Author: Ed Schröder
Date: 11:16:34 10/08/97
Go up one level in this thread
>Posted by Robert Hyatt on October 08, 1997 at 11:15:57: >In Reply to: CCC-I results posted by Walter Ravenek on October 08, 1997 at >03:39:14: >===================================================================== >CCC-I 1 2 3 4 5 6 7 8 9 10 11 > bm bm bm bm bm bm bm bm bm am bm > Rxe6 Bxg7 Ng4 Nxf7 Rc3 Qxh6 Rh3+ Bd4+ Rxh7 Qxf3 b3 >===================================================================== > 7 DarkThought 0:00 0:12 2:31 ---- ---- 4:20 0:14 ?:?? ?:?? 4:42 0:21 > 8 Rebel9 2:09 0:18 0:42 1:07 ---- ---- 0:44 ---- 0:48 1:09 0:07 > 7 Fritz5 0:04 0:14 0:08 1:50 ---- ---- 2:11 ---- ---- 0:35 0:05 > 5 Arthur 0:09 0:09 0:57 ---- ---- ---- 0:12 ---- ---- ---- 0:06 > 4 Comet 0:02 0:02 3:18 ---- ---- ---- ---- ---- ---- ?:?? 4:28 > 3 CSTal ---- 0:02 0:14 ---- ---- ---- ---- 0:07 ---- ---- ---- >===================================================================== > 7 Crafty 13.4 4:41 0:15 1:50 ---- ---- 3:50 0:11 ---- ---- 0:00 0:29 >couple of comments: position 9: I don't see the need for the >requirement to "see the mate." For example, here is what Crafty sees: > 11.229 Rxh7+ Kxh7 Qh5+ Kg8 Rxg7+ Kxg7 Bh6+ Kh7 Bg5+ Kg7 Qh6+ Kf7 > Qf6+ Kg8 Qg6+ Kh8 Bf6+ Rxf6 exf6 Qxe1+ Kxe1 Nc2+ Kf1 Ne3+ > fxe3 Rd7 Qe8+ Kh7 Qxd7+ Ne7 Qxe7+ Kg6 Qxb7 Kxf6 >That is pretty compelling evidence that it understands why Rh7+ is a >good move. It is not hard to make a program find such mates. But this >doesn't have a lot to do with how well they pay chess. Every program plays Rxh7 very quickly so no big deal. The key point of this position is the quiet move Bf6! after a lot of checks which leads to the mate. Many programs will prune Bf6. >For a comparison, take position 1. I doubt *any* of those programs saw >the tactical advantage or Re6 in a few seconds. But other moves are >-1.x, and Rxe6 is -1.x, so this is a positional judgement. Later, >they will find (maybe) that this is actually suddenly "even" in >material, and not -1. >Why count this right for positional reasons but then count the deep >mates wrong if the mate is not found? Seems inconsistent for a >tactical test. I picked some nice positions from my database because CCC suffered from topics. >Either require that a program get a specific evaluation/PV for >each position, or else count 'em right if they play the right >move, period. But let's not mix them up... Okay, you post CCC-IV :)) >These were run on my P6/200, but with a small (12m) hash table. I don't >believe a bigger one would affect results. >here is the output for the position where it is necessary to avoid Qxh6: > 12 0:55 0.000 Qxh6 Bxb4+ Kxb4 Nh5 Kc3 a5 h4 Ng3 h5 > Nf5 hxg6 Nxh6 gxh6 hxg6 Impressed... >Position 6 (Rc3) takes 5:30 on my P6/200 to find. On a PII/266 it gets >this one in under 5 minutes, but I left the times for the P6/200 as this >is a "standard" machine. On a 500mhz alpha, these are all divided by >roughly 2, based on the alpha I have tested on (actual DEC 21164/500 >workstation.) >position 10 I don't like, because Crafty never liked Qxf3 at any point, >and stuck with Nc4 from ply 1 thru 11. Selective search problem: Qxf3 Bxf3 Bxf3 Ng3 Ne4! In previous versions of Rebel Ne4 was pruned too. - Ed -
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.