Author: David Dahlem
Date: 15:27:55 09/05/04
Go up one level in this thread
On September 05, 2004 at 17:54:25, Alessandro Scotti wrote: >First, thanks to all who sent suggestions for playing the KBNK endgame: they >have been very helpful! Today I could finally do some tests and the following >has been working quite well so far: > > // bk=black king square, wk=white king square, wn=white knight square > > int res = materialScore; > > res += 5000; // Bonus for won position > > // note: evaluation scaffolded for light squared bishop > int distance_from_bishop_corner = imin( > KingDistance[bk][H1], > KingDistance[bk][A8] ); > > int b_distance_from_center = imin( > KingDistance[bk][D4], > KingDistance[bk][E4], > KingDistance[bk][D5], > KingDistance[bk][E5] ); > > int w_distance_from_center = imin( > KingDistance[wk][D4], > KingDistance[wk][E4], > KingDistance[wk][D5], > KingDistance[wk][E5] ); > > int distance_of_opposition = Distance[bk][wk]; // Taxicab distance > > int distance_from_knight = KingDistance[bk][wn]; > > res -= 50 * distance_from_bishop_corner; > > res -= 5 * distance_of_opposition; > > res -= 5 * distance_from_knight; > > // centralization: make sure white king is closer to the > // center than the black king > if( b_distance_from_center <= w_distance_from_center ) > res -= 100; > >In the test position I posted earlier (black king centralized and white king in >the corner), I was getting very long sequences, sometimes long enough to make >the position a draw by the 50 moves rule. When I added the last "centralization" >factor, I got the mate in 30-35 moves on average. >I must also add that pruning doesn't work well in this kind of position for me, >so I had to add a guard or two. Without them, I have often seen my engine >playing moves that do not make any real progress towards the mate. Does this help your engine when it plays the losing side of KBNK endgames? I think it's equally important to avoid the mate if possible. :-) Regards Dave
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.