Author: Dann Corbit
Date: 09:53:32 02/10/99
Go up one level in this thread
On February 10, 1999 at 04:25:00, Bruce Moreland wrote: >On February 09, 1999 at 22:58:41, Dann Corbit wrote: > >>Mike Byrne over on r.g.c.c. just found two more (he found a total of five!) >>negative checkmates in the first twenty rows from the rockpile. No wonder they >>are so hard to solve -- they have no solution. ;-) >> >>It is rather amazing how many of these problems have less than ideal answers. >> >>For quite some time people have been basing the strength of their computer >>programs by running against batteries of these tests, as one measure of ELO. >> >>Maybe we need to look over the tests very carefully. It could be that an 80% >>score is 100% right and a 100% score is 20% wrong! >> >>Maybe you have thought -- darn it! My program keeps getting the wrong answer >>on this problem! >> >>OTOH, maybe your program was right all along. > >I don't understand what the suite is supposed to represent. Some suites are >presented as "hard problems", and I thought that is what this was. But it seems >like a lot of them are more like "broken problems", and all running the thing >for an half-hour or so per position does is waste time. Don't you think that finding out which positions are broken in the commonly available test suites is valuable? I have found errors in every test suite. These test suites are used to judge the strength of your program. The positions that I posted were classified as "unsolved" because (up to that point) the results were unknown. By that I mean, I had already run them for a very long time, and the proposed move was not chosen, and there is no result of mate for either side. I think that finding the right answers or discovering what is wrong with the test problems is a lot more interesting than just finding an answer we already know. But I have an unusual bent, anyway.
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.