Author: Bruce Moreland
Date: 13:07:25 11/25/98
Go up one level in this thread
On November 25, 1998 at 13:57:44, Howard Exner wrote: [discussing Q vs P] >In regards to tables is there such a thing as a reduced endgame base with just >the most common scenarios? Or are the efficient to a point where there is >no performance hit during actual play? The 4-man tables don't seem to cause much problem, the 5-man tables will over-commit ram no matter what you do, so these cause more problems. I doubt it's possible to distill them to a point where you just have the good stuff and you don't thrash. Q vs P might be doable with straight heuristic knowledge. I haven't tried to do it yet. The thing is that there are cases where you have the weaker king and the pawn in a drawn configuration, but there are a few thousand permutations of where the other pieces can be, and there are several drawn configurations for the K+P as well, and I think it would be hard to actually generate a simple and correct heuristic. In the case of the wrong rook pawn, it's very easy to generate a perfect heuristic. If the king is in g7..h8 and the bishop is light-square, and the pawn or pawns are on the h-file, it's a draw, period, and there are no exception cases. It's harder to do this with Q vs P I think. >>The KBP vs KP is also a nice trick. You can move the ram a file to the right >>and that's another good one, > >I'll have to remember that one, I wasn't aware of that draw. Yes. Rare but practical, it happened to me. I also had the same color bishop thing happen to me. I had a game with Crafty where mine promoted a pawn to a bishop, when it could have easily promoted it to a queen, which Crafty would have taken, of course. Mine thought it would also take the bishop, but it didn't, and my thing played like 80 more moves with same colored bishops. It wasn't a pure case, there was other material, but it got blockaded or something. >I believe it is a good thing for chess programs to add endgame knowledge >even when the probability is rare of the position showing up. I think it >makes for a more complete product. Beyond a doubt. This stuff comes up a lot more in practice than people think, too. Once you have any of this stuff installed, you start seeing your program playing for the drawn cases when it's losing. Crafty used to have more sophisticated wrong rook pawn handling than mine did, and it was a major annoyance. Now I think mine might be better than Bob's, so he'd better watch out. Even if it only happens once in a hundred games it's still a measurable Elo delta, and it's a very embarassing game to have to sit through. bruce
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.