Author: Dann Corbit
Date: 09:38:39 07/22/99
Go up one level in this thread
On July 22, 1999 at 10:13:51, Albert Silver wrote: [snip] >If you are just looking for ready to solve exercises, and aren't too picky on >difficulty, then there is no need to buy a book. There are a number of tactical >suites, with excellent positions, designed to test a computer program's prowess. >On Dann Corbit's FTP site you can find a number, including an ECM suite with >more than 7000 positions (or so I understand) to resolve. This file: ftp://38.168.214.175/pub/Public_CAP_Results/Apocalypse/Apocalypse.epd contains 7,756 EPD positions from common test suites. The interesting thing about the file is that each position has been analyzed for at least 12 minutes of PII 300 MHz equivalent time with a computer program. It would take more than 2 months of computer time at 24 hours per day to produce it by yourself. There are a number of clear and obvious mistakes in the results (e.g. it does not matter what move you make, if the opponent plays correctly you will be checkmated) and there are also some positions that are correct and the computers fail to solve. In any case, there are 5701 problems with verifiably correct results (for the time being -- deeper analysis might prove that they are wrong!). Of the two thousand that disagree, careful study would be needed to see which of them are bugs in the test suites and which of them are weaknesses in the programs. I believe that *most* of them are bugs in the test suites, as many are obvious. Here is a humorous one: 4r2k/p2Q3p/2p1r1pR/4NpR1/3P4/2P1PP1b/P4q1P/6RK w - - The test suite position misses a mate in one! I suggest that the 5701 problems which are verified can be used to make more reliable test suites than are currently in existance (though I am fairly certain that deeper analysis will change some that do not end in mate).
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.