Author: Bruce Moreland
Date: 11:20:33 09/27/99
Go up one level in this thread
On September 27, 1999 at 13:58:55, Christophe Theron wrote: >I don't think so. The LCT test from Frederic Louguet was great when it was >published, and it gave very good results at this time (in 1995). > >But if I wanted to, Tiger could easily get 200 more ELO points on this test >suite just by changing 2 lines of code. > >The "combinational" part of it is only about mate combinations. Just extending >more on checks and mate threats does the trick. > >The LCT is unfortunately not a good test anymore... If you don't want to be >completely off, you'd better not trust it. Don't these tests *have* to give good results at the time they are released, because aren't they calibrated such that they produce results that relate to order on the SSDF list? I mean, you can make a function that returns a rating based upon the order of letters in the program's name, and this can be an absolutely perfect test at the time. People look at these tests and go, yow, what a great test, and in fact the test has been tuned to produce these results. One question is if accurate results are produced when the hardware is sped up, but even this can be estimated and tuned in to the test. You just make your scoring function return a rating that is higher by the amount the SSDF guys have predicted it should be higher, if you double hardware speed, and if you guess right, once again you have a great test, but you could still do the same thing with a function that returns a rating based upon the order of letters in the program's name, plus a little bit of math involving the processor speed. You can talk about how cool the positions are, but for the initial suite of programs tested, the positions could be anything. That solving the problems faster does have some relation with program strength in the typical case, is the only thing that saves these suites from being complete nonsense. You can test another program and get a number that's in line with the other programs, because they all play the same game in approximately the same way. bruce
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.