Author: Steven Edwards
Date: 10:15:20 09/19/03
I've been testing my toolkit's PGN parser and its opening book compiler using the 27,202 games in Crafty's gm2600.pgn file. Running on an 800 MHz notebook, the games are processed at a rate of about 25 per second including generation of the output book file. The book compiler was told to include positions from up to 48 ply deep in each game and this produced a book with 735,587 positions. From the initial array, the book moves were determined to be: (M = move; B = win_rate + draw_rate/2; C = wins losses draws; all numbers are from the viewpoint of the moving color) [M:Nc3 B:0.583333 C:[2 1 3]] [M:Nf3 B:0.554582 C:[800 451 1946]] [M:a3 B:0 C:[0 1 0]] [M:b3 B:0.475 C:[11 13 16]] [M:b4 B:0.375 C:[1 3 4]] [M:c3 B:0.3 C:[1 3 1]] [M:c4 B:0.556285 C:[789 463 1644]] [M:d3 B:0 C:[0 1 0]] [M:d4 B:0.551783 C:[3228 2014 6480]] [M:e3 B:0 C:[0 1 0]] [M:e4 B:0.562561 C:[2725 1579 4855]] [M:f4 B:0.441176 C:[4 6 7]] [M:g3 B:0.506711 C:[38 36 75]] Note that the Dunst Opening (1. Nc3) has the best expectation, although it was only played six times total. After 1. e4, the book moves are: [M:Nc6 B:0.333333 C:[3 12 12]] [M:Nf6 B:0.403509 C:[22 44 48]] [M:a6 B:1 C:[1 0 0]] [M:b6 B:0.333333 C:[0 1 2]] [M:c5 B:0.444473 C:[741 1178 2016]] [M:c6 B:0.437358 C:[140 250 488]] [M:d5 B:0.358108 C:[16 37 21]] [M:d6 B:0.401786 C:[53 119 164]] [M:e5 B:0.437899 C:[361 653 1337]] [M:e6 B:0.436968 C:[213 376 704]] [M:g6 B:0.411565 C:[29 55 63]] A few more changes to the toolkit and then it will tackle the 1,511,738 games in Crafty's enormous.pgn game file.
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.