Author: Shep
Date: 07:29:25 02/07/00
Go up one level in this thread
On February 07, 2000 at 09:40:46, Djordje Vidanovic wrote: >I am almost positive that neither of the two were tuned for _any_ particular >test. Both Morsch and Donninger, just like Christian Theron (cf. his testimony >here, on ICC) and other top programmers, believe in the overall playing strength >of their programs, and do not care much about how well it does on a test. >Naturally, tests do serve as indicators, but indicators only. There is a fine line: on one hand, there is deliberate tuning or "cooking" in order to get #1 in these test suites. I don't think any top program is programmed this way. On the other hand, programmers always need an indication if their newest settings have improved the program (without playing 100s of test games), so it is natural to turn to one or more test suites. And of course the programmer asks himself "why doesn't it find this position?", comes up with an idea, implements it and bingo! the position is found. And of course they get worried if, after a new algorithm, "oh, why doesn't solve it this position anymore?", so they try to 'fix' the hole. Ultimately, this leads to some sort of "test suite tuning", if only subconsciously. --- Shep
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.