: 1 d4 Nf6 2 c4 e6 3 Nc3 Bb4 4 Qc2 0-0 5 a3 (Bc3) 6 (Qc3)
>(b6) 7 (Bg5) (Bb7) 8 f3 h6 9 (Bh4) (d5) 10 (e3) (Nbd7).
>
>In the Grunfeld: 1 d4 Nf6 2 c4 g6 3 Nc3 d5 4 cd Nd5 5 (e4) (Nc3) 6 bc3 (Bg7)
>7 Bc4 c5 8 (Ne2) Nc6 9 (Be3) (0-0) 10 (0-0) Bg4 11 (f3) (Na5) 12 Bd3 (cd) 13
>(cd) (Be6) 14 Rc1 (Ba2) 15 (Qa4) (Be6).

Interesting set of problems.  I am going to run each of the 100 distinct
positions above (regardless of selection via parenthesis) for one hour each
using Shredder 7.04 UCI on a 64 bit AMD machine to see how well the moves are
chosen.

I wonder if any novelties will be discovered.




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.