Calibrate settings for certain endgames

[Nordlandia]

Is Komodo's default settings optimal for the following position?

[d]5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1

Perhaps some minor tweaks here and there make K10 perform well or better in that overanalyzed epic endgame.

Reduction = 50.
Contempt = 0
Dynamism = 80 or 100, (110)
Selectivity = 197 (+50%)
-----
Everything else default, - Some feedback will be appreciated.

[AdminX]

That be Spassky - Fischer if memory serves me correctly.

[Nordlandia]

That be Spassky - Fischer if memory serves me correctly.

Correct!



Notice at 1:10 Fischer snatch the poisoned pawn at h2.

https://youtu.be/ZQCAL6E-ag0

[peter]

Hi Jon!

Problem here is, komodo evalutates the drawn 7some positions as White wins, so I don't think you can get more than cosmetic changes of the evals by changing parameters. More or less selectivity won't let komodo "see" the fortresses sooner, not even with full Syzygys. I tried some Forward- Backward of positions like the one at the end of that line,

5k2/pp4pp/4pp2/1P6/8/P2KP3/5PPb/2B5 w - - 0 30

30. g3 h5 31. Ke2 Ke7 32. Kf3 Kd6 33. b6 a6 34. Kg2 Bxg3 35. Kxg3 Kd5 36. Bb2 e5 37. f4 exf4+ 38. exf4 Ke4 39. a4 h4+ 40. Kxh4 Kxf4 41. Bc1+ Ke5 42. Kh5 Kd5 43. Kg6 Kc5 44. a5 Kc6 45. Kxg7 f5 46. Kf6 Kd6 47. Kxf5

[d]8/1p6/pP1k4/P4K2/8/8/8/2B5 b - - 0 1

Analysis by Komodo 10 64-bit:

47...Ke7 48.Bh6 Kd6 49.Bf4+ Ke7 50.Ke5 Kd7 51.Be3 Kc8 52.Kd5 Kb8 53.Bd4 Ka8 54.Bf2 Kb8 55.Be3 Kc8 56.Bd4 Kb8 57.Be5+ Ka8 58.Kd6 Kb8 59.Bb2 Kc8 60.Bc3 Kb8 61.Bf6 Kc8 62.Ke6 Kb8 63.Bd4 Ka8 64.Ke5 Kb8 65.Ba1 Kc8 66.Bb2 Kd7 67.Bc3 Kd8 68.Kd6 Kc8 69.Bd4 Kb8 70.Be5 Kc8 71.Bf6 Kb8 72.Kd7 Ka8 73.Bc3 Kb8 74.Bb2 Ka8 75.Kd6 Kb8
+- (3.80) Depth: 52 00:00:16 705MN, tb=83713

And here is your setting:

8/1p6/pP1k4/P4K2/8/8/8/2B5 b - - 0 1

Analysis by Komodo 10 64-bit:

47...Kd7 48.Ke5 Kd8 49.Bg5+ Kc8 50.Be7 Kb8 51.Kd6 Kc8 52.Bf6 Kb8 53.Kd7 Ka8 54.Bg5 Kb8 55.Bf4+ Ka8 56.Kd6 Kb8 57.Ke6+ Kc8 58.Be3 Kb8 59.Ke7 Kc8 60.Bd4 Kb8 61.Kd7 Ka8 62.Bf6 Kb8 63.Be5+ Ka8 64.Bc3 Kb8 65.Ke6 Kc8 66.Ke7 Kb8 67.Bf6 Ka8 68.Bb2 Kb8 69.Kd7 Ka8 70.Kd6 Kb8 71.Bf6 Kc8 72.Bc3 Kb8 73.Bb2 Kc8 74.Ba3 Kb8 75.Kd7 Ka8 76.Ke6 Kb8 77.Bc5 Kc8 78.Be7 Kb8
+- (3.43) Depth: 59 00:00:41 2556MN

K10 needs many more moves near to the 50 move- boundary to change from evals even up to 250 (sometimes, it wasn't reproducable) to 0.00 and I don't see much differences as for keeping that in hash at Backward with different settings.
So you won't be able to get relevant changes as for the lines in output in standalone computing at the position after 30.g3 at all probably, even if you can lower the evals to whatever you want by changing the right parameters.
Here I simply lowered King's Safety, Dynamism and Magnifiying down to 20 each, Contempt 0:

8/1p6/pP1k4/P4K2/8/8/8/2B5 b - - 0 1

Analysis by Komodo 10 64-bit:

47...Ke7 48.Bf4 Kd7 49.Kf6 Kd8 50.Be5 Kc8 51.Ke6 Kd8 52.Bf6+ Kc8 53.Bc3 Kb8 54.Be1 Kc8 55.Kd6 Kb8 56.Bb4 Kc8 57.Ke6 Kb8 58.Kd5 Kc8 59.Bc3 Kb8 60.Kd6 Kc8 61.Bd2 Kb8 62.Bf4 Kc8 63.Ke6 Kd8 64.Bd6 Kc8 65.Bg3 Kd8 66.Kf6 Kd7 67.Kf5 Kd8 68.Bh4+ Kd7 69.Kf6 Kc8 70.Ke6 Kb8 71.Kd5 Kc8 72.Be7
+/= (0.58) Depth: 47 00:00:09 417MN, tb=71646

Eval looks much better, doesn't it, but so what? You need as many moves till the engines really "sees" the fortress as with default values.

Raising selectivity and reductions to the maximum didn't help in Forward- Backward neither

[Nordlandia]

Peter Martan: thanks for the long reply.

I have a thery that probing 7-Men Syzygy will add more insight to this endgame, simply because many or majority of the 6-Men positions arising frequently from endgame play is positonal draws, - easily understandable for humans.

It is interesting to find out whether 30. g3 30...Ke7 is enough to draw. Also 30...a6 is occasionally chosen by SF or Komodo.

This endgame is very tricky indeed. There is so many possible sub-variations that engines today can't comprehend which path is the best or worst.

[hgm]

You cannot expect parameter fiddling or search tricks to repair gross misevaluations such as the presence of a fortress. When the fortress would be recognized as such,even a very simple engine would probably evaluate this end-game well.

Unfortunately fortress recognition does not provide measurable Elo, or might even cost some. And Elo is the only thing customers want.

That is what you get when optimizing only an average quantity, like Elo, rather than a minimum quantity (reliability). Imagine you would have a car that would not start at all on 7 random days of the year, but even taking that into account, on average would drive 1% faster than a car that would work every day. You think anyone would buy it? Well, if it were a Chess engine, they would queue up for it...

[peter]

It is interesting to find out whether 30. g3 30...Ke7 is enough to draw. Also 30...a6 is occasionally chosen by SF or Komodo.

This endgame is very tricky indeed. There is so many possible sub-variations that engines today can't comprehend which path is the best or worst.

Here is SF Matefinder after a slow Backward from about 90 moves, where the 50- moves- boundary gets into "sight", the drawn 7some position is reached after 53 moves in this line:

5k2/pp4pp/4pp2/1P6/8/P2KP3/5PPb/2B5 w - - 0 30

30. g3 Ke7 31. a4 h5 32. Ke4 h4 33. Kf3 h3 34. Kg4 Bg1 35. Kxh3 Bxf2 36. Bd2 Kd6 37. Kg2 Bxg3 38. Bb4+ Kd5 39. Kxg3 e5 40. Kg4 Ke4 41. a5 g6 42. Be7 f5+ 43. Kg5 Kd5 44. Kxg6 f4 45. e4+ Kxe4 46. Bc5 a6 47. b6 f3 48. Kg5 Kd5 49. Bg1 Ke6 50. Kg4 Kd5 51. Kxf3 e4+ 52. Kf4 e3 53. Bxe3

After Backward:

[d]5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1

Analysis by Stockfish MateFinder 64 POPCNT:

30...Ke7 31.a4 h5 32.Ke2 h4 33.Kf3 h3 34.Kg4 Lg1 35.Kxh3 Lxf2 36.Ld2 a6 37.b6 g5 38.Kg2 Lxe3 39.Lxe3 g4 40.Kf2 Kd6 41.Ke2 f5 42.Kd3 Kd5 43.Ld2 e5 44.Le3 Kc6 45.Lg1 Kd5 46.Lf2 Kd6 47.Le1 Kc5 48.a5 Kb5 49.Ke3 Kc4 50.Ke2 Kd5 51.Lf2 f4 52.Kd2 Kc6 53.Kd3 Kb5
+/= (0.64) Tiefe: 36/47 00:01:05 2087MN, tb=1708610

[peter]

And the steps to that:

First SF Matefinder with default parameters (study mode off) and full Syzygys at your test position with empty hash (just started before 30.g3 having computed this move shortly too)

[d]5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1

Analysis by Stockfish MateFinder 64 POPCNT:

30...Ke7 31.a4 a6 32.b6 Kd7 33.Ke2 Kc6 34.a5 g5 35.Ld2 h5 36.Kf1 h4 37.Kg2 g4 38.Kxh2 h3 39.Le1 Kd5 40.f3 f5 41.e4+ Kc5 42.Ld2 Kb5 43.e5 Kc4 44.f4 Kd4 45.Le1 Ke4 46.Kh1 Ke3 47.Lb4 Kd3 48.Kg1 Kc4 49.Le1 Kb5 50.Kh1
+/- (0.80) Tiefe: 36/50 00:06:19 11055MN, tb=16306599

Here at that line's end- position, a tbs- draw:

5k2/pp4pp/4pp2/1P6/8/P2KP3/5PPb/2B5 w - - 0 30

30. g3 Ke7 31. a4 h5 32. Ke4 h4 33. Kf3 h3 34. Kg4 Bg1 35. Kxh3 Bxf2 36. Bd2 Kd6 37. Kg2 Bxg3 38. Bb4+ Kd5 39. Kxg3 e5 40. Kg4 Ke4 41. a5 g6 42. Be7 f5+ 43. Kg5 Kd5 44. Kxg6 f4 45. e4+ Kxe4 46. Bc5 a6 47. b6 f3 48. Kg5 Kd5 49. Bg1 Ke6 50. Kg4 Kd5 51. Kxf3 e4+ 52. Kf4 e3 53. Bxe3

[d]8/1p6/pP6/P2k4/5K2/4B3/8/8 b - - 0 1

Analysis by Stockfish MateFinder 64 POPCNT:

53...Kc6 54.Lc5 Kd7 55.Ke5 Kc8 56.Kd5 Kb8 57.Ld4 Kc8 58.Le5 Kd7 59.Kc5 Ke6 60.Ld6 Kd7 61.Lg3 Kc8 62.Le5 Kd7 63.Ld4 Kc8 64.Kd6 Kb8 65.Lb2 Kc8 66.Kd5 Kd7 67.Lc1 Kc8 68.Kc5 Kd7 69.Ld2 Kc8 70.Lc3 Kb8 71.Lb4 Ka8 72.Kd5 Kb8 73.Kd6 Kc8 74.La3 Kd8 75.Kd5 Kc8 76.Ke6 Kb8 77.Lc5 Kc8 78.Le3 Kb8 79.Kd5 Kc8 80.Ke5 Kd7 81.Kd4 Ke7 82.Lg5+ Ke6 83.Kc4 Kd6 84.Lh4 Kd7 85.Kc5 Kc8 86.Lg3 Kd7 87.Kb4 Kc6 88.Le1 Kd5 89.Lc3 Kd6 90.Kc4 Ke6 91.Kd4 Kd7 92.Kd5 Kc8 93.Ke6 Kb8 94.La1 Kc8 95.Kd6 Kb8 96.Ke5 Kc8 97.Ld4 Kb8 98.Kd6 Kc8 99.Le5 Kb8 100.Lf6 Kc8
+- (4.22) Tiefe: 92/101 00:00:50 1670MN, tb=969945

Here after a Backward from about 90 moves, when 0.00 of the 50- moves- boundary comes up at the same 7- pieces- position:

8/1p6/pP6/P2k4/5K2/4B3/8/8 b - - 0 1

Analysis by Stockfish MateFinder 64 POPCNT:

53...Kc6 54.Lc1 Kd7 55.Ke4 Kc8 56.Le3 Kd7 57.Ld2 Kd8 58.Kd5 Ke8 59.Lb4 Kd8 60.La3 Kd7 61.Lc1 Kc8 62.Lb2 Kb8 63.Lg7 Kc8 64.La1 Kb8 65.Le5+ Kc8 66.La1
= (0.00) Tiefe: 84/27 00:03:57 6969MN, tb=115711

And finally after Backward to test position, already shown in previous posting:

5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1

Analysis by Stockfish MateFinder 64 POPCNT:

30...Ke7 31.a4 h5 32.Ke2 h4 33.Kf3 h3 34.Kg4 Lg1 35.Kxh3 Lxf2 36.Ld2 a6 37.b6 g5 38.Kg2 Lxe3 39.Lxe3 g4 40.Kf2 Kd6 41.Ke2 f5 42.Kd3 Kd5 43.Ld2 e5 44.Le3 Kc6 45.Lg1 Kd5 46.Lf2 Kd6 47.Le1 Kc5 48.a5 Kb5 49.Ke3 Kc4 50.Ke2 Kd5 51.Lf2 f4 52.Kd2 Kc6 53.Kd3 Kb5
+/= (0.64) Tiefe: 36/47 00:01:05 2087MN, tb=1708610

Sorry for german abbrevations in outputs, forgot to change again

[Nordlandia]

This line is almost forced with draw as final result.

The neat idea with h4/h3 maneuver is that black can choose to capture either e3 or g3 pawn.

Which pawn is the correct one to capture?

Both alternatives appear adequately enough to secure the draw, i can't identify any major differences. Still i prefer Bxe3 as the bad bishop (big pawn) capture towards the center as per rule of thumb.

[d]5k2/1p4p1/pP3p2/4p3/8/P3P1P1/3B1bK1/8 b - - 0 8

--------------------------------------------

[pgn][Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1"]
[PlyCount "15"]

1... h5 2. Ke2 a6 3. b6 h4 4. Kf3 h3 5. Kg4 Bg1 6. Kxh3 Bxf2 7. Bd2 e5 8. Kg2
Bxg3 (8... Bxe3 9. Bxe3) *

[/pgn]

[pgn][Event "Blitz 1m"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Stockfish 100616 64 BMI2"]
[Black "Kajapix"]
[Result "*"]
[SetUp "1"]
[FEN "5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1"]
[PlyCount "5"]
[TimeControl "60"]

{4096MB, , EXPERT-PC} 1... h5 2. Ke2 a6 (2... Ke7 3. Kf3 Kd6 4. Kg2 Bxg3 5. b6
(5. Kxg3 Kc5 (5... Kd5 6. a4 g5 7. Ba3 e5 8. Be7 Ke6 9. Bc5 a6 $11)) 5... a6 6.
Kxg3 e5 7. Kh4 g6 8. Bd2 Kd5 $11) 3. bxa6 (3. b6 h4 4. Kf3 h3 5. Kg4 Bg1 6.
Kxh3 Bxf2 7. Bd2 e5 8. Kg2 Bxg3 {0} (8... Bxe3 9. Bxe3 $11) 9. Kxg3 $11 {
1.36/30 1}) (3. a4 h4 4. Kf3 axb5 5. axb5 h3 6. Kg4 Bg1 7. Kxh3 Bxf2 8. Bd2 Kf7
$6 (8... Ke7 $11)) 3... bxa6 $11 *

[/pgn]

Code: Select all

[Event "Blitz 1m"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Stockfish 100616 64 BMI2"]
[Black "Kajapix"]
[Result "*"]
[SetUp "1"]
[FEN "5k2/pp4pp/4pp2/1P6/8/P2KP1P1/5P1b/2B5 b - - 0 1"]
[PlyCount "5"]
[TimeControl "60"]

{4096MB, , EXPERT-PC} 1... h5 2. Ke2 a6 (2... Ke7 3. Kf3 Kd6 4. Kg2 Bxg3 5. b6
(5. Kxg3 Kc5 (5... Kd5 6. a4 g5 7. Ba3 e5 8. Be7 Ke6 9. Bc5 a6 $11)) 5... a6 6.
Kxg3 e5 7. Kh4 g6 8. Bd2 Kd5 $11) 3. bxa6 (3. b6 h4 4. Kf3 h3 5. Kg4 Bg1 6.
Kxh3 Bxf2 7. Bd2 e5 8. Kg2 Bxg3 {0} (8... Bxe3 9. Bxe3 $11) 9. Kxg3 $11 {
1.36/30 1}) (3. a4 h4 4. Kf3 axb5 5. axb5 h3 6. Kg4 Bg1 7. Kxh3 Bxf2 8. Bd2 Kf7
$6 (8... Ke7 $11)) 3... bxa6 $11 *

[Nordlandia]

Whichever pawn do you prefer to capture and describe in details please