Author: Robert Hyatt
Date: 06:35:00 02/12/98
Go up one level in this thread
On February 12, 1998 at 03:24:45, Baier Helfried wrote: >On February 11, 1998 at 08:57:04, Robert Hyatt wrote: > >>On February 11, 1998 at 02:52:29, Bernhard Bauer wrote: >> >>>On February 10, 1998 at 12:09:49, Bruce Moreland wrote: >>> >>>>Cool bug. If you want to see other bugs of this type, try setting up >>>>positions where each side has only one minor piece, but there is a mate >>>>in one. >>>> >>>>For instance something like >>>> >>>>5Bbk/8/7K/8/8/8/8/8 w - - 0 1 >>>> >>> >>>Crafty gives: >>> >>> White(1): Bg7# >>> time used: 0.18 >>> >>> Crafty offers a draw. >>> >>>So after mating Crafty offers a draw. >>> >>>>When I tried this a few years ago I found that several programs called >>>>this a draw. Maybe they still do. >>>> >>> >>>Looks like you are right. >>> >>>>Regarding 5-piece endgame databases, I have a whole bunch of them, but I >>>>don't actually use very many on ICC, since the Alpha doesn't have enough >>>>disk. >>>> >>>>One of the ones I put on the Alpha, almost as a lark, was KP vs KNN, and >>>>it has come in handy twice, which was twice more than I expected. >>>> >>> >>>I think it's a good idea to do things in a proper way. Crippling down a >>>program for speed reasons may be a bad idea and may give the customers >>>a bad feeling. It would be better to insert a line of code like: >>> >> >>you are assuming facts not in evidence. The reason *I* do what I did is >>to win games that are winnable. IE imagine a position where there are 7 >>pieces on the board. White has a king, two knights and a pawn. Black >>has >>a king knight and bishop. Crafty finds a way to win one of black's >>pieces for a pawn, and concludes "hey, I have two pieces vs one piece >>for my opponent, >>that's +3, so I'm going for it since the current score is only + 1. But >>the resulting position ends up a draw. Since it is impossible to >>determine >>which are winnable and which are drawable by search, I have to either >>say >>"draw or +3". DRAW is right 99.99999 percent of the time. >> >>It isn't a speed issue at all. IE imagine intentionally heading into >>these >>+3 types of position tactically, only to find out 50 moves later there >>was >>no way to win. It happened many times until I fixed this. I'd rather >>win >>100 and draw 1, than draw 100 and win 1. >> >>>XXX will not handle this position properly due to speed reasons. STOP. >>> >>>Here are 2 positions from V. Albillo's home page that cause difficulties >>>to crafty. >>> >>>8/1k1K4/8/8/1pN5/1N6/8/8 w - - bm Nc5+; id "VA19: 6mate"; >>>8/8/8/5K2/7k/7p/7N/7N w - - bm Nf2; id "VA20: 9mate"; >>> >>>Now I know why. >>> >>>Regards >>>Bernhard >>>>bruce > > >as far as I know KQx-k (x is Q,R,...) positions are all 1-0 positions >except for a stalemate. Does Crafty use this tablebase files or can I >delete it? > > >Regards > Actually, almost all of the 4 piece files are not needed. KQQK is obviously one. But deleting them causes one embarassing problem: If you reach a KQPK ending, it will see "mate in N" from that tablebase. But once it pushes the pawn to the 7th, it might not promote it if the KQQK tablebase is not available. It is possible it can't see deeply enough to see the mate in a blitz game (should be no problem in normal time controls). What will happen is that it will check like mad with the queen, which keeps it in the tablebase and keeps it seeing "mate in N". It will do this until either a repetition threat or 50-move rule forces it to promote to avoid the draw score. So it would probably win, but not "prettily". And it is possible that by the time it checks for 50 moves, it can't get back to the right position where it can force the promotion without repeating for a third time. bottom line: keep 'em all, only 250mb. Disk is cheap. :)
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.