Ethereal & Schooner seems to be "Big Monsters"

[Frank Quisinsky]

Hi Carldaman,

thats an interesting information.
In games vs. Wasp (Wasp 4.00 Modern x64 test-run) I saw that Wasp have problems in closed positions vs. Winter. So I wrote different times that the enigne is interesting for me. Of course I need for Wasp testing not the aggressive programs only. Winter runs without any technical problems, time managment is good! Completly new for me!

In my tourney it's very hard for engines with 2900-3000 Elo because 19 programs are clearly stronger. For better stats to fast lost / won games I need a group of 2700-2800 Elo engines. But here the same problem ... need a group for 2600-2700 Elo engines for the next and so on.

I think for engines with 2.900-3000 Elo is interesting to see how many fast lost games, topic is king safty vs clearly stronger engines. So, many good material will be produced for engines programmers with the equal FEOBOS opening book. With fast lost games engines can improved a lot. Different in the group of 2.900 - 3000 Elo engines are able to kill clearly stronger engines very fast (Texel, Chiron, Wasp).

Best
Frank

[Alayan]

Ethereal "fast wins" results are much less impressive now after 6 rounds than they were after 2, aren't they ?

But I'm glad this outlier 11 fast wins out of 80 games made you look at Ethereal games and enjoy them. Even though during development, strength is the main focus, it's rewarding to have the engine produce beautiful games and people appreciating those.

[Frank Quisinsky]

Hi Alayan,

that's right!

But if a program (Ethereal) can win so many "fast mate" games after 80 games only, more will be follow. Same for Wasp, I know that Wasp plays many of that games and for the moment place 14 _only_. So for Wasp it should go higher, maybe in the "Top-10" (opponents are to strong for Wasp) , I know that after so many test games I have.

The "fast mate" games, move average (contempt settings in sources produced often the higher move-average), short draw games are only three stats we can see a bit. More different statistics will be follow to the end of the tourney. Klaus Wlotzka will help me a bit with his Excel knowledge to do that.

Loud thinking:
If we have programs in a rating list we can't see differents to the others, the ratings by all others can't correct build. Will be the same if we have in TOP-40 of humans ratings, 10x the same player. Please thinking on it which ratings the other humans will get if Carlson is 10 times in the TOP-40.

For testing Wasp and more exactly results in Elo I need 20-30 different sparing partners in styles. The main reason for my tourney is to find out that. An other reason is to see how good works FEOBOS (at the moment I am to 90% happy with the stats, but after the work around FEOBOS it should be 100%).

Interesting are the stats to Rodent:
0 fast won, 6 fast loses only.
Often, engines not lost many games very fast are good king attacker. Rodent is in the group of engines in my tourney the weakest engine but lost only 6 games very fast. I saw Rodent stats in a group of equal engines and Rodent won a lot of fast games. The group of opponents is for Rodent to strong but engine hold one of the main strengths (good king safty with many pieces on board).

Interesting is Texel:
Texel can win very fast and lost very fast.
Same for Hakkapeliitta.
Not the same for Spark or Wasp (Spark have in a group of engines, max. 150 Elo stronger as the own strength the factor 6:1 (6x more fast won games as fast lost games), Wasp comes here with the factor 8:1! Chiron have the factor 5:1, Andscacs the factor 5:1, Pepito the factor 4:1 ... indeed all are wonderful engines with different other strengths Wasp have. So I am sure end of the tourney Wasp have here better statistics as Chiron and Pepito and arounnd the same Andscacs have with 100 Elo points more (fast won games, not many fast lost games).

After 6 rounds:
From the newer engines (I have not many games), is "Schooner" for the moment most interesting for myself. Just a great engine in his style after all I can see. Yesterday I make two new statistics to middelgame points (not ready for a publication). Sensational what Pepito do here. With time I like Pepito more and more. For the moment is Pepito one of TOP-5 Sparing partners in testing Wasp, what for a wonderful work the programmer do.

OK, 6 rounds are not enough ... thats quiet clear.

Best
Frank

PS: What Stockfish do in the middlegame is from an other planet. Yesterday I am looking in detail after the first 6 rounds.

[D Sceviour]

And now Schooner.
The game is for humans for sure very complicated.
But Schooner like such things (saw that in other games too, here the perfect example) and it seems the style is very speculative with aggressive knight, rooks and bishop moves. Not sure for the moment but the strength in middlegames seems to produced not with aggressive pawns. That is indeed very interesting! Need more of such games because such a style is new for me.

[pgn][Event "FCP Tourney-2020"]
[Site "Trier"]
[Date "2020.06.28"]
[Round "1.35"]
[White "Schooner 2.2 SSE x64"]
[Black "Igel 2.5.0 BMI2 x64"]
[Result "1-0"]
[ECO "B88"]
[PlyCount "87"]
[EventDate "2020.??.??"]
[EventType "tourn"]
[EventRounds "50"]
[EventCountry "GER"]
[SourceTitle "FCP Tourney-2020"]
[Source "Frank Quisinsky"]
[SourceDate "2020.07.02"]
[SourceVersion "2"]
[SourceVersionDate "2020.07.02"]
[SourceQuality "1"]

1. e4 {book 0s} c5 {book 0s} 2. Nf3 {book 0s} e6 {book 0s} 3. d4 {book 0s} cxd4
{book 0s} 4. Nxd4 {book 0s} Nc6 {book 0s} 5. Nc3 {book 0s} d6 {book 0s} 6. Bc4
{book 0s} Nf6 {book 0s} 7. Bb3 {book 0s} a6 {book 0s} 8. f4 {book 0s} Qc7 {
book 0s} 9. Be3 {book 0s} b5 {book 0s} 10. Qf3 {book 0s} Bb7 {book 0s} 11. f5 {
[%eval 37,27] [%emt 0:00:31]} Ne5 {[%eval 1,27] [%emt 0:02:10]} 12. Qf4 {
[%eval 48,29] [%emt 0:02:05] (Qh3)} b4 {[%eval 28,27] [%emt 0:00:55]} 13. Nce2
{[%eval 41,26] [%emt 0:00:28] (Na4)} exf5 {[%eval 0,26] [%emt 0:01:02]} 14.
Ba4+ {[%eval 35,28] [%emt 0:01:27] (Nxf5)} Kd8 {[%eval 0,25] [%emt 0:00:56]}
15. Qxf5 {[%eval 47,28] [%emt 0:01:17]} Bxe4 {[%eval 0,26] [%emt 0:01:43]} 16.
Qg5 {[%eval 57,31] [%emt 0:05:14]} Kc8 {[%eval -42,28] [%emt 0:01:54]} 17. O-O
{[%eval 53,27] [%emt 0:00:15]} Qb7 {[%eval -58,26] [%emt 0:00:57]} 18. c3 {
[%eval 93,26] [%emt 0:00:21] (Bf4)} a5 {[%eval 0,27] [%emt 0:01:24]} 19. cxb4 {
[%eval 81,26] [%emt 0:00:14] (Rac1)} h6 {[%eval -29,24] [%emt 0:00:30]} 20. Qg3
{[%eval 129,28] [%emt 0:00:14]} axb4 {[%eval 0,23] [%emt 0:00:27]} 21. Bb5 {
[%eval 122,27] [%emt 0:00:47] (Bb3)} h5 {[%eval 0,28] [%emt 0:01:05]} 22. h4 {
[%eval 133,29] [%emt 0:00:32] (a4)} Kb8 {[%eval -111,22] [%emt 0:00:56]} 23. a4
{[%eval 98,27] [%emt 0:00:55] (Bd2)} bxa3 {[%eval -49,23] [%emt 0:00:34]} 24.
bxa3 {[%eval 127,25] [%emt 0:00:24]} Nfg4 {[%eval 0,23] [%emt 0:00:47]} 25. Bf4
{[%eval 211,27] [%emt 0:00:15]} Ra5 {[%eval 0,24] [%emt 0:00:32]} 26. Rfb1 {
[%eval 289,28] [%emt 0:00:23] (a4)} Bxb1 {[%eval 0,28] [%emt 0:00:25]} 27. Rxb1
{[%eval 328,30] [%emt 0:00:18]} Kc8 {[%eval 0,28] [%emt 0:00:10]} 28. Qe1 {
[%eval 401,28] [%emt 0:00:11] (Rc1+)} Ra7 {[%eval 0,24] [%emt 0:00:33]} 29. Nc3
{[%eval 342,27] [%emt 0:00:12] (Rc1+)} Ra5 {[%eval 0,24] [%emt 0:00:33]} 30.
Ne4 {[%eval 485,27] [%emt 0:00:10] (Nce2)} Ra7 {[%eval 17,23] [%emt 0:00:31]}
31. Nf2 {[%eval 716,26] [%emt 0:00:19] (Rc1+)} Qd5 {[%eval 166,25] [%emt 0:00:
12]} 32. Rc1+ {[%eval 920,28] [%emt 0:00:11]} Rc7 {[%eval 168,23] [%emt 0:00:
27]} 33. Rxc7+ {[%eval 1140,34] [%emt 0:00:09]} Kxc7 {[%eval 665,26] [%emt 0:
00:14]} 34. Qa5+ {[%eval 1336,38] [%emt 0:00:40]} Kb8 {[%eval 665,28] [%emt 0:
00:24]} 35. Nc6+ {[%eval 1452,39] [%emt 0:00:43]} Nxc6 {[%eval 747,29] [%emt 0:
00:11]} 36. Qb6+ {[%eval 1516,39] [%emt 0:00:21]} Kc8 {[%eval 707,27] [%emt 0:
00:21]} 37. Bxc6 {[%eval 1492,39] [%emt 0:00:11]} Qf5 {[%eval 848,30] [%emt 0:
00:08]} 38. Ne4 {[%emt 0:00:08] +M-1376/28} Ne5 {[%eval 848,31] [%emt 0:00:09]}
39. Nc5 {[%emt 0:00:07] +M-1377/33 (Bxe5)} Nf3+ {[%eval 32756,44] [%emt 0:00:
19]} 40. Kh1 {[%emt 0:00:01] +M-1378/99} Qb1+ {[%eval 32758,45] [%emt 0:00:05]}
41. Qxb1 {[%emt 0:00:00] +M-1379/99} Ne5 {[%eval 32760,50] [%emt 0:00:09]} 42.
Qb7+ {[%emt 0:00:00] +M-1380/99 (Bxe5)} Kd8 {book 0s} 43. Bxe5 {[%emt 0:00:00]
+M-1381/99} dxe5 {[%eval 32764,68] [%emt 0:00:05]} 44. Qd7# {[%emt 0:00:00]
+M-1382/99} 1-0

[/pgn]

Here is an additional programmers note. Schooner has always had an extra wide null move search window. This invites piece sacrifices, and this makes a very aggressive attacking style at just the proper moment in the game. There is no loss (or gain) in overall elo strength for doing this; it only affects the style. With the large additional depths now available, Schooner can often announce mate in less than 30 moves.

The observation that the pawns are less aggressive is interesting. Perhaps I will try a user null move search flag that can change the style to "aggressive with pawns".

[carldaman]

Hi Alayan,

that's right!

But if a program (Ethereal) can win so many "fast mate" games after 80 games only, more will be follow. Same for Wasp, I know that Wasp plays many of that games and for the moment place 14 _only_. So for Wasp it should go higher, maybe in the "Top-10" (opponents are to strong for Wasp) , I know that after so many test games I have.

The "fast mate" games, move average (contempt settings in sources produced often the higher move-average), short draw games are only three stats we can see a bit. More different statistics will be follow to the end of the tourney. Klaus Wlotzka will help me a bit with his Excel knowledge to do that.

Loud thinking:
If we have programs in a rating list we can't see differents to the others, the ratings by all others can't correct build. Will be the same if we have in TOP-40 of humans ratings, 10x the same player. Please thinking on it which ratings the other humans will get if Carlson is 10 times in the TOP-40.

For testing Wasp and more exactly results in Elo I need 20-30 different sparing partners in styles. The main reason for my tourney is to find out that. An other reason is to see how good works FEOBOS (at the moment I am to 90% happy with the stats, but after the work around FEOBOS it should be 100%).

Interesting are the stats to Rodent:
0 fast won, 6 fast loses only.
Often, engines not lost many games very fast are good king attacker. Rodent is in the group of engines in my tourney the weakest engine but lost only 6 games very fast. I saw Rodent stats in a group of equal engines and Rodent won a lot of fast games. The group of opponents is for Rodent to strong but engine hold one of the main strengths (good king safty with many pieces on board).

Interesting is Texel:
Texel can win very fast and lost very fast.
Same for Hakkapeliitta.
Not the same for Spark or Wasp (Spark have in a group of engines, max. 150 Elo stronger as the own strength the factor 6:1 (6x more fast won games as fast lost games), Wasp comes here with the factor 8:1! Chiron have the factor 5:1, Andscacs the factor 5:1, Pepito the factor 4:1 ... indeed all are wonderful engines with different other strengths Wasp have. So I am sure end of the tourney Wasp have here better statistics as Chiron and Pepito and arounnd the same Andscacs have with 100 Elo points more (fast won games, not many fast lost games).

After 6 rounds:
From the newer engines (I have not many games), is "Schooner" for the moment most interesting for myself. Just a great engine in his style after all I can see. Yesterday I make two new statistics to middelgame points (not ready for a publication). Sensational what Pepito do here. With time I like Pepito more and more. For the moment is Pepito one of TOP-5 Sparing partners in testing Wasp, what for a wonderful work the programmer do.

OK, 6 rounds are not enough ... thats quiet clear.

Best
Frank

PS: What Stockfish do in the middlegame is from an other planet. Yesterday I am looking in detail after the first 6 rounds.

Interesting, but isn't Pepito rated less than 2500? Not that I would mind this kind of testing, just curious.

[Frank Quisinsky]

Ah ... Pedone, not Pepito.



I am to old for all this.

Best
Frank

But Pepito with his free sources in winboard times was also an event.

[Frank Quisinsky]

Hi,

thanks for the Schooner information to the aggressive style Schooner produced.
I like Schooner a lot and I am looking most of times the Schooner, Wasp, Ethereal games in my still running tourney.

Two problems with Schooner:

1. The 203 move problem. Very often after 203 moves (without resign mode, Shredder GUI, have a look in the tournament rules) the game ended for Schooner and Shredder GUI crashed.

FCP Tourney-2020
http://www.amateurschach.de/main/_fcp-tourney-2020.htm

2. I saw that in Wasp test-runs that Schooner give to fast games draw. I have to test with Contempt = 0 (all with Contempt = 0) but maybe Schooner need here a dynamic contempt or a little bit in main settings (without contempt UCI settings).

Yes, with aggressive pawns much more is possible. The strength of Wasp, Wasp try to search the way to go with pawns as soon as possible for open positions. Made the engine in endgames perhaps a bit weaker but the produced style is fantastic.

Nice to see your interest on aggressive computer chess with the idea you wrote about it. That's all the time what I like, the aggressive middlegame will make engines much more interesting for humans.

So, Schooner ist more as on the way only ...
Thanks for this fantastic software development!

I am very happy with Schooner (produced style is a complete other if I compare stats with other TOP-41, and the style is definitiv fantastic), but the two small problems should be fixed.

Best
Frank

[AndrewGrant]

1. The 203 move problem. Very often after 203 moves (without resign mode, Shredder GUI, have a look in the tournament rules) the game ended for Schooner and Shredder GUI crashed.
Frank

Might be worth noting that a typical uci string of "position startpos moves ..." followed by 203 moves by both players just barely comes in at > 2048 characters. Too small a buffer size is a good suspect.

[Frank Quisinsky]

Hi Andrew,

thanks that you try to give helps here.
Very important that the problem will be fixed.

Maybe the programmer of Schooner can make an update with 2.2(a), so I can changed the version in my tourney with a bugfix only. If the programmer made more changes, I have to wait for testing if the tournament is over.

Andscacs like helps.
The programmer of Andscacs like to give many helps for the others too.
A pity that I can read nothing about him since a longer time, website is down also.

Thank you Andrew!!
And thank you for Ethereal.
I have really new nice toys for testing here and have many fun.

Give me energy back in times LCO is very boring for myself.

Best
Frank

[silentshark]

And now Schooner.
The game is for humans for sure very complicated.
But Schooner like such things (saw that in other games too, here the perfect example) and it seems the style is very speculative with aggressive knight, rooks and bishop moves. Not sure for the moment but the strength in middlegames seems to produced not with aggressive pawns. That is indeed very interesting! Need more of such games because such a style is new for me.

[pgn][Event "FCP Tourney-2020"]
[Site "Trier"]
[Date "2020.06.28"]
[Round "1.35"]
[White "Schooner 2.2 SSE x64"]
[Black "Igel 2.5.0 BMI2 x64"]
[Result "1-0"]
[ECO "B88"]
[PlyCount "87"]
[EventDate "2020.??.??"]
[EventType "tourn"]
[EventRounds "50"]
[EventCountry "GER"]
[SourceTitle "FCP Tourney-2020"]
[Source "Frank Quisinsky"]
[SourceDate "2020.07.02"]
[SourceVersion "2"]
[SourceVersionDate "2020.07.02"]
[SourceQuality "1"]

1. e4 {book 0s} c5 {book 0s} 2. Nf3 {book 0s} e6 {book 0s} 3. d4 {book 0s} cxd4
{book 0s} 4. Nxd4 {book 0s} Nc6 {book 0s} 5. Nc3 {book 0s} d6 {book 0s} 6. Bc4
{book 0s} Nf6 {book 0s} 7. Bb3 {book 0s} a6 {book 0s} 8. f4 {book 0s} Qc7 {
book 0s} 9. Be3 {book 0s} b5 {book 0s} 10. Qf3 {book 0s} Bb7 {book 0s} 11. f5 {
[%eval 37,27] [%emt 0:00:31]} Ne5 {[%eval 1,27] [%emt 0:02:10]} 12. Qf4 {
[%eval 48,29] [%emt 0:02:05] (Qh3)} b4 {[%eval 28,27] [%emt 0:00:55]} 13. Nce2
{[%eval 41,26] [%emt 0:00:28] (Na4)} exf5 {[%eval 0,26] [%emt 0:01:02]} 14.
Ba4+ {[%eval 35,28] [%emt 0:01:27] (Nxf5)} Kd8 {[%eval 0,25] [%emt 0:00:56]}
15. Qxf5 {[%eval 47,28] [%emt 0:01:17]} Bxe4 {[%eval 0,26] [%emt 0:01:43]} 16.
Qg5 {[%eval 57,31] [%emt 0:05:14]} Kc8 {[%eval -42,28] [%emt 0:01:54]} 17. O-O
{[%eval 53,27] [%emt 0:00:15]} Qb7 {[%eval -58,26] [%emt 0:00:57]} 18. c3 {
[%eval 93,26] [%emt 0:00:21] (Bf4)} a5 {[%eval 0,27] [%emt 0:01:24]} 19. cxb4 {
[%eval 81,26] [%emt 0:00:14] (Rac1)} h6 {[%eval -29,24] [%emt 0:00:30]} 20. Qg3
{[%eval 129,28] [%emt 0:00:14]} axb4 {[%eval 0,23] [%emt 0:00:27]} 21. Bb5 {
[%eval 122,27] [%emt 0:00:47] (Bb3)} h5 {[%eval 0,28] [%emt 0:01:05]} 22. h4 {
[%eval 133,29] [%emt 0:00:32] (a4)} Kb8 {[%eval -111,22] [%emt 0:00:56]} 23. a4
{[%eval 98,27] [%emt 0:00:55] (Bd2)} bxa3 {[%eval -49,23] [%emt 0:00:34]} 24.
bxa3 {[%eval 127,25] [%emt 0:00:24]} Nfg4 {[%eval 0,23] [%emt 0:00:47]} 25. Bf4
{[%eval 211,27] [%emt 0:00:15]} Ra5 {[%eval 0,24] [%emt 0:00:32]} 26. Rfb1 {
[%eval 289,28] [%emt 0:00:23] (a4)} Bxb1 {[%eval 0,28] [%emt 0:00:25]} 27. Rxb1
{[%eval 328,30] [%emt 0:00:18]} Kc8 {[%eval 0,28] [%emt 0:00:10]} 28. Qe1 {
[%eval 401,28] [%emt 0:00:11] (Rc1+)} Ra7 {[%eval 0,24] [%emt 0:00:33]} 29. Nc3
{[%eval 342,27] [%emt 0:00:12] (Rc1+)} Ra5 {[%eval 0,24] [%emt 0:00:33]} 30.
Ne4 {[%eval 485,27] [%emt 0:00:10] (Nce2)} Ra7 {[%eval 17,23] [%emt 0:00:31]}
31. Nf2 {[%eval 716,26] [%emt 0:00:19] (Rc1+)} Qd5 {[%eval 166,25] [%emt 0:00:
12]} 32. Rc1+ {[%eval 920,28] [%emt 0:00:11]} Rc7 {[%eval 168,23] [%emt 0:00:
27]} 33. Rxc7+ {[%eval 1140,34] [%emt 0:00:09]} Kxc7 {[%eval 665,26] [%emt 0:
00:14]} 34. Qa5+ {[%eval 1336,38] [%emt 0:00:40]} Kb8 {[%eval 665,28] [%emt 0:
00:24]} 35. Nc6+ {[%eval 1452,39] [%emt 0:00:43]} Nxc6 {[%eval 747,29] [%emt 0:
00:11]} 36. Qb6+ {[%eval 1516,39] [%emt 0:00:21]} Kc8 {[%eval 707,27] [%emt 0:
00:21]} 37. Bxc6 {[%eval 1492,39] [%emt 0:00:11]} Qf5 {[%eval 848,30] [%emt 0:
00:08]} 38. Ne4 {[%emt 0:00:08] +M-1376/28} Ne5 {[%eval 848,31] [%emt 0:00:09]}
39. Nc5 {[%emt 0:00:07] +M-1377/33 (Bxe5)} Nf3+ {[%eval 32756,44] [%emt 0:00:
19]} 40. Kh1 {[%emt 0:00:01] +M-1378/99} Qb1+ {[%eval 32758,45] [%emt 0:00:05]}
41. Qxb1 {[%emt 0:00:00] +M-1379/99} Ne5 {[%eval 32760,50] [%emt 0:00:09]} 42.
Qb7+ {[%emt 0:00:00] +M-1380/99 (Bxe5)} Kd8 {book 0s} 43. Bxe5 {[%emt 0:00:00]
+M-1381/99} dxe5 {[%eval 32764,68] [%emt 0:00:05]} 44. Qd7# {[%emt 0:00:00]
+M-1382/99} 1-0

[/pgn]

Here is an additional programmers note. Schooner has always had an extra wide null move search window. This invites piece sacrifices, and this makes a very aggressive attacking style at just the proper moment in the game. There is no loss (or gain) in overall elo strength for doing this; it only affects the style. With the large additional depths now available, Schooner can often announce mate in less than 30 moves.

The observation that the pawns are less aggressive is interesting. Perhaps I will try a user null move search flag that can change the style to "aggressive with pawns".

Interesting.. What do you mean extra wide null move search window. So you don't search around beta-1, beta, but something wider? I don't get how that would change things to do with piece sacs.