Scan's games at the 2015 Olympiad

[Rein Halbersma]

Scan - Moby Dam after 26. 42-38

Here Moby Dam made a tactical error in an already difficult position: 26... 11-17? 27. 31-27 6-11 28. 38-33 18-23 29. 33x22 17x28 30. 37-31 11-17 31. 27-22 28-32 32. 22x11 16x7


and although the position look superficially quiet, the outpost on <32> is doomed and Scan would flawlessly find the win here. If Moby Dam was missing eval patterns for such outposts on the 7th row, then it's understandable that its search could not compensate for this on the 26th move (the actual loss of the piece happened on move 42, so 32 plies away and that still requires another deep search to discover).

In the first diagram, a draw was possible with 26... 12-17 27. 31-27 2-8 28. 38-33 8-13 29. 33x22 17x28


and here white cannot immediately play 30. 37-31? because of the threat 30... 18-22! 31. 27x20 15x33. After 30. 40-35 11-17! black can easily defend his outpost on <28> (31. 37-31 17-22! =). Moby Dam must have seen this tactic, but its eval must have judged the 2nd diagram better than the 3rd. Perhaps Harm can share his search log for the 26th move.

[Rein Halbersma]

JDraughts-Scan after 22. 39x28

Here Scan made a blunder with 22... 1-7? and allowed the combination 23. 27-22! 18x27 24. 36-31 27x36 25. 37-31 36x27 26. 28-23 19x37 27. 42x2


With so few pieces remaining, a king is at least worth 2.5 men, so regardless of positional compensation and king freedom, this position should have been scored negatively by Scan (easily fixable of course). Luckily enough for black, it appears that white could not win the endgame. The game continued with 27... 14-19 28. 48-42 9-14 29. 38-33 21-26

and after 30. 33-29 the game eventually got drawn because black could break through to the promotion line. But also preventing the breakthrough with 30. 43-38 26-31 31. 42-37 31x42 32. 38x47 19-24 33. 34-30 13-19 34. 2x13 19x8 35. 30x10 15x4 would have lead to a drawn position.

[Rein Halbersma]

TD King-Scan after 49... 22-27

The moves 50. 41-36, 50. 50-44 and 50. 50-45 are all a draw here. But also the game continuation is still good enough and the decisive mistake happened one move later:
50. 39-34 27-31 51. 28-23? with a losing endgame. According to Kingsrow a draw could still be obtained with 51. 34-30! 31-36 52. 30x19 36x47 53. 19-14 9-13 54. 14x25 47-36 55. 28-23

and incredibly, the black pieces are just badly enough coordinated that they cannot prevent white from getting to the promotion line.

[Fabien Letouzey]

Thanks Rein!

I guess the other games were lopsided or uneventful. I was impressed by Sjende Blyn which played very solidly.

JDraughts also had a very good tournament; I'm looking forward to seeing its games.

[Rein Halbersma]

Thanks Rein!

I guess the other games were lopsided or uneventful. I was impressed by Sjende Blyn which played very solidly.

JDraughts also had a very good tournament; I'm looking forward to seeing its games.

Yes, the other games against the bottom half of the table were very lopsided. Scan also positionally dominated the games against Damage and Sjende Blyn, but no winning advantage as far as I could see. The game against Dragon was an equal draw IMO.

Because the top 5 drew each other, and won almost everything against the bottom 5, the games against Moby Dam and TD King were crucial wins for Scan. These programs (and Dam 2.2 in the past) are usually quite hard to beat, and have also won tournaments, but often they ended up just below the top, so they often "decide" who wins the tournament by losing against the ultimate winner and drawing against the other top programs.

With so few games, you need a little bit of luck that a program that is down positionally, eventually runs into a variation that turns out to be lost, and this happened twice. Of course, it was Scan's credit to put those programs under pressure in the first place. I also think that TD King would have drawn with 7 piece dbs and Damage might have lost without them. So if Scan would have 7 or 8 piece dbs, then I think it would add quite a bit of strength as well.

[Harm Jetten]

In the lower iterations for move 26... the best move changes between 2-8, 3-8 and 11-17.
Below is the iteration for depth 18 where Moby Dam discovers it's in trouble. Time ran out during the re-search of 11-17. The value of a man is 10000000.

Code: Select all

[...]
17.complete, 19770 ms, score=-1872 move=3-8 
18.0 score=-9895560 pv=3-8 38-32 8-13 32x23 18x29 34x23 13-19 40-34 19x28 31-27 15-20 36-31 14-19 30-25 20-24 41-36 11-17 27-22 24-30 22x33 
     .   b   .   b   .
   b   .   .   .   .
     .   b   .   .   .
   b   b   .   b   .
     .   .   .   .   w
   .   .   .   .   b
     w   .   w   w   .
   w   w   .   w   .
     .   .   .   .   .
   .   .   w   w   .   (b to move)
18.1 merit=-9893248 pv=11-17 31-26 18-22 38-33 
     .   b   b   b   .
   b   .   .   .   .
     .   b   .   b   b
   b   b   .   .   .
     .   b   .   .   .
   w   .   b   .   w
     .   .   w   w   .
   w   w   .   w   w
     w   .   .   .   .
   .   .   w   w   .   (b to move)
18.1 re-search
reached depth=18 move=1
nodes total=716044288 nonleaf=338897115 leaf=377147173
moves calls=677882390 generated=1644957787
tt probes=218159406 hits=37854048 bestmoves=32803455
etc tests=1140535 tthits=514395 cuts=186824
egdb err=0 2pc=0 3pc=10 4pc=267 5pc=2185 6pc=15818
evals=338779222 score=-9893248
move=26... 11-17 
     .   b   b   b   .
   b   .   .   .   .
     .   b   .   b   b
   b   b   b   .   .
     .   .   .   .   .
   .   .   b   .   w
     w   .   .   w   .
   w   w   w   w   w
     w   .   .   .   .
   .   .   w   w   .   (w to move)
engine used 62118 ms, leaving 766143 ms for 49 moves

[Rein Halbersma]

In the lower iterations for move 26... the best move changes between 2-8, 3-8 and 11-17.
Below is the iteration for depth 18 where Moby Dam discovers it's in trouble. Time ran out during the re-search of 11-17. The value of a man is 10000000.

Thanks for posting this! Very unlucky to stop searching when rescue was perhaps just around the corner. At which search depth and after how many seconds does Moby Dam find 26... 12-17?

BTW, chess engine Crafty used to have the following comment in its time control code when failing low at the root

Code: Select all

 *   We are in trouble at the root.  We have now used the   *
 *   allocated time limit, yet the score for the current    *
 *   iteration is still worse than the score for the        *
 *   previous iteration.  We will continue to search until  *
 *   we use 6x the normal target time in an effort to avoid *
 *   playing a move that might end up losing the game.      *

[Harm Jetten]

When I try to reproduce the search from this position, it takes much more time and gives different best moves at various depths. It's missing TT information built up during previous moves and pondering.
You can find the time allocation in function set_budget() in search.c.

[Harm Jetten]

Jaap Bus analyzed the Scan - Moby Dam game at his blog, see http://endgamesandmore.blogspot.de/2015 ... puter.html .
His conclusion is that 26... 11-17 was not fatal, Moby Dam could still have reached a draw.
At move 33... instead of 12-18 it should have played 15-20 and then 4-10 or vice versa.



I let the current version of Moby Dam loose on this position and it took 1.5 hours to switch from 12-18 to 4-10.
Much too deep to find under tournament time controls, alas.

[Fabien Letouzey]

I let the current version of Moby Dam loose on this position and it took 1.5 hours to switch from 12-18 to 4-10.
Much too deep to find under tournament time controls, alas.

There is something deep about this move. Scan needs about 30s to find it. It didn't see it on the previous ply during the tournament though.

[BertTuyt]

For Damage the position is also not straightforward.
Im a few days in Netherlands, and was able to test the position with my faster 8 core machine (and a 7P DB).

At ply 28 after 280 seconds it switched from 12-18 to 15-20.
The score -0.10 is somewhat better for White, but at this stage looks quite drawish.

Bert

[Jelle Wiersma]

At one moment 12-18 looks ok, and then it suddenly becomes bad.
The current version of Sjende Blyn (6p database) requires 139 seconds to switch from 12-18 to 15-20.