Excessive pruning by SF??

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

whereagles
Posts: 565
Joined: Thu Nov 13, 2014 12:03 pm

Excessive pruning by SF??

Post by whereagles »

From Keres-Hort (1972)

5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - - 23 57

The puzzle solution is Rxh8!

Usually Stockfish finds these things in like 0.1 sec, but this time after 10 mins it's still stuck on Bc5 for +7.70 45/81 (5-men syzygy, 2 cores @2.67 MHz, SF6 POPCNT).

After I force it to Rxh8, it goes +123.4 in like 0.1 sec :)

Is this the so-called excessive pruning?

(By the way, how do put in the chess diagram..? Sorry - newbie here :P )
Top
Modern Times
Posts: 3546
Joined: Thu Jun 07, 2012 11:02 pm

Re: Excessive pruning by SF??

Post by Modern Times »

Stockfish finds Rxh8 in a few seconds here.

Code: Select all


Engine: Stockfish 170615 x64 8CPU (512 MB)
by Tord Romstad, Marco Costalba and Joona
 27/37	 0:01 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (3.285.999) 2478  TB:8.849
 28/37	 0:01 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (4.019.935) 2644  TB:11.057
 29/37	 0:01 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (4.935.474) 2825  TB:13.954
 30/37	 0:02 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (6.111.289) 2943  TB:19.953
 31/37	 0:02 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (7.374.890) 3070  TB:24.709
 32/37	 0:02 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (8.974.571) 3286  TB:30.436
 33/38	 0:03 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (12.450.261) 3533  TB:46.875
 34/39	 0:03 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (14.621.767) 3704  TB:56.490
 35/40	 0:04 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (18.695.888) 4028  TB:76.042
 36/41	 0:05 	+2.32 	1.Rd8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Kc4 Bb6 7.Kd3 Bc7 8.Ke4 Bd6 9.Rd8 Ba3 10.Rb8 Bd6 11.Re8 Bf8 12.Kf3 Bd6 13.Rc8 Bb4 14.Rb8 (24.723.604) 4445  TB:104.912
 37/42	 0:06 	+2.38++	1.Rxh8 (29.618.024) 4801  TB:124.994
 37/42	 0:06 	+2.44++	1.Rxh8 (29.801.446) 4795  TB:126.045
 37/42	 0:06 	+2.54++	1.Rxh8 (29.966.244) 4798  TB:126.692
 37/42	 0:06 	+2.68++	1.Rxh8 (30.108.255) 4808  TB:126.881
 37/42	 0:06 	+2.89++	1.Rxh8 (30.296.133) 4820  TB:127.337
 37/42	 0:06 	+3.20++	1.Rxh8 (30.515.487) 4831  TB:127.824
 37/42	 0:06 	+3.67++	1.Rxh8 (30.789.872) 4846  TB:128.703
 37/42	 0:06 	+4.37++	1.Rxh8 (31.140.681) 4865  TB:130.434
 37/42	 0:06 	+5.42++	1.Rxh8 (32.036.562) 4897  TB:136.942
 37/42	 0:06 	+7.00++	1.Rxh8 (32.953.008) 4930  TB:147.522
 37/42	 0:06 	+9.36++	1.Rxh8 (34.158.611) 4966  TB:158.593
 37/42	 0:07 	+12.89++	1.Rxh8 (35.333.901) 4983  TB:170.120
 37/42	 0:07 	+18.20++	1.Rxh8 (39.548.109) 4963  TB:219.056
 37/42	 0:08 	+26.16++	1.Rxh8 (40.856.268) 4990  TB:231.689
 37/42	 0:08 	+38.09++	1.Rxh8 (41.441.622) 5002  TB:236.978
 37/42	 0:08 	+55.99++	1.Rxh8 (42.141.183) 5029  TB:242.233
 37/42	 0:08 	+33.17--	1.Rxh8 Kxh8 (42.304.524) 5031  TB:243.316
 37/50	 0:08 	+62.70++	1.Rxh8 (43.362.772) 5077  TB:252.704
 37/50	 0:08 	+28.43--	1.Rxh8 Kxh8 (43.524.427) 5081  TB:253.482
 37/50	 0:08 	+77.65++	1.Rxh8 (43.853.007) 5093  TB:255.391
 37/56	 0:08 	+123.41	1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc6 Ra8 8.Bd4 Rb8 9.Kc7 Ra8 10.Kd7 f5 11.e8Q Rd8+ 12.Ke7 Rxe8+ 13.Kxe8 Kg8 14.Bh8 (46.420.054) 5168  TB:285.473
Top
whereagles
Posts: 565
Joined: Thu Nov 13, 2014 12:03 pm

Re: Excessive pruning by SF??

Post by whereagles »

Thanks.

I tried it on computer that could run SF6 BMI2 (but otherwise similar specs) and now it did find Rxh8 in a few secs as well.

Actually, it found a couple of similar moves, which were nothing but some piece shuffling before engaging in Rxh8 later on. After like 3 mins it finally hooked on Rxh8, since it was the fastest way to win (#28).

Still intrigued why similar computers yielded different results. Seems like pruning may not be the culprit after all.. or does BMI2 prune differently..?

Cheers :)

PS. Thanks Gabor. Will do that. Edit: it didn't let me edit after 15 mins.. oops.
Top
kgburcham
Posts: 2016
Joined: Sun Feb 17, 2008 4:19 pm

Re: Excessive pruning by SF??

Post by kgburcham »

this version does ok

[D] 5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - -


Engine: Stockfish 070615 64 BMI2 (8192 MB)
by Tord Romstad, Marco Costalba and Joona
Found 172 tablebases.
<snip>
38/44 0:03 +2.29 1.Rc8 Kh7 2.e6 Bxd4 3.e7 Re3 4.e8Q Rxe8 5.Rxe8 Kg7 6.Re2 Bc3 7.Kc5 Ba5 8.Kd6 Kf8 9.Rb2 Bc3 10.Rc2 Be1 11.Kd5 Kg7 12.Rb2 Ba5 13.Rb7 Bc3 14.Ke4 (54.639.591) 15198 TB:294.849
39/45 0:03 +2.35++ 1.Rxh8 (62.374.601) 15707 TB:339.550
39/45 0:03 +2.41++ 1.Rxh8 (62.561.340) 15707 TB:340.214
39/45 0:03 +2.51++ 1.Rxh8 (62.798.881) 15711 TB:341.013
39/45 0:04 +2.65++ 1.Rxh8 (62.995.044) 15383 TB:341.357
39/47 0:04 +2.86++ 1.Rxh8 (63.274.543) 15380 TB:342.231
39/47 0:04 +3.17++ 1.Rxh8 (63.587.831) 15392 TB:343.125
39/47 0:04 +3.64++ 1.Rxh8 (63.974.318) 15404 TB:344.315
39/47 0:04 +4.34++ 1.Rxh8 (64.653.093) 15422 TB:347.987
39/47 0:04 +5.39++ 1.Rxh8 (65.912.236) 15432 TB:360.520
39/50 0:04 +6.96++ 1.Rxh8 (67.218.086) 15448 TB:370.144
39/50 0:04 +9.32++ 1.Rxh8 (69.469.155) 15520 TB:394.510
39/50 0:04 +12.86++ 1.Rxh8 (72.726.317) 15663 TB:421.217
39/50 0:04 +18.17++ 1.Rxh8 (76.781.652) 15847 TB:456.484
39/50 0:04 +26.13++ 1.Rxh8 (79.359.463) 15948 TB:483.103
39/50 0:05 +38.06++ 1.Rxh8 (80.546.348) 15652 TB:493.252
39/50 0:05 +55.96++ 1.Rxh8 (83.456.385) 15716 TB:514.392
39/50 0:05 +82.81++ 1.Rxh8 (87.782.853) 15943 TB:541.980
39/50 0:05 +61.42-- 1.Rxh8 Kxh8 (88.257.491) 15939 TB:548.493
39/50 0:05 +92.87++ 1.Rxh8 (89.541.520) 16032 TB:551.110
39/52 0:05 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Re8 6.Kc6 Ra8 7.Kd6 Ra6+ 8.Kc7 Ra8 9.Kc6 Kh7 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (94.648.245) 16243 TB:605.301
40/52 0:05 +123.29-- 1.Rxh8 Kxh8 (95.018.991) 16228 TB:607.519
40/52 0:05 +123.36++ 1.Rxh8 (96.784.008) 16307 TB:628.414
40/52 0:06 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Re8 6.Kc6 Ra8 7.Kd6 Ra6+ 8.Kc7 Ra8 9.Kc6 Kh7 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (101.390.080) 16227 TB:724.425
41/52 0:06 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Re8 6.Kc6 Ra8 7.Kd6 Ra6+ 8.Kc7 Ra8 9.Kc6 Kh7 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (103.441.064) 16277 TB:745.066
42/52 0:06 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rb8+ 5.Kc6 Ra8 6.Kd7 Ra7+ 7.Kd6 Ra6+ 8.Kc7 Ra8 9.Kc6 Kh7 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (106.241.136) 16369 TB:781.108
43/52 0:06 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Kh7 5.Kc6 Ra6+ 6.Kd7 Ra7+ 7.Kc8 Ra8+ 8.Kb7 Rg8 9.Kc6 Ra8 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (110.308.084) 16505 TB:833.773
44/52 0:06 +123.36 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Kh7 5.Kc6 Ra6+ 6.Kd7 Ra7+ 7.Kc8 Ra8+ 8.Kb7 Rg8 9.Kc6 Ra8 10.Bd4 Rb8 11.Kc7 Ra8 12.Kd7 Ra4 13.e8Q Rxd4+ 14.Ke7 (112.165.149) 16563 TB:855.087
45/52 0:08 +123.37 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc6 Ra8 8.Bd4 Rb8 9.Kc7 Ra8 10.Kd7 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kxf7 Rxf4+ 14.Ke7 (146.972.092) 17536 TB:1.359.180
46/52 0:08 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (156.646.462) 17841 TB:1.463.511
47/52 0:08 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (161.822.551) 17988 TB:1.515.981
48/52 0:09 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (167.633.352) 17945 TB:1.572.209
49/52 0:09 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (170.870.720) 18018 TB:1.605.483
50/52 0:09 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (176.310.293) 18118 TB:1.674.873
51/52 0:10 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (182.685.091) 18213 TB:1.741.374
52/52 0:10 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (190.996.442) 18277 TB:1.865.463
53/52 0:10 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (198.458.895) 18384 TB:1.962.028
54/52 0:11 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (208.155.122) 18393 TB:2.071.997
55/52 0:11 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (221.199.505) 18571 TB:2.262.513
56/52 0:12 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (230.005.204) 18590 TB:2.376.668
57/52 0:13 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (258.667.723) 18926 TB:2.745.262
58/52 0:14 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (271.689.436) 19059 TB:2.888.383
59/52 0:15 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (299.009.454) 19262 TB:3.307.696
60/52 0:19 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (372.339.404) 19431 TB:4.925.732
61/52 0:20 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 fxg5 10.fxg5 Ra4 11.e8Q Rxd4+ 12.Ke7 Re4+ 13.Kf8 Rf4+ 14.Qf7+ (401.953.772) 19557 TB:5.387.896
62/52 0:27 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 Ra4 10.e8Q Rxd4+ 11.Ke7 fxg5 12.Qf7+ Kh6 13.fxg5+ Kh5 14.Qh7+ (563.444.143) 20602 TB:9.275.592
63/52 0:30 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 Ra4 10.e8Q Rxd4+ 11.Ke7 fxg5 12.Qf7+ Kh6 13.fxg5+ Kh5 14.Qh7+ (640.314.835) 20801 TB:10.758.631
64/52 0:33 +123.41 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8 5.Kb6 Kh7 6.Kb7 Rg8 7.Kc7 Ra8 8.Bd4 f6 9.Kd7 Ra4 10.e8Q Rxd4+ 11.Ke7 fxg5 12.Qf7+ Kh6 13.fxg5+ Kh5 14.Qh7+ (705.097.339) 20956 TB:12.116.953
Top
syzygy
Posts: 5557
Joined: Tue Feb 28, 2012 11:56 pm

Re: Excessive pruning by SF??

Post by syzygy »

SzG wrote:Just put the fen between [D] and [/D].
There is actually no need to put the [/D] behind the fen, and the [D] tag remains visible either way. I agree this is not entirely consistent with how BBCode generally works.
[D]5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - - 23 57
Nuno Sousa wrote:Still intrigued why similar computers yielded different results. Seems like pruning may not be the culprit after all.. or does BMI2 prune differently..?
Either you've tried different versions of SF or, perhaps more likely, you're doing multithreaded searches which inherently are non-deterministic (i.e. give a different result each time).

Bc5 seems to be clearly winning, so I'm not sure Rxh8 can be said to be "better".
Top
User avatar
MikeB
Posts: 4889
Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania

Re: Excessive pruning by SF??

Post by MikeB »

whereagles wrote:From Keres-Hort (1972)

[d]5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - - 23 57

The puzzle solution is Rxh8!

Usually Stockfish finds these things in like 0.1 sec, but this time after 10 mins it's still stuck on Bc5 for +7.70 45/81 (5-men syzygy, 2 cores @2.67 MHz, SF6 POPCNT).

After I force it to Rxh8, it goes +123.4 in like 0.1 sec :)

Is this the so-called excessive pruning?

(By the way, how do put in the chess diagram..? Sorry - newbie here :P )

interesting position, but the game is already clearly won and not a good test position to check for excessive pruning. Set your pv to 5 and you will see five clearly winning lines Rd8, Rc8, Rxh8, Re8 and Bc5 are all plus 10
Top
User avatar
Eelco de Groot
Posts: 4561
Joined: Sun Mar 12, 2006 2:40 am
Full name:   

Re: Excessive pruning by SF??

Post by Eelco de Groot »

Nobody has mentioned Mate scores yet :P I suppose this is more or less correct, but I was a bit worried because in Multi PV Rainbow Serpent somehow lost track of Bc5 as one of the winning moves. Somehow move was either not checked (very bad) or did not Fail High (still bad). Very strange. It must be some kind of bug but I do not know if it is a Rainbow Serpent bug. Because the starter also mentioned Bc5, I am not sure it is purely Rainbow Serpent here or I have to revert Fail High re-searches in PV nodes as most likely culprit. But it shouldn't have this effect.


[D]5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - -

Engine: Sf20150607_003 MOD MP (512 MB)
by Tord Romstad, Marco Costalba and Joona Kiiski

53 127:33 +M23 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8
5.Kb6 Re8 6.Kb7 Kh7 7.Kc7 Ra8 8.Bd4 f6
9.Kd7 fxg5 10.fxg5 Rb8 11.e8Q Rd8+
12.Ke7 Rxe8+ 13.Kxe8 Kg8 14.Bc5 (35.805.619.209) 4678

53 127:33 +M24 1.Rc8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Re8 7.Kb7 Kh7
8.Kc7 Ra8 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (35.805.619.209) 4678

53 127:33 +M24 1.Re8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Re8 7.Kb7 Kh7
8.Kc7 Ra8 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (35.805.619.209) 4678

53 127:33 +M24 1.Rd8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Re8 7.Kb7 Kh7
8.Kc7 Ra8 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (35.805.619.209) 4678

52 127:33 +39.46 1.Bc5 Rb3+ 2.Kc6 Rc3 3.Rc8 Kh7 4.Kd7 Bg7
5.Bd6 Rf3 6.Rc7 Kg8 7.Ra7 Rf1 8.Ke8 Rxf4
9.e6 Re4 10.e7 Bf8 11.Kd7 Bxe7
12.Bxe7 f5 (35.805.619.209) 4678

52 127:33 +1.74 1.Kc6 Ra6+ 2.Kb7 Kxf8 3.Bc5+ Ke8
4.Kxa6 Kd7 5.Kb7 Bg7 6.Bb4 Bh8 7.Bc3 Ke6
8.Kc7 Ke7 9.Bb4+ Ke8 10.Kc6 Bg7
11.Bd6 Kd8 12.Bc5 Ke8 13.Bb4 Kd8
14.Kb7 (35.805.619.209) 4678

A little bit earlier 1. Bc5 had a score of 45.30 and it definitely should not happen that the score falls below that in a next iteration.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
Top
User avatar
Eelco de Groot
Posts: 4561
Joined: Sun Mar 12, 2006 2:40 am
Full name:   

Re: Excessive pruning by SF??

Post by Eelco de Groot »

Rainbow Serpent now at depth 62 thinks that 1.Bc5 is a Mate in 39. It went from Mate in 40 to Mate in 41 to Mate in 39, so that is not completely the way it should be but with these long distances also not very bad. The search is very slightly changed; this is version 004, not the GitHub version, that's 002. It would be nice if it could be confirmed with a matefinder program but it is probably not possible with Stockfish sec within the maximum number of iterations. Timewise it is not such a problem I think because Stockfish' iterations will go much faster, judging by what I saw from Kim.


[D]5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - -

Engine: Rainbow Serpent Sf20150607_004 MOD MP (512 MB)
by Tord Romstad, Marco Costalba and Joona Kiiski

62 784:10 +M23 1.Rxh8 Kxh8 2.e6+ Kg8 3.e7 Ra8 4.Bf6 Rc8
5.Kb6 Rb8+ 6.Kc6 Ra8 7.Kc7 Kh7 8.Bd4 f6
9.Kd7 fxg5 10.fxg5 Rb8 11.e8Q Rd8+
12.Ke7 Rxe8+ 13.Kxe8 Kg8 14.Bc5 (255.541.213.935) 5431

62 784:10 +M24 1.Rc8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Rb8+ 7.Kc6 Ra8
8.Kc7 Kh7 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (255.541.213.935) 5431

62 784:10 +M24 1.Re8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Rb8+ 7.Kc6 Ra8
8.Kc7 Kh7 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (255.541.213.935) 5431

62 784:10 +M24 1.Rd8 Kh7 2.Rxh8+ Kxh8 3.e6+ Kg8
4.e7 Ra8 5.Bf6 Rc8 6.Kb6 Rb8+ 7.Kc6 Ra8
8.Kc7 Kh7 9.Bd4 f6 10.Kd7 fxg5
11.fxg5 Rb8 12.e8Q Rd8+ 13.Ke7 Rxe8+
14.Kxe8 (255.541.213.935) 5431

62 784:10 +M39 1.Bc5 Ra2 2.Kc6 Rc2 3.Rc8 Kh7 4.Kd7 Bg7
5.Bd6 Rg2 6.Rc7 Kg8 7.Ra7 Rf2 8.Ke8 Rxf4
9.e6 Re4 10.e7 Bf8 11.Kd8 Bxe7+
12.Rxe7 Rd4 13.Kc7 Rc4+ 14.Kb7 (255.541.213.935) 5431

61 784:10 +1.74 1.Kc6 Ra6+ 2.Kb7 Kxf8 3.Bc5+ Ke8
4.Kxa6 Kd7 5.Kb5 Bg7 6.Kb6 Bh8 7.Bd4 Ke7
8.Kc6 Bg7 9.Kd5 Bh8 10.Bc5+ Kd7
11.Bb4 Bg7 12.Ba3 Ke8 13.Kc6 Kd8
14.Bd6 (255.541.213.935) 5431
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
Top
Sergei
Posts: 3
Joined: Wed Jun 24, 2015 9:43 pm

Re: Excessive pruning by SF??

Post by Sergei »

whereagles wrote:From Keres-Hort (1972)

5R1b/5pk1/6p1/1K2P1P1/3B1P2/r7/8/8 w - - 23 57

The puzzle solution is Rxh8!

Usually Stockfish finds these things in like 0.1 sec, but this time after 10 mins it's still stuck on Bc5 for +7.70 45/81 (5-men syzygy, 2 cores @2.67 MHz, SF6 POPCNT).
My similar configuration (5-men syzygy, single-threaded, i5 2.7GHz, SF6 POPCNT, 4GB hash) cannot find Rxh8 either. But when I switch to 6-men syzygy, Rxh8 is found in about 15 s.

So 6-men table base makes a difference. By the way, were there any matches between SF equipped with 5-men syzygy and 6-men? I wonder what was the score.
Top

Return to “Computer Chess Club: General Topics”