Author: Christophe Theron
Date: 03:40:39 01/04/98
Go up one level in this thread
On January 03, 1998 at 19:20:56, Bruce Moreland wrote:
>
>On January 03, 1998 at 17:16:50, Moritz Berger wrote:
>
>>>processing with a setting of 2 plies (meaning it tests for an extra ply
>>>to verify the program doesn't change it's mind)
>>
>>I used the same settings.
>
>This isn't a good way to do these tests, since this will result in false
>quick solutions sometimes.
>
>To get an idea of how often this can produce an incorrect result, I just
>looked over an ECM dump, my program missed 267 problems (20 seconds per
>problem on a P6/200), but in 36 of the problems that it missed, it found
>and held the correct answer for an extra ply, then changed its mind
>later.
>
>A few false correct answers (or even one) in something like LCT2 can
>change that pseudo-Elo result a lot.
In fact not.
The LCT2 is not just a set of position. It also has an associated
protocol that prevents problems like these.
You have to run the program on each position for 10 minutes and record
the LAST time the program changed for the right move.
For example:
in POS01, the key move is d5. If your program finds d5 in 1s, then
changes its mind and plays o-o, then goes back on d5 in 256 seconds and
would still play d5 at the 600th second (10mn), you record 256s for
POS01. This has nothing to do with plies, depth...
I will post the complete LCT2 protocol on CCC very soon. Frederic
Louguet, author of the test, mailed me the complete english text. I'll
also try to explain the motivations behind this test, because I'm
expecting many people to begin endless angry threads about it. :)
Christophe
This page took 0.01 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.