Author: Robert Hyatt
Date: 10:39:17 05/15/99
Go up one level in this thread
On May 15, 1999 at 10:54:23, Francis Monkman wrote: > >On May 15, 1999 at 10:33:47, Robert Hyatt wrote: > >>I consider a program to solve a tactical 'win' when (a) they produce a PV that >>shows that they understood why the move was important and (b) the score shows >>that the program understood how good (or bad) the move was. But just picking >>the "right" move is something that can happen for lots of reasons, totally >>unrelated to playing strength. > >I think for a piece sacrifice to be reading above zero (in the time it takes >LGG) means >something -- also, that although it was not correct about the best line after >about 7-ply, it certainly 'got the drift' -- coincidentally, very much in the >manner being discussed just prior >to this "The reason why etc". Surely it must be clear that NO program is going >to throw away a knight 'for the wrong reason'? > >Francis that misses the point. With an eval of +.2, it is _not_ throwing away anything at all. The resulting position is, in that program's opinion, 'roughly equal'. IE if it had a PV that showed a knight being lost, and no material being won to restore the material balance, and _still_ had an evaluation of +.23, then either its evaluation is grossly wrong, or it is grossly right. A simple example is win at chess #2. Crafty gets that right at 3-4 plies, because it has evaluation code that carefully assesses a pair of connected passed pawns on the 6th against a rook and king, and if it decides that the king is far enough away, it knows that the lone rook can't stop one from promoting, and so it playx Rxb2 instantly. If you let it search for 10-15 seconds, it will discover that Rxb2 wins tactically as well because it can actually see the pawn promoting (after all the checks by white end). Both solutions are correct, because one is for the right positional idea (it sees rxb2, Rxb2 and can certainly say -4 for that. But then it looks further and positionally decides that the pawns are worth more than the -4, and hence it 'chooses' the right more for the right reason. That was my point. IE you might take crafty, search that same position and compare the scores for the move you like (rd1) and for the move crafty likes (h5) and see how close the scores are. If they are 'close' (and I'll bet they are) then crafty is simply using king-side safety considerations to choose a move that it thinks is slightly better. It may be right or wrong. But I would only consider it wrong if some program can actually prove that Rd1 is a winning move, not just a legal move that doesn't lose material...
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.