Author: Bob Durrett
Date: 16:40:55 12/03/02
Go up one level in this thread
On December 03, 2002 at 19:07:10, Pat King wrote: >On December 03, 2002 at 15:20:06, Bob Durrett wrote: > >>On December 03, 2002 at 13:09:42, Pat King wrote: >> >><snip> >> >>As a practical matter, how useful are >>>those less-than scores? >>>[snip] >>> >>>Pat >> >>In my experience, the "second best" move and sometimes third are *occasionally* >>the best moves!!!! >> >Of course, else your program is playing perfect chess, and that ain't gonna >happen. The question was, rephrased, is it worth knowing that some variation is >AT LEAST such and such worse than the PV, since the method I proposed wouldn't >give an exact score for other than the best move. Also, given the uncertainty in >the scores, is that sorted list of alternative moves likely to be in the right >order? It would be good if a chess programmer could answer this question. I can only speak as an interested user whose programming experience is obsolete. "AT LEAST such and such worse than the PV" implies to me a recognition that sometimes a position has more than one good move such that the difference in merit of the good moves may be very small. In other words, two or more moves may have "about" the same value. But my perception is that current chess engines select one, even if the others are exactly as good. If the programmer were to allow moves that were almost as good as the move chosen, then a tree of "best move sequences" might be constructed. Compared to the tree of all possible continuations, this would be a very small tree even if it had dozens of branches. Generally, all the "tricks" [like alpha/beta] are to minimize the number of nodes evaluated except for the nodes in the one and only "best" move sequence. On the other hand, I would expect that creation of the "good moves" tree suggested above would not carry a huge penalty. It is not clear to me that a suitable search algorithm could not be produced to construct that tree. That's why I asked about a GENERALIZED alpha/beta. It certainly goes against ALL of my intuition to believe that the chess engine would have to essentially "start over" for each alternative move as Bob Hyatt seems to suggest!!!!!!! After such a tree of good move sequences were constructed, it would be a simple matter to evaluate the leaf nodes to see which was best, which was second, etceteras. Then that information could be displayed. Current engines try to construct a tree, which is nothing but one long crooked [but "best"] branch. Why is it so hard to construct a small barren tree with several branches???? Surely that is NOT so difficult! I guess the stock reply might be: "Well, if you think it's so easy then why don't you just do it." Obviously, a non-programmer cannot do that. Bob D.
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.