Scorpio & egbb issue (OSX)

syzygy · Post by **syzygy** » Sun Feb 02, 2014 3:32 pm

Max wrote:
syzygy wrote:
hgm wrote:
Max wrote:I didn't realize that Scorpio needs some more search time, even with egbbs.
The problem is that WDL info offers zero help in end-games like KBNK, where every position where white does not obviously hang a piece will be won. And knowing that you must not blunder away a piece doesn't help you finding the mate, or any simple 3-ply search without eval would already be able to do it. So KBNK remains exactly as difficult with or without EGBB.
The use of bitbases likely makes it harder to win KBNK. If they are probed during the searched, it is overhead in return for no information. If they are used to prune branches, it is likely the search won't be able to make progress where without pruning it could. When KBNK is on the board, bitbases should be disabled.
Thanks for your explanations!

This applies for Syzygy-bases too, right? Stockfish with Syzygy-bases didn't always choose the shortest way to win here.

If you have the DTZ tables, the engine is guaranteed to deliver mate within the 50-move rule from any won position with 6 pieces or less. There is no guarantee that it will be the shortest mate possible.

If you only have WDL, then SF+TB will try to win KBNK based on SF's regular search and evaluation. With enough time it will manage, but there is no guarantee.

Daniel Shawul · Post by **Daniel Shawul** » Sun Feb 02, 2014 3:41 pm

syzygy wrote:
hgm wrote:
Max wrote:I didn't realize that Scorpio needs some more search time, even with egbbs.
The problem is that WDL info offers zero help in end-games like KBNK, where every position where white does not obviously hang a piece will be won. And knowing that you must not blunder away a piece doesn't help you finding the mate, or any simple 3-ply search without eval would already be able to do it. So KBNK remains exactly as difficult with or without EGBB.
The use of bitbases likely makes it harder to win KBNK. If they are probed during the searched, it is overhead in return for no information. If they are used to prune branches, it is likely the search won't be able to make progress where without pruning it could. When KBNK is on the board, bitbases should be disabled.

No, speak for your bitbases. My Scorpio bitbases handle them fine. Honestly I don't know how he managed to reproduce an non-winning KBNK game because I can't with any decent time control! I have my theories though

Daniel Shawul · Post by **Daniel Shawul** » Sun Feb 02, 2014 4:00 pm

Nope. Here it wins with 40 moves in 10 seconds. And even more with Stockfish.

[Event "Computer Chess Game"]
[Site "DANIEL-PC"]
[Date "2014.02.02"]
[Round "-"]
[White "Fruit 2.1"]
[Black "Scorpio_09:52:46"]
[Result "0-1"]
[TimeControl "40/10"]
[FEN "8/8/4K2P/7n/3k4/8/8/1b6 w - - 12 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . .
. . . . K . . P
. . . . . . . n
. . . k . . . .
. . . . . . . .
. . . . . . . .
. b . . . . . .
white to play
--------------}
1. Ke7 {-6.45/11} Bh7 {+50.52/16 0.3} 2. Ke6 {-6.53/12 0.3} Ke4
{+51.32/18 0.3} 3. Kd6 {-6.84/12 0.1} Kf4 {+52.12/18 0.3} 4. Ke7
{-6.85/13 0.1} Kg5 {+52.99/18 0.3} 5. Ke6 {-7.18/12 0.2} Kxh6
{+53.79/19 0.1} 6. Kf7 {-7.23/13 0.2} Bf5 {+49.96/15 0.2} 7. Ke7
{-7.31/12 0.2} Kg5 8. Kd6 {-7.35/13 0.3} Nf4 9. Ke5 {-7.32/12 0.3} Ne2
{+49.96/15 0.1} 10. Kd6 {-7.32/12 0.2} Kf6 {+49.96/15 0.3} 11. Kd5
{-7.37/13 0.1} Bc2 {+49.96/15 0.2} 12. Kc4 {-7.38/11 0.2} Ke5
{+49.96/15 0.1} 13. Kc5 {-7.42/12 0.3} Nc3 {+50.06/15 0.2} 14. Kc4
{-7.42/13 0.3} Nd5 {+50.19/15 0.2} 15. Kc5 {-7.42/12 0.3} Bb3
{+50.11/15 0.1} 16. Kc6 {-7.46/12 0.2} Nf4 {+50.09/15 0.2} 17. Kb5
{-7.45/12 0.3} Nd3 {+50.14/15 0.2} 18. Kc6 {-7.53/12 0.1} Ba4+
{+50.16/15 0.2} 19. Kb6 {-7.57/14 0.1} Kd6 {+50.56/14 0.2} 20. Ka5
{-7.64/14 0.1} Bc6 {+50.21/15 0.2} 21. Kb6 {-7.66/14 0.1} Nb2
{+50.16/15 0.3} 22. Ka5 {-7.59/11 0.2} Kc5 {+50.16/15 0.1} 23. Ka6
{-7.64/12 0.1} Bf3 {+50.24/15 0.2} 24. Ka7 {-7.64/12 0.2} Kc6
{+50.24/15 0.1} 25. Kb8 {-99.84/13 0.4} Bg4 {+50.24/15 0.2} 26. Ka7
{-99.86/14 0.1} Nc4 {+99.86/16 0.2} 27. Kb8 {-99.88/16 0.1} Kb6
{+99.88/16 0.2} 28. Ka8 {-99.90/36 0.1} Nd6 {+99.90/17 0.2} 29. Kb8
{-99.92/39 0.1} Bc8 {+99.92/19 0.2} 30. Ka8 {-99.94/4 0.1} Kc7
{+99.94/38 0.1} 31. Ka7 {-99.96/38 0.1} Nb5+ {+99.96/64 0.2} 32. Ka8
{-99.98/37 0.1} Bb7# {+99.98/64 0.1}
{Xboard adjudication: Checkmate} 0-1

[Event "Computer Chess Game"]
[Site "DANIEL-PC"]
[Date "2014.02.02"]
[Round "-"]
[White "Fruit 2.1"]
[Black "Scorpio_09:52:46"]
[Result "0-1"]
[TimeControl "40/10"]
[FEN "8/8/4K2P/7n/3k4/8/8/1b6 w - - 12 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . .
. . . . K . . P
. . . . . . . n
. . . k . . . .
. . . . . . . .
. . . . . . . .
. b . . . . . .
white to play
--------------}
1. Ke7 {-6.45/11} Bh7 {+50.52/15 0.3} 2. Ke6 {-6.53/12 0.3} Ke4
{+51.32/18 0.4} 3. Kd6 {-6.84/13 0.1} Kf5 {+52.12/18 0.2} 4. Kd5
{-6.97/11 0.3} Kg6 {+52.92/19 0.1} 5. Kc4 {-7.14/12 0.2} Kxh6
{+53.75/19 0.2} 6. Kd5 {-7.23/13 0.1} Ng3 {+49.94/15 0.2} 7. Kd4
{-7.23/11 0.2} Be4 {+49.54/16 0.1} 8. Ke3 {-7.31/12 0.2} Kg5
{+49.54/16 0.2} 9. Kd4 {-7.36/13 0.1} Kf4 {+49.54/15 0.1} 10. Kc4
{-7.39/12 0.3} Nf5 {+49.54/15 0.1} 11. Kb4 {-7.41/12 0.2} Ne3
{+49.96/15 0.2} 12. Kb5 {-7.42/12 0.2} Ke5 {+50.51/14 0.3} 13. Kc5
{-7.46/12 0.1} Bd5 {+50.46/14 0.3} 14. Kb5 {-7.46/13 0.1} Kd4
{+50.51/14 0.3} 15. Kb6 {-7.45/12 0.1} Kc4 {+50.91/13 0.2} 16. Kc7
{-7.45/12 0.3} Nf5 {+50.14/15 0.1} 17. Kd7 {-7.55/13 0.2} Bf7 18. Kc6
{-7.51/11 0.3} Be8+ {+50.19/15 0.2} 19. Kb6 {-7.55/13 0.3} Kd5
{+49.79/16 0.1} 20. Kc7 {-7.59/12 0.2} Ng7 {+49.76/16 0.1} 21. Kd8
{-7.55/13 0.3} Kd6 {+49.76/16 0.3} 22. Kc8 {-7.59/14 0.1} Bb5
{+49.76/16 0.2} 23. Kb7 {-7.64/13 0.2} Bd7 {+49.76/16 0.2} 24. Kb6
{-7.64/13 0.3} Nf5 {+49.76/16 0.3} 25. Kb7 {-7.69/13 0.3} Ne3
{+50.24/15 0.2} 26. Kb6 {-7.72/13 0.1} Nc4+ {+50.24/15 0.3} 27. Kb7
{-99.86/14 0.1} Bh3 {+50.24/15 0.3} 28. Ka6 {-99.88/13 0.2} Kc6
{+99.80/16 0.2} 29. Ka7 {-99.90/29 0.1} Bc8 {+99.90/17 0.2} 30. Kb8
{-99.92/42 0.1} Nd6 {+99.92/22 0.1} 31. Ka7 {-99.94/39 0.1} Kc7
{+99.94/31 0.1} 32. Ka8 {-99.96/36 0.1} Bb7+ {+99.96/64 0.2} 33. Ka7
{-99.98/37 0.1} Nc8# {+99.98/64 0.1}
{Xboard adjudication: Checkmate} 0-1

Daniel Shawul · Post by **Daniel Shawul** » Sun Feb 02, 2014 4:10 pm

And it wins with this too. With lower time controls Fruit hangs before it is about to get mated. So I wonder how I can't reproduce one non-win so far.

[Event "Computer Chess Game"]
[Site "DANIEL-PC"]
[Date "2014.02.02"]
[Round "-"]
[White "Fruit 2.1"]
[Black "Scorpio_09:52:46"]
[Result "0-1"]
[TimeControl "40/5"]
[FEN "8/8/4K2P/7n/3k4/8/8/1b6 w - - 12 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . .
. . . . K . . P
. . . . . . . n
. . . k . . . .
. . . . . . . .
. . . . . . . .
. b . . . . . .
white to play
--------------}
1. Ke7 {-6.22/10} Bh7 {+50.52/13 0.2} 2. Ke6 {-6.33/10 0.1} Ke4
{+51.32/12 0.1} 3. Ke7 {-6.53/11 0.1} Kf5 {+52.14/13 0.1} 4. Kd7
{-6.85/12 0.1} Kg5 {+52.99/13 0.1} 5. Kc7 {-7.05/11 0.2} Kxh6
{+53.90/16 0.1} 6. Kd7 {-7.23/12 0.1} Be4 {+49.96/15 0.1} 7. Kd6
{-7.26/11 0.1} Bg2 {+50.34/14 0.1} 8. Ke5 {-7.28/11 0.1} Kg5
{+49.94/15 0.1} 9. Kd4 {-7.32/12 0.1} Kf4 {+49.94/15 0.1} 10. Kc4
{-7.37/12 0.1} Bb7 {+50.34/14 0.1} 11. Kd4 {-7.37/11 0.1} Ng3
{+49.94/15 0.1} 12. Kc3 {-7.37/10 0.1} Ba8 {+49.94/15 0.1} 13. Kd4
{-7.37/10 0.1} Ne2+ {+50.34/14 0.1} 14. Kd3 {-7.37/11 0.1} Bf3
{+49.96/15 0.1} 15. Kc4 {-7.39/12 0.1} Ke5 {+50.34/14 0.1} 16. Kd3
{-7.40/10 0.1} Nd4 {+50.36/14 0.1} 17. Kd2 {-7.37/10 0.1} Be4
{+50.76/13 0.1} 18. Kc3 {-7.41/10 0.1} Bg6 {+49.96/15 0.1} 19. Kc4
{-7.40/10 0.1} Bf5 {+49.96/15 0.1} 20. Kc5 {-7.41/9 0.1} Nc2
{+50.74/12 0.1} 21. Kc4 {-7.41/10 0.1} Ne1 22. Kb5 {-7.40/10 0.1} Be6
{+50.86/13 0.1} 23. Ka4 {-7.41/11 0.1} Nc2 {+50.94/13 0.1} 24. Kb5
{-7.51/11 0.1} Kd5 {+50.94/13 0.1} 25. Kb6 {-7.50/10 0.1} Bf7
{+50.94/13 0.1} 26. Kc7 {-7.46/10 0.1} Nd4 {+50.96/13 0.1} 27. Kd7
{-7.51/11 0.1} Nc6 {+50.16/15 0.1} 28. Kc7 {-7.55/11 0.1} Ne5
{+50.19/15 0.1} 29. Kb6 {-7.55/10 0.1} Be8 {+50.56/14 0.1} 30. Kc7
{-7.55/9 0.1} Bc6 {+50.99/12 0.1} 31. Kd8 {-7.64/11 0.1} Kd6
{+50.59/14 0.1} 32. Kc8 {-7.69/12 0.1} Nc4 {+50.56/14 0.1} 33. Kd8
{-7.64/9 0.1} Nb2 {+50.59/14 0.1} 34. Kc8 {-7.53/4 0.1} Na4 {+50.24/15 0.1}
35. Kd8 {-6.93/1 0.1} Nc5 {+50.24/15 0.1} 36. Kc8 {-7.43/1 0.1} Bd7+
{+50.24/15 0.1} 37. Kb8 {-7.38/1 0.1} Kc6 {+99.90/16 0.1} 38. Ka7
{-7.13/1 0.1} Kc7 {+99.92/18 0.1} 39. Ka8 {-7.42/1 0.1} Kb6 {+99.94/36 0.2}
40. Kb8 {-7.19/1 0.1} Na6+ {+99.96/64 0.1} 41. Ka8 {-99.98/4 0.1} Bc6#
{+99.98/64 0.1}
{Xboard adjudication: Checkmate} 0-1

[Event "Computer Chess Game"]
[Site "DANIEL-PC"]
[Date "2014.02.02"]
[Round "-"]
[White "Fruit 2.1"]
[Black "Scorpio_09:52:46"]
[Result "0-1"]
[TimeControl "40/5"]
[FEN "8/8/4K2P/7n/3k4/8/8/1b6 w - - 12 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . .
. . . . K . . P
. . . . . . . n
. . . k . . . .
. . . . . . . .
. . . . . . . .
. b . . . . . .
white to play
--------------}
1. Ke7 {-6.22/10} Bh7 {+50.52/12 0.1} 2. Ke6 {-6.33/11 0.2} Ke4
{+51.32/13 0.1} 3. Ke7 {-6.47/11 0.1} Kf4 {+52.14/13 0.1} 4. Kd6
{-6.84/11 0.1} Kg5 {+52.94/15 0.2} 5. Kd5 {-6.97/11 0.1} Kxh6
{+53.77/13 0.1} 6. Kd4 {-7.21/12 0.1} Kg5 7. Ke5 {-7.28/12 0.1} Nf4
{+50.34/14 0.1} 8. Kd4 {-7.32/12 0.1} Kf5 {+50.34/14 0.1} 9. Kc4
{-7.37/10 0.1} Nh5 {+50.36/14 0.1} 10. Kd4 {-7.32/11 0.1} Ng7
{+49.54/16 0.1} 11. Kd5 {-7.36/11 0.1} Bg8+ {+49.56/16 0.1} 12. Kd6
{-7.37/12 0.1} Ke4 {+50.84/12 0.1} 13. Ke7 {-7.43/11 0.1} Nf5+
{+50.84/13 0.1} 14. Kf6 {-7.46/11 0.2} Kf4 {+50.86/13 0.1} 15. Kg6
{-7.49/12 0.1} Ke5 {+50.46/14 0.1} 16. Kh5 {-7.51/10 0.1} Bf7+ 17. Kg4
{-7.52/11 0.1} Ne7 {+50.84/13 0.1} 18. Kf3 {-7.37/9 0.1} Bg6
{+50.86/13 0.1} 19. Kf2 {-7.39/10 0.1} Be4 {+50.89/12 0.1} 20. Ke2
{-7.41/11 0.1} Kd4 {+50.09/15 0.1} 21. Kf2 {-7.46/12 0.1} Nd5
{+51.29/11 0.1} 22. Kg3 {-7.45/10 0.1} Nf6 {+50.94/12 0.1} 23. Kf4
{-7.42/10 0.1} Ng8 {+50.89/12 0.1} 24. Kg5 {-7.46/11 0.1} Ke5
{+50.54/14 0.1} 25. Kg4 {-7.46/11 0.1} Nh6+ {+50.46/14 0.1} 26. Kg3
{-7.48/12 0.1} Nf5+ 27. Kf2 {-7.48/11 0.1} Kd4 {+50.89/12 0.1} 28. Ke2
{-7.48/13 0.1} Kc3 {+50.94/13 0.1} 29. Ke1 {-7.48/11 0.1} Bf3
{+50.94/12 0.1} 30. Kf2 {-7.48/9 0.1} Bd1 {+50.16/15 0.1} 31. Kg2
{-7.45/9 0.1} Kd2 {+51.36/11 0.1} 32. Kf2 {-7.55/10 0.1} Be2
{+50.57/14 0.1} 33. Kg1 {-7.57/12 0.1} Ke3 {+52.91/6 0.1} 34. Kg2
{-7.55/12 0.1} Bb5 {+50.59/14 0.2} 35. Kg1 {-7.55/9 0.1} Kf3
{+50.59/14 0.2} 36. Kh2 {-99.90/25 0.1} Bf1 {+99.90/15 0.1} 37. Kg1
{-99.92/41 0.1} Ne3 {+99.92/19 0.1} 38. Kh2 {-99.94/39 0.1} Kf2
{+99.94/19 0.2} 39. Kh1 {-99.96/38 0.1} Bg2+ {+99.96/18 0.1} 40. Kh2
{-99.98/36 0.1} Nf1# {+99.98/64 0.1}
{Xboard adjudication: Checkmate} 0-1

Max · Post by **Max** » Sun Feb 02, 2014 4:25 pm

Daniel Shawul wrote:Honestly I don't know how he managed to reproduce an non-winning KBNK game because I can't with any decent time control! I have my theories though

Try it against an engine who knows about the right corner!

An example against Arasan 16.3 with Gaviotabases at 30 moves in 60 seconds.

Code: Select all

&#91;Event "Computer Chess Game"&#93;
&#91;Site "Home"&#93;
&#91;Date "2014.02.02"&#93;
&#91;Round "-"&#93;
&#91;White "Arasan 16.3 &#40;Gaviota&#41;"&#93;
&#91;Black "Scorpio 2014"&#93;
&#91;Result "1/2-1/2"&#93;
&#91;TimeControl "30/60"&#93;
&#91;FEN "8/7b/7n/4k3/2K5/8/8/8 w - - 0 1"&#93;
&#91;SetUp "1"&#93;

&#123;--------------
. . . . . . . .
. . . . . . . b
. . . . . . . n
. . . . k . . .
. . K . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
white to play
--------------&#125;
1. Kc3 &#123;-99,48/3&#125; Ng4 &#123;+46,41/20 4&#125; 2. Kc4 &#123;-99,50/3 0,3&#125; Bg8+
&#123;+47,14/19 1,8&#125; 3. Kd3 &#123;-99,48/3 0,2&#125; Bf7 &#123;+46,56/20 1,8&#125; 4. Kc3
&#123;-99,50/3 0,3&#125; Nf6 &#123;+47,14/19 1,8&#125; 5. Kc2 &#123;-99,50/3 0,3&#125; Kd4
&#123;+47,01/19 1,8&#125; 6. Kb2 &#123;-99,52/3 0,2&#125; Kd3 &#123;+46,99/19 1,8&#125; 7. Kb1
&#123;-99,54/3 0,2&#125; Nd7 &#123;+46,97/19 1,7&#125; 8. Kb2 &#123;-99,54/3 0,3&#125; Nc5
&#123;+47,01/19 1,5&#125; 9. Ka3 &#123;-99,56/3 0,3&#125; Be8 &#123;+47,09/19 1,7&#125; 10. Kb4
&#123;-99,54/3 0,3&#125; Kd4 &#123;+46,59/20 1,6&#125; 11. Ka3 &#123;-99,56/3 0,2&#125; Kc3
&#123;+46,71/20 1,9&#125; 12. Ka2 &#123;-6,62/2 0,3&#125; Ne6 &#123;+46,56/20 1,9&#125; 13. Ka1
&#123;-99,58/3 0,3&#125; Nf4 &#123;+46,71/20 1,7&#125; 14. Kb1 &#123;-99,56/3 0,3&#125; Bb5
&#123;+46,56/20 1,7&#125; 15. Ka1 &#123;-99,56/3 0,3&#125; Bc4 &#123;+46,14/21 1,9&#125; 16. Kb1
&#123;-6,70/2 0,3&#125; Ne6 &#123;+46,19/21 1,9&#125; 17. Ka1 &#123;-99,60/3 0,3&#125; Bd3
&#123;+45,96/21 1,9&#125; 18. Ka2 &#123;-6,66/2 0,3&#125; Nd8 &#123;+46,62/21 1,9&#125; 19. Ka1
&#123;-99,56/3 0,3&#125; Be4 &#123;+45,84/23 2,0&#125; 20. Ka2 &#123;-6,38/2 0,3&#125; Nc6
&#123;+45,87/22 1,9&#125; 21. Ka1 &#123;-99,60/3 0,3&#125; Ne5 &#123;+46,90/21 1,7&#125; 22. Ka2
&#123;-6,70/2 0,3&#125; Nc4 &#123;+46,19/21 1,7&#125; 23. Ka1 &#123;-6,62/2 0,3&#125; Ne3 &#123;+46,18/21 2,1&#125;
24. Ka2 &#123;-6,74/2 0,3&#125; Kd3 &#123;+46,14/20 2,1&#125; 25. Kb2 &#123;-99,56/3 0,2&#125; Bd5
&#123;+46,15/20 2,1&#125; 26. Kb1 &#123;-99,58/3 0,3&#125; Ke4 &#123;+46,13/20 2,0&#125; 27. Ka1
&#123;-99,56/3 0,2&#125; Nc2+ &#123;+46,60/20 1,7&#125; 28. Kb1 &#123;-99,58/3 0,2&#125; Nb4
&#123;+45,80/21 1,9&#125; 29. Ka1 &#123;-99,56/3 0,3&#125; Nd3 &#123;+46,06/22 2,5&#125; 30. Kb1
&#123;-6,70/2 0,3&#125; Nc5 &#123;+46,01/21 2,8&#125; 31. Kc2 &#123;-99,54/3 0,2&#125; Na4
&#123;+46,13/20 1,9&#125; 32. Kc1 &#123;-99,54/3 0,3&#125; Kd3 &#123;+46,41/20 1,6&#125; 33. Kb1
&#123;-99,56/3 0,3&#125; Nc5 &#123;+45,71/21 1,9&#125; 34. Ka1 &#123;-99,58/3 0,3&#125; Ne4
&#123;+45,92/21 1,9&#125; 35. Kb1 &#123;-99,56/3 0,3&#125; Kd4 &#123;+45,79/21 1,9&#125; 36. Kb2
&#123;-99,54/3 0,2&#125; Ng5 &#123;+46,13/20 1,9&#125; 37. Kc2 &#123;-99,52/3 0,3&#125; Ke5
&#123;+46,13/20 1,7&#125; 38. Kb1 &#123;-99,50/3 0,2&#125; Be6 &#123;+46,18/20 2,0&#125; 39. Ka1
&#123;-99,50/3 0,3&#125; Kf6 &#123;+46,76/20 2,0&#125; 40. Kb1 &#123;-99,48/3 0,3&#125; Bf5+
&#123;+46,11/20 2,0&#125; 41. Ka1 &#123;-99,48/3 0,2&#125; Nf3 &#123;+46,66/19 1,6&#125; 42. Kb2
&#123;-99,48/3 0,3&#125; Ne1 &#123;+46,08/18 2,0&#125; 43. Kb3 &#123;-99,48/3 0,3&#125; Ke6
&#123;+45,31/18 2,0&#125; 44. Ka2 &#123;-99,50/3 0,2&#125; Nf3 &#123;+45,32/19 2,0&#125; 45. Kb2
&#123;-99,50/3 0,3&#125; Be4 &#123;+44,58/20 2,0&#125; 46. Ka3 &#123;-99,50/3 0,3&#125; Kd5
&#123;-0,02/26 1,6&#125; 47. Kb4 &#123;-99,52/3 0,2&#125; Nh4 &#123;-0,02/64 0,6&#125; 48. Kb3
&#123;-99,50/3 0,2&#125; Nf3 &#123;-0,02/64 0,1&#125; 49. Kb4 &#123;-99,52/3 0,2&#125; Nh4 &#123;-0,02/64 0,2&#125;
50. Kb3 &#123;+0,00/3 0,2&#125; Nf3 &#123;-0,02/64 0,2&#125;
&#123;Draw claim&#58; 50-move rule&#125; 1/2-1/2

-Max

syzygy · Post by **syzygy** » Sun Feb 02, 2014 4:26 pm

Daniel Shawul wrote:
syzygy wrote:
hgm wrote:
Max wrote:I didn't realize that Scorpio needs some more search time, even with egbbs.
The problem is that WDL info offers zero help in end-games like KBNK, where every position where white does not obviously hang a piece will be won. And knowing that you must not blunder away a piece doesn't help you finding the mate, or any simple 3-ply search without eval would already be able to do it. So KBNK remains exactly as difficult with or without EGBB.
The use of bitbases likely makes it harder to win KBNK. If they are probed during the searched, it is overhead in return for no information. If they are used to prune branches, it is likely the search won't be able to make progress where without pruning it could. When KBNK is on the board, bitbases should be disabled.
No, speak for your bitbases. My Scorpio bitbases handle them fine. Honestly I don't know how he managed to reproduce an non-winning KBNK game because I can't with any decent time control! I have my theories though

So how do Scorpio bitbases help with KBNK where all positions are won?

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.

Daniel Shawul · Post by **Daniel Shawul** » Sun Feb 02, 2014 4:45 pm

So how do Scorpio bitbases help with KBNK where all positions are won?

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.

Confused much? My dll doesn't search for the engine, the engine does it but it replaces its eval pure KBNK mate table. How can that be worse than an engine could implement?? Yeah I know you an 100Gb solution for it but stop spreading lies about my EGBBs.

syzygy · Post by **syzygy** » Sun Feb 02, 2014 5:04 pm

Daniel Shawul wrote:
So how do Scorpio bitbases help with KBNK where all positions are won?

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.
Confused much? My dll doesn't search for the engine, the engine does it but it replaces its eval pure KBNK mate table.

Is that any different from what I wrote ("your dll returns heuristic values that are supposed to help")?

How can that be worse than an engine could implement??

Again, the bitbase information for KBNK ("everything wins") can't possibly help. You can of course attempt to replace the engine's evaluation for specific endgames, thinking you can do a better job at it than the engine's author can, but is that the purpose of bitbases?

Daniel Shawul · Post by **Daniel Shawul** » Sun Feb 02, 2014 5:22 pm

syzygy wrote:
Daniel Shawul wrote:
So how do Scorpio bitbases help with KBNK where all positions are won?

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.
Confused much? My dll doesn't search for the engine, the engine does it but it replaces its eval pure KBNK mate table.
Is that any different that what I wrote ("your dll returns heuristic values that are supposed to help")?

How can that be worse than an engine could implement??
Again, the bitbase information for KBNK ("everything wins") can't possibly help. You can of course attempt to replace the engine's evaluation for specific endgames, thinking you can do a better job at it than the engine's author can, but is that the purpose of bitbases?

Now you are getting philospophical. You said the engine searches better than the dll, which I pointed out that dll produces evaluations only. For KBNK mating you only need one table and nothing else. Infact mobility and other irrelevant evals could elongate finiding the mate. The dll provides you just that so it CAN'T be any worse. Whatever else the engine does, like prunings/reductions is upto the engines. HG in the other thread clamined he can do it in 3 plies, but I know my scorpio's pruning and reductions probably won't allow that. So your statement is simply wrong, you can do as well using search-dll_eval than a native search-eval combo. The 'philosophical argument' you raised is moot, as one can get WDL scores from >0, <0, =0. No one who uses Scorpio bitbases does that however, because it doesn't make sense to go backwards.

hgm · Post by **hgm** » Sun Feb 02, 2014 5:38 pm

Note the 3-ply estimate was only for not blundering away material. With two ply you can unwittingly place B and N such that the bare King can step in between to fork them.

[d]8/8/8/3N4/5B2/3k4/8/3K4 b
Ply 2: Ke4! Ply 3: stand pat.

So a 3-ply search tells you exactly the same as what a WDL bitbase for KBNK could tell you. I have no idea if a heuristic recognized could be made that is good enough to always win at 3 ply. (Without using large tables. With a DTM table you obviously only need a 1-ply search to play perfectly.)

Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

40 moves in 5 seconds

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)

Re: Scorpio & egbb issue (OSX)