Author: Robert Hyatt
Date: 11:41:24 11/28/00
Go up one level in this thread
On November 28, 2000 at 13:41:46, Enrique Irazoqui wrote: >On November 28, 2000 at 13:19:11, Ed Schröder wrote: > >>On November 28, 2000 at 13:00:02, Dann Corbit wrote: >> >>>A number of authors have stated in this forum that they use the WAC and other >>>test suites to tune their programs. This may or may not make them play better, >>>as you know. Consider that the Rebel settings for solving positions and the >>>Rebel settings for playing the strongest chess are different. >>> >>>Therefore, to tune purely to solve test suites is probably not the best way to >>>create a strong playing program (though it does produce decent chess). >> >>Using wellknown test positions to test if your program changes do >>a better job is something else than Bob said: >> >>quote: >> >> I have seen (a) programs tuned to choose the right move to improve >> their test result scores artificially; >> >>end quote >> >>Note the word artificially which implies cheating. >> >>This of course may be the case but then I would like to see it >>supported by examples. >> >>Ed > >There have been examples of programs cooking test positions. And once, on ICC, a >programmer told me in public that if I would publish my own test suite he would >cook it on the spot. So... > >Enrique Crafty is actually pretty notorious for this "on its own." If you forget to turn off position learning, you will be _amazed_ what it will solve after you run a suite a few times. :) Screws up my debugging all the time. In fact, my development directory .craftyrc file has "learn=0" to stop that nonsense. With this program, cooking would be trivial. And I would be smart enough to cook each position four times, using the "flip/flop" commands in crafty, so that you could reverse colors, mirror the board across the a-h files, and _still get the right answer. :) Of course, adding one unimportant pawn would wreck it, a good way to catch this stuff.
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.