Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Vas why the 59 ply limit ?
- - By Jim Walker (***) Date 2007-06-21 10:31
I'm beginning to wonder if the 59 ply limit of Rybka was part of the problem in drawing with Zappa in the WCCC 2007.  Analysis on my slow XP3100 shows Rybka was not that far away from finding the winning moves.  So it makes me wonder if on your fast hardware were you hitting the 59 ply limit around that time.  I also wonder why this limit exist.
Jim
Parent - - By Vasik Rajlich (Silver) Date 2007-06-21 13:47
It's to prevent various overflows. It wouldn't be hard to increase it - are there really positions where the limit is reached and more searching would help?

Vas
Parent - - By Lee Ma Hong (**) Date 2007-06-21 15:58
Paging correspondence chess players who run infinite analysis for 24 hours ...
Parent - By vroger007 (**) Date 2007-06-21 17:32
or game analysis during 3 days :-)
...  no, seriously, it's rare but there certainly are such cases every now and then!
Roger
Parent - - By Uri Blass (*****) Date 2007-06-21 18:37
For the first question the answer is positive
In this case more searching is not going to help to find a better move.
Only you know if more searching can help to find a draw score.

New game - Rybka 2.3.2a 32-bit
8/8/7K/4k2P/7P/7P/7P/8 b - - 0 1


Analysis by Rybka 2.3.2a 32-bit :

1...Ke5-f6 2.Kh6-h7
  +-  (9.19)   Depth: 5   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7
  +-  (9.26)   Depth: 6   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6
  +-  (9.56)   Depth: 7   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8
  +-  (9.24)   Depth: 8   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5
  +-  (9.22)   Depth: 9   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7
  +-  (9.27)   Depth: 10   00:00:13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.h3-h4
  +-  (9.56)   Depth: 11   00:00:13  1kN
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.h3-h4 Kf7-f8
  +-  (9.27)   Depth: 12   00:00:13  2kN
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.Kh7-h8 Kf7-f8 6.h6-h7 Kf8-f7
  +-  (9.22)   Depth: 13   00:00:13  3kN
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.Kh7-h8 Kf7-f8 6.h6-h7 Kf8-f7 7.h5-h6
  +-  (9.62)   Depth: 14   00:00:14  7kN
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.Kh7-h8 Kf7-f8 6.h6-h7 Kf8-f7 7.h5-h6 Kf7-f8
  +-  (9.27)   Depth: 15   00:00:14  13kN, tb=1
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.h5-h6 Kf7-f8 4.h4-h5 Kf8-f7 5.Kh7-h8 Kf7-f8 6.h6-h7 Kf8-f7 7.h5-h6 Kf7-f8 8.h3-h4
  +-  (9.27)   Depth: 16   00:00:14  25kN, tb=5
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.h6-h7
  +-  (9.22)   Depth: 17   00:00:14  43kN, tb=13
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.h6-h7
  +-  (9.62)   Depth: 18   00:00:15  63kN, tb=33
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.h6-h7
  +-  (9.27)   Depth: 19   00:00:15  85kN, tb=70
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.Kh8-h7
  +-  (9.27)   Depth: 20   00:00:16  110kN, tb=133
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-h8 Kf8-f7 7.Kh8-h7 Kf7-f8 8.Kh7-h8
  +-  (8.66)   Depth: 21   00:00:17  144kN, tb=255
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-h8 Kf8-f7 7.Kh8-h7 Kf7-f8 8.Kh7-h8
  +-  (9.04)   Depth: 22   00:00:19  192kN, tb=441
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-h8 Kf8-f7 7.Kh8-h7 Kf7-f8 8.Kh7-h8
  +-  (8.66)   Depth: 23   00:00:21  269kN, tb=661
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-h8 Kf8-f7 7.Kh8-h7 Kf7-f8 8.Kh7-h8
  +-  (8.83)   Depth: 24   00:00:25  407kN, tb=920
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.h3-h4 Kf7-f8 8.Kh7-g6
  +-  (6.79)   Depth: 25   00:00:32  619kN, tb=1323
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.h3-h4 Kf7-f8 8.Kh7-g6
  +-  (6.77)   Depth: 26   00:00:45  1002kN, tb=1930
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.h3-h4 Kf7-f8 8.Kh7-g6
  +-  (6.62)   Depth: 27   00:01:06  1621kN, tb=3284
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.h3-h4
  +-  (4.93)   Depth: 28   00:01:28  2320kN, tb=5395
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.h4-h5 Kf8-f7 7.Kh7-h8 Kf7-f8 8.h3-h4
  +-  (4.93)   Depth: 29   00:01:28  2320kN, tb=5400
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.Kg6-g5 Kg8-h8 8.Kg5-f5
  +-  (4.93)   Depth: 30   00:01:28  2320kN, tb=5408
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.Kg6-g5 Kg8-h8 8.Kg5-f5
  +-  (4.93)   Depth: 31   00:01:28  2320kN, tb=5416
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.Kg6-g5 Kg8-h8 8.Kg5-g4
  +-  (4.93)   Depth: 32   00:01:28  2321kN, tb=5426
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.Kg6-g5 Kg8-h8 8.Kg5-g4
  +-  (4.93)   Depth: 33   00:01:28  2321kN, tb=5435
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 34   00:01:28  2321kN, tb=5444
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 35   00:01:29  2321kN, tb=5453
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 36   00:01:29  2322kN, tb=5466
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 37   00:01:29  2324kN, tb=5477
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 38   00:01:29  2324kN, tb=5489
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 39   00:01:29  2324kN, tb=5500
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 40   00:01:29  2325kN, tb=5514
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 41   00:01:29  2325kN, tb=5526
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 42   00:01:29  2326kN, tb=5554
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 43   00:01:29  2326kN, tb=5569
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 44   00:01:29  2326kN, tb=5579
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 45   00:01:29  2327kN, tb=5590
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 46   00:01:29  2327kN, tb=5601
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 47   00:01:29  2327kN, tb=5612
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 48   00:01:29  2328kN, tb=5625
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 49   00:01:29  2328kN, tb=5636
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 50   00:01:29  2328kN, tb=5648
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 51   00:01:29  2328kN, tb=5660
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 52   00:01:29  2329kN, tb=5673
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 53   00:01:29  2329kN, tb=5689
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 54   00:01:29  2330kN, tb=5703
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 55   00:01:29  2330kN, tb=5716
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 56   00:01:29  2331kN, tb=5730
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 57   00:01:29  2331kN, tb=5744
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 58   00:01:29  2332kN, tb=5764
1...Ke5-f6 2.Kh6-h7 Kf6-f7 3.Kh7-h8 Kf7-f8 4.h5-h6 Kf8-f7 5.Kh8-h7 Kf7-f8 6.Kh7-g6 Kf8-g8 7.h4-h5 Kg8-h8 8.Kg6-f6
  +-  (4.93)   Depth: 59   00:01:29  2332kN, tb=5780

(,  15.06.2007)
Parent - - By Vasik Rajlich (Silver) Date 2007-06-25 14:10
More searching here won't help.

By the way - it seems that Rybka 2.3.2a can have some problems with repetition detection. It's a side-effect of some changes and will need some more work.

Vas
Parent - - By Uri Blass (*****) Date 2007-06-25 14:45
I wonder if simply not using hash for pruning when the remaining depth is bigger than 15 is a simple solution that is good enough
both for the 50 move rule problem and for the repetition problem.

I think that usually you can expect pruning by hash soon so you are not going to lose significant time by not pruning in the first plies.
You can try it even with constant depth of 10 at blitz and if it works than it suggests that it can also work for 15 at long time controls.

Uri
Parent - - By Vasik Rajlich (Silver) Date 2007-06-25 15:13
Yes, hashing is of course the problem. Your solution is interesting, I will have to think about it. A depth == 30 entry could need significant search to get to the depth == 15 entries, and child entries might be lost from the hash table.

Vas
Parent - - By Uri Blass (*****) Date 2007-06-25 19:49
Rybka practically does not get depth 30 in middle game positions and I assume that even in the case that she get depth 30 most child entries at depth 15 are not going to be lost because they are more important entries relative to entries at small depth.

I guess that about 99.9% of the data about pruning comes from depth that is smaller than depth 15 and you have enough space to save the entries at depth 15 with almost no loss in data.

Uri
Parent - - By Uri Blass (*****) Date 2007-06-26 04:08 Edited 2007-06-26 04:12
I can add that the number of plies when you do not prune may be dependent on the size of hash tables and the tree.

You can simply count the number of nodes that you search in the first n plies of the search and remember it inside an array  and you can decide not to use hash for pruning in the next iteration in the first n plies if  the number of nodes in previous iteration in the same plies is both smaller than x% of the hash size and smaller than y% of the size of the tree.

x% of the hash size is needed to having small risk of forgetting information and y% of the size of the tree is needed in order not to spend too much time on nodes that you already searched.

I do not think that it is the best solution but this solution is probably relatively simple solution and better than nothing.

If this solution seems too complicate to implement in 2.3.2b you can certainly not use hash for pruning when the remaining depth is not bigger than 15 and give the user the option to change it by changing parameters(in long analysis the program may lose information if it searches to big depth of 30 plies but correspondence players may prefer it relative to the risk of not seeing a draw by repetition or by fifty move rule regardless of depth.

Uri
Parent - - By Vasik Rajlich (Silver) Date 2007-06-27 13:43
Yes, I can think of some simpler adaptive algorithms for this. For example, the maximum transposition table pruning depth could be the depth of the last search minus X, where X is something like 4 or 6.

However, it's really not a very good solution. For one thing, you'll still miss a lot of repetitions deeper in the search. These types of repetitions can be important, although users may not notice them and post them on your forum. :)

What this topic needs is a good day of just thinking. It's really amazing how complex the transposition table can be. I'd estimate that I've spent 150-200 hours of my life just thinking about the transposition table - not working or debugging, but just thinking about maybe 200 lines of code total.

Vas
Parent - - By Uri Blass (*****) Date 2007-06-27 15:51
I basically want chess programs to see everything if you search deep enough so not pruning the first 4 plies is not a solution for analysis.
A simple solution for analysis may be pruning only in the last n plies when the user can choose n.

I agree that hash is not a simple thing and in movei I still do not use hash for pruning simply because I was interested more in improving movei
in other ways(originally it was because of path dependent evaluation but I found that my path depedent evaluation did not help much).
I wonder how many rating points programs earn from using hash for pruning.

Uri
Parent - By Vasik Rajlich (Silver) Date 2007-06-29 13:27
Yes, true, the D-4 plies solution is not very good. The problem with not pruning in last N plies is that you're going to have worse search performance as the search tree gets bigger.

Hash pruning is a huge pain in the butt but it's worth it. I'm sure that the benefit is at least 50 Elo. Besides, even 20 Elo would be too much to ignore. You have to be willing to tackle these types of things, otherwise your search will be too inefficient.

Vas
Parent - By Kapaun (****) Date 2007-06-25 15:15 Edited 2007-06-25 15:28
Yes, I noticed that, too. You will need to have a close look on the variants, because sometimes she smuggles a repetition variant into the output, but not as a draw. Can be pretty annoying...
Parent - By Jim Walker (***) Date 2007-06-21 22:41
Well I noticed on my XP3100 that when I got the position to where it was around mate in 30 or 31 somewhere in that area, Rybka was able to find a mate score fairly quickly.  So my thinking was that on your faster hardware it's possible that more plies might have allowed Rybka to find a mate in 37-39.  As hardware gets faster I believe you will have to consider this possibility.  If Rybka is spitting out moves in 30 seconds because it has hit the upper limit you have set and it really has much more time why not let it go?  Just a thought.  This will only happen in very late endgames of course (like the one with Zap).
Regards,
Jim
Parent - - By John Hyden (*) Date 2007-06-22 19:14 Edited 2007-06-22 19:17
Yes Vas, there exist such positions. I recommend making the depth limit an optional setting, 6500+ can be the maximum moves in a chess game if there is 50 move rule. I will post one example:

White to play and win (series of moves)-

bBrb1B2/P1n1r2p/1Kp1Pb1p/2pk1P1p/5P2/1P2pP2/1pP1P3/1R4n1 w - -


Here 59 plies is not enough to solve this, even 100 plies is not enough. I challenge you to solve this. I will post the answer if I don't get a reply or you fail to solve it. It maybe a rare position but there maybe a lot of this just waiting to be played. And maybe there is no program that can solve this clearly at the moment.

Some Questions:
1. Does the depth of the search really increases the accuracy in all positions? Or does it hurt the accuracy in some types of (maybe rare) positions?
2.How do you describe the search of your program? How do they solve the positions that maybe beyond their search depth?
Parent - By John Hyden (*) Date 2007-06-22 19:38
I tested this on programs, and they give me weird evaluations.
Parent - - By Lukas Cimiotti (Bronze) Date 2007-06-22 19:56
This is an artificial position, that cannot happen in reality. How do you get those triple pawns? Only by taking opponents pieces. And the only piece that is missing for black is the queen,  but there is a 2. black bishop. So this position is impossible to reach.
Parent - By Uri Blass (*****) Date 2007-06-22 21:43
I agree that the position is illegal.

8 white pawns and 2 white bishops on black squares is illegal.

Uri
Parent - - By John Hyden (*) Date 2007-06-23 19:31
I don't care if that is an impossible position, that is just an example and a matter of "ifs". What I mean is that there might be similar positions like this that is playable or has already been played (who knows). How could you solve a very deep position efficiently? And is it possible to solve it even without going through lots of variations?

Anyone who have solved it? I will give the answer soon if no one responds. 100 plies are not enough here.
Parent - - By Uri Blass (*****) Date 2007-06-23 23:24
Fritz gui cares about it and simply does not accept your position.
Parent - By Uri Blass (*****) Date 2007-06-23 23:59
I find that rybka can analyze it under shredder gui and it seems that you did not understand Vas question.
Vas asked:
"are there really positions where the limit is reached and more searching would help?"

The limit is not reached in the relevant position so you did not give a good example
Assuming branching factor of 2 rybka is not going to reach depth 59 in the next 1000000 years.

bBrb1B2/P1n1r2p/1Kp1Pb1p/2pk1P1p/5P2/1P2pP2/1pP1P3/1R4n1 w - -


Engine: Rybka 2.3.2a 32-bit (512 MB)
by Vasik Rajlich

8.00   0:01   -6.23    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.b4 (75.624) 66

9.00   0:01   -6.08    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.b4 b1Q (84.171) 65

10.00  0:01   -6.47    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.b4 b1Q 6.Rxb1 cxb4+ 7.Ka5 (103.718) 66

11.00  0:01   -6.45    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.b4 b1Q 6.Rxb1 cxb4+ 7.Ka5 Kc5 (129.160) 67

12.01  0:03   -6.72    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.b4 b1Q 6.Rxb1 cxb4+ 7.Ka5 Kc5 (268.664) 72

13.01  0:08   -7.04    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h2 6.f6 Kxe6 7.fxe7 Bxe7 (630.688) 75

14.01  0:17   -6.57    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 h2 7.Ka5 h4 (1.199.258) 70

15.01  0:27   -6.58    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 h2 7.Rh1 h6 (1.900.893) 70

16.01  0:38   -6.46    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 h2 7.Rh1 h6 (2.605.472) 70

17.01  0:59   -6.53    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 h4 7.f6 Kxe6 (4.000.345) 69

18.01  1:48   -5.61    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 h6 7.Ka3 Bb7 (7.233.584) 68

19.01  2:53   -5.48    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h5 6.Ka4 Bb7 7.Ka3 h4 (11.529.199) 67

20.01  4:56   -5.92    1.Rd1+ Bd4 2.c4+ Kd6 3.Rxg1 h4 4.Rd1 h3
                       5.Ka5 h2 6.Kb6 h5 7.Ka5 h6 (19.929.291) 68
Parent - - By Gaмßito (****) Date 2007-06-24 02:07
Hi,

Give us the solution. 2.3.2a still continue thinking that Rd1+ is the best move for white here. 

21.01   9:28   -6.06   1.Td1+ Ad4 2.c4+ Rd6 3.Txg1 h4 4.Td1 h3 5.Ra5 h2 6.Rb6 h5 7.Ra5 h6 (28.204.484) 50
22.01  23:57   -5.17   1.Td1+ Ad4 2.c4+ Rd6 3.Txg1 h4 4.Td1 h3 5.Ra5 h2 6.Th1 Ac3+ 7.Ra4 Ab7 (68.464.919) 48
23.01  58:44   -5.27   1.Td1+ Ad4 2.c4+ Rd6 3.Txg1 h4 4.Td1 h3 5.Ra5 h5 6.Ra4 h4 7.Th1 Ab7 (168.853.341) 49
24.01  91:29   -5.36   1.Td1+ Ad4 2.c4+ Rd6 3.Txg1 h4 4.Td1 h3 5.Ra5 h5 6.Ra4 h4 7.Th1 Ab7 (259.666.807) 48

Regards,
Gambito.
Parent - - By John Hyden (*) Date 2007-06-25 15:13 Edited 2007-06-25 15:19
Its checkmate in 290 moves (longest ever discovered yet). See this for more details: http://www-lipn.univ-paris13.fr/~banderier/Chess/index.html
Parent - By George Tsavdaris (****) Date 2007-06-25 17:45
I thought this one with mate in 292 is the longest one!

q5nn/1p2p3/p1k1P1p1/6Pp/PKp1p1pP/8/2P1P1PP/3Q4 w - - 0 1


1. Qd7+ Kb6 2. a5+ Ka7 3. Qd4+ Kb8 4. Qd8+ Ka7 5. Qb6+ Kb8 6. c3 Kc8 7. Qc5+
Kb8 8. Qe5+ Kc8 9. Qd5 Kc7 10. Qd7+ Kb8 11. Qd8+ Ka7 12. Qb6+ Kb8 13. e3 Kc8
14. Qc5+ Kb8 15. Qe5+ Kc8 16. Qd5 Kc7 17. Qd7+ Kb8 18. Qd8+ Ka7 19. Qb6+ Kb8
20. g3 Kc8 21. Qc5+ Kb8 22. Qe5+ Kc8 23. Qd5 Kc7 24. Qd7+ Kb8 25. Qd8+ Ka7 26.
Qb6+ Kb8 27. Ka3 Kc8 28. Qc5+ Kb8 29. Qe5+ Kc8 30. Qd5 Kc7 31. Qd7+ Kb8 32.
Qd8+ Ka7 33. Qb6+ Kb8 34. Kb2 Kc8 35. Qc5+ Kb8 36. Qe5+ Kc8 37. Qd5 Kc7 38.
Qd7+ Kb8 39. Qd8+ Ka7 40. Qb6+ Kb8 41. Kc2 Kc8 42. Qc5+ Kb8 43. Qe5+ Kc8 44.
Qd5 Kc7 45. Qd7+ Kb8 46. Qd8+ Ka7 47. Qb6+ Kb8 48. Kd2 Kc8 49. Qc5+ Kb8 50.
Qe5+ Kc8 51. Qd5 Kc7 52. Qd7+ Kb8 53. Qd8+ Ka7 54. Qb6+ Kb8 55. Ke2 Kc8 56.
Qc5+ Kb8 57. Qe5+ Kc8 58. Qd5 Kc7 59. Qd7+ Kb8 60. Qd8+ Ka7 61. Qb6+ Kb8 62.
Kf2 Kc8 63. Qc5+ Kb8 64. Qe5+ Kc8 65. Qd5 Kc7 66. Qd7+ Kb8 67. Qd8+ Ka7 68.
Qb6+ Kb8 69. Kg2 Kc8 70. Qc5+ Kb8 71. Qe5+ Kc8 72. Qd5 Kc7 73. Qd7+ Kb8 74.
Qd8+ Ka7 75. Qb6+ Kb8 76. h3 gxh3+ 77. Kxh3 Kc8 78. Qc5+ Kb8 79. Qe5+ Kc8 80.
Qd5 Kc7 81. Qd7+ Kb8 82. Qd8+ Ka7 83. Qb6+ Kb8 84. Kg2 Kc8 85. Qc5+ Kb8 86.
Qe5+ Kc8 87. Qd5 Kc7 88. Qd7+ Kb8 89. Qd8+ Ka7 90. Qb6+ Kb8 91. Kf2 Kc8 92.
Qc5+ Kb8 93. Qe5+ Kc8 94. Qd5 Kc7 95. Qd7+ Kb8 96. Qd8+ Ka7 97. Qb6+ Kb8 98.
Ke2 Kc8 99. Qc5+ Kb8 100. Qe5+ Kc8 101. Qd5 Kc7 102. Qd7+ Kb8 103. Qd8+ Ka7
104. Qb6+ Kb8 105. Kd2 Kc8 106. Qc5+ Kb8 107. Qe5+ Kc8 108. Qd5 Kc7 109. Qd7+
Kb8 110. Qd8+ Ka7 111. Qb6+ Kb8 112. Kc2 Kc8 113. Qc5+ Kb8 114. Qe5+ Kc8 115.
Qd5 Kc7 116. Qd7+ Kb8 117. Qd8+ Ka7 118. Qb6+ Kb8 119. Kb2 Kc8 120. Qc5+ Kb8
121. Qe5+ Kc8 122. Qd5 Kc7 123. Qd7+ Kb8 124. Qd8+ Ka7 125. Qb6+ Kb8 126. Ka3
Kc8 127. Qc5+ Kb8 128. Qe5+ Kc8 129. Qd5 Kc7 130. Qd7+ Kb8 131. Qd8+ Ka7 132.
Qb6+ Kb8 133. Kb4 Kc8 134. Qc5+ Kb8 135. Qe5+ Ka7 136. Qd4+ Kb8 137. Qxh8 Kc7
138. Qe5+ Kc8 139. Qd5 Kb8 140. Qd8+ Ka7 141. Qb6+ Kb8 142. Ka3 Kc8 143. Qc5+
Kb8 144. Qe5+ Kc8 145. Qd5 Kc7 146. Qd7+ Kb8 147. Qd8+ Ka7 148. Qb6+ Kb8 149.
Kb2 Kc8 150. Qc5+ Kb8 151. Qe5+ Kc8 152. Qd5 Kc7 153. Qd7+ Kb8 154. Qd8+ Ka7
155. Qb6+ Kb8 156. Kc2 Kc8 157. Qc5+ Kb8 158. Qe5+ Kc8 159. Qd5 Kc7 160. Qd7+
Kb8 161. Qd8+ Ka7 162. Qb6+ Kb8 163. Kd2 Kc8 164. Qc5+ Kb8 165. Qe5+ Kc8 166.
Qd5 Kc7 167. Qd7+ Kb8 168. Qd8+ Ka7 169. Qb6+ Kb8 170. Ke2 Kc8 171. Qc5+ Kb8
172. Qe5+ Kc8 173. Qd5 Kc7 174. Qd7+ Kb8 175. Qd8+ Ka7 176. Qb6+ Kb8 177. Kf2
Kc8 178. Qc5+ Kb8 179. Qe5+ Kc8 180. Qd5 Kc7 181. Qd7+ Kb8 182. Qd8+ Ka7 183.
Qb6+ Kb8 184. Kg2 Kc8 185. Qc5+ Kb8 186. Qe5+ Kc8 187. Qd5 Kc7 188. Qd7+ Kb8
189. Qd8+ Ka7 190. Qb6+ Kb8 191. Kh3 Kc8 192. Qc5+ Kb8 193. Qe5+ Kc8 194. Qd5
Kc7 195. Qd7+ Kb8 196. Qd8+ Ka7 197. Qb6+ Kb8 198. g4 Kc8 199. Qc5+ Kb8 200.
Qe5+ Kc8 201. Qd5 Kc7 202. Qd7+ Kb8 203. Qd8+ Ka7 204. Qb6+ Kb8 205. gxh5 gxh5
206. Kg2 Kc8 207. Qc5+ Kb8 208. Qe5+ Kc8 209. Qd5 Kc7 210. Qd7+ Kb8 211. Qd8+
Ka7 212. Qb6+ Kb8 213. Kf2 Kc8 214. Qc5+ Kb8 215. Qe5+ Kc8 216. Qd5 Kc7 217.
Qd7+ Kb8 218. Qd8+ Ka7 219. Qb6+ Kb8 220. Ke2 Kc8 221. Qc5+ Kb8 222. Qe5+ Kc8
223. Qd5 Kc7 224. Qd7+ Kb8 225. Qd8+ Ka7 226. Qb6+ Kb8 227. Kd2 Kc8 228. Qc5+
Kb8 229. Qe5+ Kc8 230. Qd5 Kc7 231. Qd7+ Kb8 232. Qd8+ Ka7 233. Qb6+ Kb8 234.
Kc2 Kc8 235. Qc5+ Kb8 236. Qe5+ Kc8 237. Qd5 Kc7 238. Qd7+ Kb8 239. Qd8+ Ka7
240. Qb6+ Kb8 241. Kb2 Kc8 242. Qc5+ Kb8 243. Qe5+ Kc8 244. Qd5 Kc7 245. Qd7+
Kb8 246. Qd8+ Ka7 247. Qb6+ Kb8 248. Ka3 Kc8 249. Qc5+ Kb8 250. Qe5+ Kc8 251.
Qd5 Kc7 252. Qd7+ Kb8 253. Qd8+ Ka7 254. Qb6+ Kb8 255. Kb4 Kc8 256. Qc5+ Kb8
257. Kxc4 Qa7 258. Qxa7+ Kxa7 259. Kc5 b6+ 260. axb6+ Kb7 261. c4 a5 262. Kb5
a4 263. Kxa4 Kxb6 264. Kb4 Kc6 265. c5 Kd5 266. Kb5 Kxe6 267. Kc6 Kf7 268. Kd7
e5 269. c6 Ne7 270. c7 Nd5 271. c8=Q Nb6+ 272. Kd8 Nxc8 273. Kxc8 Ke8 274. Kc7
Ke7 275. Kc6 Ke6 276. Kc5 Ke7 277. Kd5 Kf7 278. Kxe5 Kg7 279. Kxe4 Kg6 280. Kf4
Kf7 281. Kf5 Kg7 282. e4 Kf7 283. g6+ Ke7 284. e5 Kf8 285. e6 Ke7 286. g7 Kd6
287. g8=Q Kc5 288. e7 Kd4 289. e8=Q Kc5 290. Qe5+ Kb4 291. Qgb8+ Kc4 292. Qbb5#
1-0

Has this been refuted and proved to be shorter?
Parent - - By Gaмßito (****) Date 2007-06-22 21:09
Hi, 
 
How can you prove that 6500 is the maximum number of moves in the chess game? Some weeks ago, we arrive to the conclusion that it is impossible to get, even,  more than 6000.

Regards,
Gambito.
Parent - - By John Hyden (*) Date 2007-06-23 19:40
"The longest game is 6350 moves, allowing 50 moves between each pawn
move or capture. The longest tournament game on record between masters
lasted 168 moves, and the shortest four moves. (Chernev, Curious Chess
Facts, The Black Knight Press, 1937.)"


Source - http://www.pi.infn.it/%7Ecarosi/chess/shannon.txt

Just rounded it off.
Parent - By turbojuice1122 (Gold) Date 2007-06-23 23:14
I think you'll find that the author is quite wrong here--I had at first arrived at that number, and then realized that you eventually lose some of the pawns that you'd promote and eventually capture by the algorithm that leads to 6350--or something--I'm thinking about other stuff right now, so I don't have time to go through the argument again, but a number of us in this forum eventually agreed on a number that was less than that--it might have been 6100.
Parent - - By Gaмßito (****) Date 2007-06-24 00:16
Thanks for the info, but how I can reach to 6350 moves?? I do not see any examples there and I will like to see at least one example.

Regards,
Gambito.
Parent - - By John Hyden (*) Date 2007-06-25 15:25
You don't see examples because it is impossible right now to produce that long game without any blunders. You have to play perfectly, probably if you have 32-men tablebase you will find out that game. But if you want both sides to cooperate it is also unclear right now because a move might affect other moves hundreds of plies far ahead and therefore making your game inaccurate.
Parent - By turbojuice1122 (Gold) Date 2007-06-25 18:00
You don't need to play perfectly.  In fact, a good mathematical exercise (a very difficult one) would be to figure out how many possible ways are there to create this longest possible game.  My guess is that the answer is a * 10^b, where a is a real number between 1 and 10 and b is a positive integer that is greater than 3 or 4.
Parent - By Uly (Gold) Date 2007-06-26 08:24
I thought Uri had proof that the longest game had 5898 moves and he gave an example:

http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=10760#pid10760

(I think that the idea is to make idle moves until the 50 move rule is going to hit, then do something useful)
Parent - By George Tsavdaris (****) Date 2007-06-22 22:38

>Yes Vas, there exist such positions.


How do you know that this is a position that Vas was referring?
In order to be one, it should not just have a solution which is above 59 plies, it has to have the following property:
Rybka to search 59 plies and not find the solution.

So in order your position to be the one that Vas was referring, you have to provide 59 ply analysis of Rybka that does not show the solution.

>It maybe a rare position but there maybe a lot of this just waiting to be played.


It's not a rare Chess position. It's an improbable Chess position! It just can't happen during a (legal) Chess game.....
You didn't say of course Chess position but only position, but here we are speaking about Chess....

>And maybe there is no program that can solve this clearly at the moment.


We can't expect from the Chess programs to solve non-Chess positions.....
Parent - - By Vasik Rajlich (Silver) Date 2007-06-25 14:13
Uri's answer applies here - a bigger limit would not help you here, unless you had a lot of time on your hands :)

Re. Rybka's search, more search is better more often than it's worse, at least in typical practical positions.

Vas
Parent - - By John Hyden (*) Date 2007-06-25 15:18
So you can't solve it unless you have plenty of time. I will give you an easier one:

White to play and win (series of moves):

q5nn/1pk1p3/p3P1p1/P3rpP1/1Kp1pP1p/8/2PQ1PP1/8 w - -


Another one that is unclear to your programs. I hope you solve it and Happy Birthday!
Parent - By vroger007 (**) Date 2007-06-25 21:27
hi,
the sad thing about this position is that rybka gives a move repetition and no draw score:
1. (-0.90): 4.Qd8+ Ka7 5.Qb6+ Kb8 6.Qd8+ Ka7 7.Qb6+ Kb8 8.Qd8+ Ka7 9.Qb6+ Kb8 10.Qd8+ Ka7
doesn't this justify a 2.3.2b ?
(note: that's the PV after I manually enter the first couple of moves Qd7+ Kb8 Qd8+ Ka7 Qb6+ Kb8)
Parent - - By John Hyden (*) Date 2007-07-01 18:58
No reply or comment to the problem Vas? I have no time time to wait until your 150th birthday so I will just post the answers. Easier than the previous one. Not mate in 290 but in 257.

q5nn/1pk1p3/p3P1p1/P3rpP1/1Kp1pP1p/8/2PQ1PP1/8 w - -


See this for more info: http://chessmind.powerblogs.com/files/blathy1.htm . IMHO I think this is where brute intuition persist rather than brute calculation. Or combined?
Parent - - By Uri Blass (*****) Date 2007-07-01 19:51
No reply because your positions are not interesting.
It is well known that there are positions that chess programs are unable to solve.

The question was if bigger depth than depth 59 could help and the positions that you give are not relevant because rybka is not going to solve them regardless of depth and I do not care about problems that it may solve after million of years here.

I will give another challange to posters.
1)Are there positions when rybka finds the best move but needs depth 59 to get it?
2)Are there positions when rybka get depth 59 and play a wrong move?

Uri
Parent - By Vasik Rajlich (Silver) Date 2007-07-03 18:22
Yes, I am looking for a position where the depth limit prevents Rybka from finding the solution in a practical setting.

Vas
Up Topic Rybka Support & Discussion / Rybka Discussion / Vas why the 59 ply limit ?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill