Author: Sune Fischer
Date: 17:08:04 02/06/02
Go up one level in this thread
On February 06, 2002 at 19:14:43, Dann Corbit wrote: >On February 06, 2002 at 18:54:28, Sune Fischer wrote: >[snip] >>I'm not saying that it would be easy to get all those test positions, but if you >>had them the test would be nice. >>I suppose one could do a 14 ply search as an approximation to the best move, it >>wouldn't change the distribution all that much I think. > >The problem (as I see it) of running test suites with only ce answers is that >the chess engine will [not infrequently] find a better solution or another >solution which is marked as "wrong" by the program looking for an answer. > >If you have not traced all the way to checkmate, then the best answer is not >certain. Listen, the idea is very simple :) A) We have X positions where we _know_ the best move. (of cause we can discuss how to get such a set of positions, but that wasn't really the point here, it is an *assumtion*, a mind/thought experiment if you will :) B) Run the engine brute force limitet to n ply. This will produce the previous posted distribution. C) How to interpret this distribution: The 1-ply searcher will often pick the right move, but for the *wrong* reasons in the sense that it cannot see the deeper tacticks. My point is, that this does not matter, because it will still play the *right* move! Pure luck, but that is okay. So the distribution should IMO reflect the chances of *being lucky*. The deeper you search, the more often you will pick the right move. Remember that even a 12 ply search is still making a guess at the right move just as the 1 ply searcher did, but more often the 12-ply'er guesses right. I haven't got any idea as to how fast the distribution converges or what kind of distribution we might be talking about, but interesting to find out I think. -S.
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.