Enough is Enough

Posts : 119 Join date : 2022-07-19

Perhaps adding some options to exclude pawn up positions, since lots of pawn up endgames are draws and we "ASSUME" engines to play perfect so pawn up might not be winning, and rook for knight can be a draw, so up 2 pts material I'm sure is both common for engines.

Posts : 132 Join date : 2022-03-01 Location : Berlin

texium wrote:: Perhaps adding some options to exclude pawn up positions, since lots of pawn up endgames are draws and we "ASSUME" engines to play perfect so pawn up might not be winning, and rook for knight can be a draw, so up 2 pts material I'm sure is both common for engines.

I did this already. With 2 pawns, the numbers of bad draws halfed for most of engines, the ranking considering EAS-points did (nearly) not change.

Everybody can change the EAS-tool as much as he likes.

Switching to 2 pawns advantage = bad draw is very easy: Open the EAS batch-file with any text-editor and find these lines in the code:

REM *** Find all draws, were engine had material advantage (=bad draw)
pgn-extract --quiet -Tw%engine% -y1_pawnsac_black enginedraws.pgn -abad_draws2.pgn >NUL
pgn-extract --quiet -Tb%engine% -y1_pawnsac_white enginedraws.pgn -abad_draws2.pgn >NUL

Now, change these lines to:
REM *** Find all draws, were engine had material advantage (=bad draw)
pgn-extract --quiet -Tw%engine% -y2_pawnsac_black enginedraws.pgn -abad_draws2.pgn >NUL
pgn-extract --quiet -Tb%engine% -y2_pawnsac_white enginedraws.pgn -abad_draws2.pgn >NUL

(only the -y1_pawnsac_black changed to -y2_pawnsac_black and -y1_pawnsac_white changed to -y2_pawnsac_white, rest stays unchanged - very easy)

This is done in just 1 minute!

Thats the good thing about programming in batch-language: Its just an interpreter-language and no compiler or development-environments-installments or anything else is needed. Anybody can change anything without knowing anything about compiling etc... Just use an editor and do it! All you need are programming skills on school-level (Python, Basic, Pascal or so). And learning the batch language a bit. Here a nice site for doing so:
https://www.tutorialspoint.com/batch_script/index.htm

And you should definitly know pgn-extract and its manual, of course. Otherwise, you will not understand, what the EAS-tool is doing.
https://www.cs.kent.ac.uk/people/staff/djb/pgn-extract/

Subject: Re: Enough is Enough Fri Sep 20, 2024 8:51 am

pohl4711 wrote:: Another interesting point here is Stockfish final HCE, which is the only non-neural net engine in my UHO-full ratinglist: Stockfish final HCE has an enourmous high bad draw value (26.11%), which is 3x higher than Stockfish 16.1 and around 5x higher than the (much weaker!) Patricia engine. Why is that? That is exactly the point, where it gets interesting (and what even Ed sadly does not understand): A non-neural net engine has much less positional understanding. The neural-net engines understand, that just one pawn more in many endgames is not enough for a win. So, they try to avoid these endgames (a possible way is, to avoid captures (= avoid going towards endgame) until a second pawn was won). Stockfish final HCE doesnt have this understanding, so it believes in winning, when having one pawn more, no matter if it is an endgame or not). So, it has much higher number of bad draws, even though, Stockfish final HCE is a tactical monster and +114 Celo stronger than Revenge 1 and +261 Celo stronger than Patricia...

Oh.....Stefan, how dare you Very Happy

, you are going to eat those words Very Happy

Consider my version of the bad-draw-list.

Code:: Bad draw List [game decided margin = 300]

PGN database : pgn\uho_ratinglist_games.pgn
Count Bad Draws : 58.100

Engine Won Draw Loss Perc BAD
RofChade 3.1 avx2 2291 7126 5583 39.0% 33
CSTal 2.0 avx2 2939 7497 4564 44.6% 23
Rebel EAS avx2 2263 7366 5371 39.6% 16
Seer 2.7.0 avx2 2245 7139 5616 38.8% 16
RubiChess 230918 av 3514 7477 4009 48.3% 15
Revenge 3.0 avx2 1941 7292 5767 37.2% 14
Uralochka 3.40a avx 2122 7108 5770 37.8% 12
Koivisto 9.2 avx2 2599 7395 5006 42.0% 10
Caissa 1.14 avx2 3153 7405 4442 45.7% 8
Ethereal 14.25 nnue 4112 7521 3367 52.5% 7
Clover 6 avx2 2239 7425 5336 39.7% 6
Torch 1 popavx2 6527 7105 1368 67.2% 2
Berserk 12 avx2 5377 7255 2368 60.0% 1
Stockfish 231107 av 7439 6947 614 72.7% 0
Stockfish 16 230630 7199 7100 701 71.7% 0
KomodoDragon 3.3 av 5846 7230 1924 63.1% 0

Counted 163 bad draws, press a key to continue...

It does not need to look at the material imbalance, it ONLY uses the scores in the PGN.

When an engine X has a positive score of >=3 pawns for 5 consecutive moves and the opponent engine Y in the 5 consecutive moves is behind in score with 3 pawns and engine X only draws then that is a bad draw. Currently I only count them (no penalties yet) and as you can see it has not much to with aggressiveness but purely strength, Stockfish and Komodo have zero bad draws.

Besides the BAD column I need a second one called WORSE or so, surely there must be a few cases engine X has that winning score >= 3 pawns for 5 consecutive moves and even manages to lose the game.

So indeed, we trust the NNUE score a lot more than the HCE score and BTW the margin of 3 pawns of course is flexible and user defined.

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:

pohl4711 wrote:: Another interesting point here is Stockfish final HCE, which is the only non-neural net engine in my UHO-full ratinglist: Stockfish final HCE has an enourmous high bad draw value (26.11%), which is 3x higher than Stockfish 16.1 and around 5x higher than the (much weaker!) Patricia engine. Why is that? That is exactly the point, where it gets interesting (and what even Ed sadly does not understand): A non-neural net engine has much less positional understanding. The neural-net engines understand, that just one pawn more in many endgames is not enough for a win. So, they try to avoid these endgames (a possible way is, to avoid captures (= avoid going towards endgame) until a second pawn was won). Stockfish final HCE doesnt have this understanding, so it believes in winning, when having one pawn more, no matter if it is an endgame or not). So, it has much higher number of bad draws, even though, Stockfish final HCE is a tactical monster and +114 Celo stronger than Revenge 1 and +261 Celo stronger than Patricia...

Oh.....Stefan, how dare you Very Happy

, you are going to eat those words Very Happy

Consider my version of the bad-draw-list.

Code:: Bad draw List [game decided margin = 300]

PGN database : pgn\uho_ratinglist_games.pgn
Count Bad Draws : 58.100

Engine Won Draw Loss Perc BAD
RofChade 3.1 avx2 2291 7126 5583 39.0% 33
CSTal 2.0 avx2 2939 7497 4564 44.6% 23
Rebel EAS avx2 2263 7366 5371 39.6% 16
Seer 2.7.0 avx2 2245 7139 5616 38.8% 16
RubiChess 230918 av 3514 7477 4009 48.3% 15
Revenge 3.0 avx2 1941 7292 5767 37.2% 14
Uralochka 3.40a avx 2122 7108 5770 37.8% 12
Koivisto 9.2 avx2 2599 7395 5006 42.0% 10
Caissa 1.14 avx2 3153 7405 4442 45.7% 8
Ethereal 14.25 nnue 4112 7521 3367 52.5% 7
Clover 6 avx2 2239 7425 5336 39.7% 6
Torch 1 popavx2 6527 7105 1368 67.2% 2
Berserk 12 avx2 5377 7255 2368 60.0% 1
Stockfish 231107 av 7439 6947 614 72.7% 0
Stockfish 16 230630 7199 7100 701 71.7% 0
KomodoDragon 3.3 av 5846 7230 1924 63.1% 0

Counted 163 bad draws, press a key to continue...

It does not need to look at the material imbalance, it ONLY uses the scores in the PGN.

When an engine X has a positive score of >=3 pawns for 5 consecutive moves and the opponent engine Y in the 5 consecutive moves is behind in score with 3 pawns and engine X only draws then that is a bad draw. Currently I only count them (no penalties yet) and as you can see it has not much to with aggressiveness but purely strength, Stockfish and Komodo have zero bad draws.

Besides the BAD column I need a second one called WORSE or so, surely there must be a few cases engine X has that winning score >= 3 pawns for 5 consecutive moves and even manages to lose the game.

So indeed, we trust the NNUE score a lot more than the HCE score and BTW the margin of 3 pawns of course is flexible and user defined.

To be honest, I really dont like the idea, to use any engine evals in the EAS-tool. Not only, because evals can be very different - depending on the engine, but also not all games have evals in the comments and doing any evaluations in the EAS-tool would be way too slow. And nobody can be sure about how any GUIs in the future will write the evals in the pgn-comments (text-format, Point of View (White or color of the engine), or change of evals to WDL-scores as new standard - how knows?). So, this approach of using engine-evals can heavily fail in the future. My EAS-tool will not, as it considers only the played moves and the game result. And this will work, as long as the pgn-format itself stays unchanged...
And, addtionally, the EAS-tool as it is can be used for human games, too (even though, the sac-search can lead to some false finds, when humans blunder or lose on time).
So, for me, engine evals are taboo for my EAS-tool. Of course, you can do, whatever you want in your tool. But I strongly recommend not to use engine evals.
Not using any evals is one of the best features of all of my tools, IMHO.

Posts : 119 Join date : 2022-07-19

pohl4711 wrote:

Admin wrote:

pohl4711 wrote:: Another interesting point here is Stockfish final HCE, which is the only non-neural net engine in my UHO-full ratinglist: Stockfish final HCE has an enourmous high bad draw value (26.11%), which is 3x higher than Stockfish 16.1 and around 5x higher than the (much weaker!) Patricia engine. Why is that? That is exactly the point, where it gets interesting (and what even Ed sadly does not understand): A non-neural net engine has much less positional understanding. The neural-net engines understand, that just one pawn more in many endgames is not enough for a win. So, they try to avoid these endgames (a possible way is, to avoid captures (= avoid going towards endgame) until a second pawn was won). Stockfish final HCE doesnt have this understanding, so it believes in winning, when having one pawn more, no matter if it is an endgame or not). So, it has much higher number of bad draws, even though, Stockfish final HCE is a tactical monster and +114 Celo stronger than Revenge 1 and +261 Celo stronger than Patricia...

Oh.....Stefan, how dare you Very Happy

, you are going to eat those words Very Happy

Consider my version of the bad-draw-list.

Code:: Bad draw List [game decided margin = 300]

PGN database : pgn\uho_ratinglist_games.pgn
Count Bad Draws : 58.100

Engine Won Draw Loss Perc BAD
RofChade 3.1 avx2 2291 7126 5583 39.0% 33
CSTal 2.0 avx2 2939 7497 4564 44.6% 23
Rebel EAS avx2 2263 7366 5371 39.6% 16
Seer 2.7.0 avx2 2245 7139 5616 38.8% 16
RubiChess 230918 av 3514 7477 4009 48.3% 15
Revenge 3.0 avx2 1941 7292 5767 37.2% 14
Uralochka 3.40a avx 2122 7108 5770 37.8% 12
Koivisto 9.2 avx2 2599 7395 5006 42.0% 10
Caissa 1.14 avx2 3153 7405 4442 45.7% 8
Ethereal 14.25 nnue 4112 7521 3367 52.5% 7
Clover 6 avx2 2239 7425 5336 39.7% 6
Torch 1 popavx2 6527 7105 1368 67.2% 2
Berserk 12 avx2 5377 7255 2368 60.0% 1
Stockfish 231107 av 7439 6947 614 72.7% 0
Stockfish 16 230630 7199 7100 701 71.7% 0
KomodoDragon 3.3 av 5846 7230 1924 63.1% 0

Counted 163 bad draws, press a key to continue...

It does not need to look at the material imbalance, it ONLY uses the scores in the PGN.

When an engine X has a positive score of >=3 pawns for 5 consecutive moves and the opponent engine Y in the 5 consecutive moves is behind in score with 3 pawns and engine X only draws then that is a bad draw. Currently I only count them (no penalties yet) and as you can see it has not much to with aggressiveness but purely strength, Stockfish and Komodo have zero bad draws.

Besides the BAD column I need a second one called WORSE or so, surely there must be a few cases engine X has that winning score >= 3 pawns for 5 consecutive moves and even manages to lose the game.

So indeed, we trust the NNUE score a lot more than the HCE score and BTW the margin of 3 pawns of course is flexible and user defined.

To be honest, I really dont like the idea, to use any engine evals in the EAS-tool. Not only, because evals can be very different - depending on the engine, but also not all games have evals in the comments and doing any evaluations in the EAS-tool would be way too slow. And nobody can be sure about how any GUIs in the future will write the evals in the pgn-comments (text-format, Point of View (White or color of the engine), or change of evals to WDL-scores as new standard - how knows?). So, this approach of using engine-evals can heavily fail in the future. My EAS-tool will not, as it considers only the played moves and the game result. And this will work, until the pgn-format itself stays unchanged...
And, addtionally, the EAS-tool as it is can be used for human games, too (even though, the sac-search can lead to some false finds, when humans blunder or lose on time).
So, for me, engine evals are taboo for my EAS-tool. Of course, you can do, whatever you want in your tool. But I strongly recommend not to use engine evals.
Not using any evals is one of the best features of all of my tools, IMHO.

Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

Subject: Re: Enough is Enough Fri Sep 20, 2024 10:29 am

I also worked on the shorties, same 3 pawn principle, when an engine X on move 30 has a positive score of >=3 pawns for 5 consecutive moves and the opponent engine Y in the 5 consecutive moves is behind in score with 3 pawns and engine X wins then move 30 is used to calculate an EAS score. Rule is that the winning position must occur before move 40 else it can not count as a shortie.

EAS points are based on move number, in our example case move 30.

Code::    int shorties[] = {
   1000,1000,1000,1000,1000,1000,1000,1000,1000,1000, // 00-09
   999, 999, 999, 999, 999, 999, 999, 999, 999, 999, // 10-19
   999, 900, 800, 700, 600, 500, 450, 400, 350, 300, // 20-29
   250, 210, 180, 150, 120, 95, 70, 50, 35, 25}; // 30-39

So move 30 gets a bonus of 250 points.

Examples on - https://drive.google.com/file/d/1DJiWi51yLvn8OFfJwq2FOX-cLj8UFHNU/view?usp=drive_link

Last, I have looked into the sacrifices part, I think I can't do that better.

Posts : 132 Join date : 2022-03-01 Location : Berlin

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

IMHO this would defintily fail, because of a lot too much players (EAS-Tool is limited to 950 different players in one database). But what can be done is:

Filtering all games of one player. Put these games in one new pgn-file and then use the Gauntlet-version of my EAS-Tool to get the EAS-score of this player. (The Gauntlet version of the EAS-Tool only evaluates the player with the most games in the pgn-gamebase)

Repeat this for different players, you are interested in, and then compare these EAS-scorings.

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:: Last, I have looked into the sacrifices part, I think I can't do that better.

I doubt that. As I mentioned before, a C++ solution could be much better, than mine:

"You just have to count material in pawn-units after each ply and look, if the winning color has less pawn-units on the board for 8 consecutive plies. This would work even better than my solution, because in pgn-extract, there are some different piece-patterns with the same amount of pawn-units less (or more) for one color, and switching between them, resets the counter of consecutive plies, which can lead to overlooking some sacs... no way to avoid this for me, because of pgn-extract. But a pure "pawn-unit-counting"-solution in C++ would avoid these problem."

Subject: Re: Enough is Enough Fri Sep 20, 2024 11:01 am

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

A complete mega base might take days, an elite base is recommended.

https://www.pgnmentor.com/files.html#players

Subject: Re: Enough is Enough Sat Sep 21, 2024 8:00 am

From https://www.pgnmentor.com/files.html#players

Code:: Player EAS
Steinitz, William 114.355
Lasker, Emanuel 100.654
Carlsen, Magnus 82.632
Alekhine, Alexander 82.401
Fischer, Robert James 70.500
Capablanca, Jose Raul 58.828
Euwe, Max 53.967
Rubinstein, Akiba 52.981
Tal, Mihail 51.900
Kasparov, Gary 45.600
Bronstein, David 44.533
Botvinnik, Mikhail 44.222
Anand,V 43.820
Spassky, Boris V 40.075
Nimzowitsch, Aaron 39.243
Karpov, Anatoly 27.636
Smyslov, Vassily 26.959
Petrosian, Tigran V 26.739

Code:: Morphy, Paul 195.323 Not enough games, EAS-score not reliable [50+ wins and 30+ draws needed]

Sadly.

Lemme know which other players (from the link!) should be added to the list.

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

A complete mega base might take days, an elite base is recommended.

https://www.pgnmentor.com/files.html#players

Great site, thanks Ed!

I downloaded some famous players and made an EAS-Ratinglist (on my Experiments-subsite):

https://www.sp-cc.de/experiments.htm

Subject: Re: Enough is Enough Sat Sep 21, 2024 8:26 am

pohl4711 wrote:

Admin wrote:: Last, I have looked into the sacrifices part, I think I can't do that better.

I doubt that. As I mentioned before, a C++ solution could be much better, than mine:

"You just have to count material in pawn-units after each ply and look, if the winning color has less pawn-units on the board for 8 consecutive plies. This would work even better than my solution, because in pgn-extract, there are some different piece-patterns with the same amount of pawn-units less (or more) for one color, and switching between them, resets the counter of consecutive plies, which can lead to overlooking some sacs... no way to avoid this for me, because of pgn-extract. But a pure "pawn-unit-counting"-solution in C++ would avoid these problem."

I know, but... say you have discovered one less pawn-unit and start to check it for 8 consecutive moves (not plies) and while in the process of doing so you discover 4 less pawn-units, how do you go from there?

Subject: Re: Enough is Enough Sat Sep 21, 2024 8:30 am

pohl4711 wrote:

Admin wrote:

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

A complete mega base might take days, an elite base is recommended.

https://www.pgnmentor.com/files.html#players

Great site, thanks Ed!

I downloaded some famous players and made an EAS-Ratinglist (on my Experiments-subsite):

https://www.sp-cc.de/experiments.htm

How odd, we have different results, I am still using version 5.21

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:

pohl4711 wrote:

Admin wrote:: Last, I have looked into the sacrifices part, I think I can't do that better.

I doubt that. As I mentioned before, a C++ solution could be much better, than mine:

"You just have to count material in pawn-units after each ply and look, if the winning color has less pawn-units on the board for 8 consecutive plies. This would work even better than my solution, because in pgn-extract, there are some different piece-patterns with the same amount of pawn-units less (or more) for one color, and switching between them, resets the counter of consecutive plies, which can lead to overlooking some sacs... no way to avoid this for me, because of pgn-extract. But a pure "pawn-unit-counting"-solution in C++ would avoid these problem."

I know, but... say you have discovered one less pawn-unit and start to check it for 8 consecutive moves (not plies) and while in the process of doing so you discover 4 less pawn-units, how do you go from there?

I am not sure, that I understand you.
Do you mean these lines in the 1_pawnsac files?
12 q1r1-l4-p* q=r=l=p1>=
12 q1r2l4p5- q=r=l=p1>=
12 q1r2l3p6- q=r=l=p1>=
12 q1r2l2-p* q=r=l=p1>=

The idea is, to try to avoid giving EAS-points for gambits in the opening, if the pawn can be recaptured soon. These lines mean: When the board is still nearly full, then a pawn-sac is detected after 12 plies, not 8.

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:

pohl4711 wrote:

Admin wrote:

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

A complete mega base might take days, an elite base is recommended.

https://www.pgnmentor.com/files.html#players

Great site, thanks Ed!

I downloaded some famous players and made an EAS-Ratinglist (on my Experiments-subsite):

https://www.sp-cc.de/experiments.htm

How odd, we have different results, I am still using version 5.21

I used the Gambit-EAS Tool and fixed the short-wins bonus-point-limit to 40 moves. In the folder for engine developers, the EAS-Tools have the option to hardcode this limit:
REM **************************************************************************************
REM *** special hardcoded shortwin_movelimit here, change to other values, if you want ***
REM *** set it to 0, to deactivate the hardcode-override
set /A hardlimit=40
REM **************************************************************************************

Subject: Re: Enough is Enough Sat Sep 21, 2024 10:42 am

Code:: Bad draw List [game decided margin = 300]       |
   |
PGN database : pgn\uho_ratinglist_games.pgn |
Count Bad Draws : 58.100 | bad avg.win
   | Rank EAS-Score sacs shorts draws moves Engine/player
Engine Won Draw Loss Perc BAD | -----------------------------------------------------------
RofChade 3.1 avx2 2291 7126 5583 39.0% 33 | 1 90601 08.46% 71 Stockfish 231107 avx2
CSTal 2.0 avx2 2939 7497 4564 44.6% 23 | 2 84681 09.65% 71 Stockfish 16 230630
Rebel EAS avx2 2263 7366 5371 39.6% 16 | 3 64516 13.78% 69 Torch 1 popavx2
Seer 2.7.0 avx2 2245 7139 5616 38.8% 16 | 4 56169 15.78% 76 Uralochka 3.40a avx2
RubiChess 230918 av 3514 7477 4009 48.3% 15 | 5 56169 15.72% 74 Revenge 3.0 avx2
Revenge 3.0 avx2 1941 7292 5767 37.2% 14 | 6 56169 15.01% 72 KomodoDragon 3.3 avx2
Uralochka 3.40a avx 2122 7108 5770 37.8% 12 | 7 51984 16.49% 87 Ethereal 14.25 nnue
Koivisto 9.2 avx2 2599 7395 5006 42.0% 10 | 8 51984 16.06% 75 RubiChess 230918 avx2
Caissa 1.14 avx2 3153 7405 4442 45.7% 8 | 9 48400 17.85% 82 CSTal 2.0 avx2
Ethereal 14.25 nnue 4112 7521 3367 52.5% 7 | 10 44944 18.65% 80 Berserk 12 avx2
Clover 6 avx2 2239 7425 5336 39.7% 6 | 11 41616 19.03% 84 Rebel EAS avx2
Torch 1 popavx2 6527 7105 1368 67.2% 2 | 12 38809 20.66% 86 Koivisto 9.2 avx2
Berserk 12 avx2 5377 7255 2368 60.0% 1 | 13 38809 20.31% 79 Caissa 1.14 avx2
Stockfish 231107 av 7439 6947 614 72.7% 0 | 14 35721 21.24% 78 Clover 6 avx2
Stockfish 16 230630 7199 7100 701 71.7% 0 | 15 30625 23.35% 79 RofChade 3.1 avx2
KomodoDragon 3.3 av 5846 7230 1924 63.1% 0 | 16 26244 25.98% 75 Seer 2.7.0 avx2

Counted 163 bad draws, press a key to continue...

A comparison between your bad draws code and mine, the results are very similar. What they also have in common is that's it's not about aggressiveness but about playing strength, weaker engines out-searched as best guess.

Other that that it's ~10 times faster in C++ and I wrote the code in 15 minutes.

Subject: Re: Enough is Enough Sat Sep 21, 2024 10:47 am

pohl4711 wrote:

Admin wrote:

pohl4711 wrote:

Admin wrote:

texium wrote:: Speaking of...has someone run this on the mega base or an elite player base to find the most aggressive players?

A complete mega base might take days, an elite base is recommended.

https://www.pgnmentor.com/files.html#players

Great site, thanks Ed!

I downloaded some famous players and made an EAS-Ratinglist (on my Experiments-subsite):

https://www.sp-cc.de/experiments.htm

How odd, we have different results, I am still using version 5.21

I used the Gambit-EAS Tool and fixed the short-wins bonus-point-limit to 40 moves. In the folder for engine developers, the EAS-Tools have the option to hardcode this limit:
REM **************************************************************************************
REM *** special hardcoded shortwin_movelimit here, change to other values, if you want ***
REM *** set it to 0, to deactivate the hardcode-override
set /A hardlimit=40
REM **************************************************************************************

Okay, downloaded and will use the latest version.

Subject: Re: Enough is Enough Sat Sep 21, 2024 11:01 am

pohl4711 wrote:

Admin wrote:

pohl4711 wrote:

Admin wrote:: Last, I have looked into the sacrifices part, I think I can't do that better.

I doubt that. As I mentioned before, a C++ solution could be much better, than mine:

"You just have to count material in pawn-units after each ply and look, if the winning color has less pawn-units on the board for 8 consecutive plies. This would work even better than my solution, because in pgn-extract, there are some different piece-patterns with the same amount of pawn-units less (or more) for one color, and switching between them, resets the counter of consecutive plies, which can lead to overlooking some sacs... no way to avoid this for me, because of pgn-extract. But a pure "pawn-unit-counting"-solution in C++ would avoid these problem."

I know, but... say you have discovered one less pawn-unit and start to check it for 8 consecutive moves (not plies) and while in the process of doing so you discover 4 less pawn-units, how do you go from there?

I am not sure, that I understand you.
Do you mean these lines in the 1_pawnsac files?
12 q1r1-l4-p* q=r=l=p1>=
12 q1r2l4p5- q=r=l=p1>=
12 q1r2l3p6- q=r=l=p1>=
12 q1r2l2-p* q=r=l=p1>=

The idea is, to try to avoid giving EAS-points for gambits in the opening, if the pawn can be recaptured soon. These lines mean: When the board is still nearly full, then a pawn-sac is detected after 12 plies, not 8.

Yes, we should avoid short tactical sacrifices, they are not real.

I mean these kind of situations, consider an example -

14. e5 (pawn sac) dxe5 15. Bd3 c5; 16.Bxh7+ (bishop sac)

How does pgn-extract handle this?
1. Does it still check 14.e5 for 12 plies?
2. Does it also checks 16.Bxh7+ for 12 plies?

Subject: Re: Enough is Enough Sat Sep 21, 2024 11:50 am

Code:: Engine Shorties List [game decided margin = 300] |
   |
PGN database : pgn\uho_ratinglist_games.pgn |
Calculate EAS : 61.800 |
   |
Engine Won Loss Perc EAS |
Stockfish 231107 av 7439 614 92.4% 15588 | 1 54468 26.93% 69 Torch 1 popavx2
Stockfish 16 230630 7199 701 91.1% 15039 | 2 47939 23.43% 71 Stockfish 16 230630
Uralochka 3.40a avx 2122 5770 26.9% 12346 | 3 46014 22.58% 71 Stockfish 231107 avx2
Rebel EAS avx2 2263 5371 29.6% 11600 | 4 38673 21.14% 72 KomodoDragon 3.3 avx2
Torch 1 popavx2 6527 1368 82.7% 11505 | 5 35802 22.77% 74 Revenge 3.0 avx2
KomodoDragon 3.3 av 5846 1924 75.2% 11001 | 6 32352 19.13% 76 Uralochka 3.40a avx2
Revenge 3.0 avx2 1941 5767 25.2% 10888 | 7 23921 17.76% 75 RubiChess 230918 avx2
CSTal 2.0 avx2 2939 4564 39.2% 10300 | 8 19767 15.32% 75 Seer 2.7.0 avx2
Caissa 1.14 avx2 3153 4442 41.5% 9954 | 9 12230 12.64% 78 Clover 6 avx2
Seer 2.7.0 avx2 2245 5616 28.6% 9634 | 10 10752 11.61% 79 Caissa 1.14 avx2
Koivisto 9.2 avx2 2599 5006 34.2% 9264 | 11 10572 12.13% 79 RofChade 3.1 avx2
Clover 6 avx2 2239 5336 29.6% 9002 | 12 8623 10.79% 82 CSTal 2.0 avx2
Ethereal 14.25 nnue 4112 3367 55.0% 7588 | 13 6789 09.47% 80 Berserk 12 avx2
RubiChess 230918 av 3514 4009 46.7% 7313 | 14 6223 10.43% 84 Rebel EAS avx2
RofChade 3.1 avx2 2291 5583 29.1% 7282 | 15 436 08.44% 87 Ethereal 14.25 nnue
Berserk 12 avx2 5377 2368 69.4% 6354 | 16 0 06.89% 86 Koivisto 9.2 avx2
   |
Counted 24015 shorties, press a key to continue... |

A comparison between your shorties code and mine, while the results are similar there are a couple of exceptions.

1. In your list Torch tops, in mine REBEL EAS is above Torch.
2. My list is more kinder towards Koivisto Very Happy

, probably because my list counts 2.5 x more shorties.
3. Uralochka (26.9%) and Revenge (25.2%) very high listed despite their bad performance in this elo pool, great, that's what EAS is about also.

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:: Yes, we should avoid short tactical sacrifices, they are not real.

I mean these kind of situations, consider an example -

14. e5 (pawn sac) dxe5 15. Bd3 c5; 16.Bxh7+ (bishop sac)

How does pgn-extract handle this?
1. Does it still check 14.e5 for 12 plies?
2. Does it also checks 16.Bxh7+ for 12 plies?

You should not try to copy my piece-patterns exactly. Additionally I am not 100% sure, if other 08-ply-filtering piece patterns do not "override" these 4 lines with 12 plies in some cases.

You should instead realize just the idea behind my (bad) solution. For you, this is much easier, because you are just counting the pawn-units, both sides have on the board. IMHO, you should do it like this:

If the winning color of the game has exactly one pawn less than the loosing color, you must count, how many pawn-units overall are still on the board. (Maximum is (no Kings!) (queens) 9*2, (rooks) 5*4, (Bishop/Knights) 3*8, (pawns) 1*16 = 78 pawnunits).
If 10 (or 9 or so) pawn-units or less have already disappeared, you should count up to 12 plies (because the board is still very full = early phase of the game = opening), in all other cases only up to 8 plies. Thats it!

All other (and higher) sacs, also counting up to 8 plies. Of course.

That realizes my idea: 1 pawn-sac in the opening: 12 consecutive plies. And 8 consecutive plies for all other sacs in all other game-phases.

Subject: Re: Enough is Enough Sun Sep 22, 2024 10:53 am

Your sac system is not stupid.

The definition of a sacrifice is not simple and the info in the PGN is not sufficient for a 100% solution, it would have been helpful if each move should contain the full mainline, but that's impossible with cutechess.

I will give it a try reading the PGN into memory and make it one big mainline and see if I can extract something useful from it.

And this is not going to be a 15 minutes job. scratch

Subject: Re: Enough is Enough Fri Sep 27, 2024 8:45 am

@Sfefan, I have been working on the sac part of the EAS util and so far get total different results than you.

https://drive.google.com/file/d/1Wv9wJXRalIOyRVHcMPtqIpj9x1fcH4v7/view?usp=drive_link

It's an older rating list, 120.000 games.

Code:: Rank EAS-Score sacs shorts draws moves Engine/player
-----------------------------------------------------------
   1 56520 21.92% 71 Stockfish 231107 avx2
   2 52760 21.16% 71 Stockfish 16 230630
   3 37975 14.61% 76 Uralochka 3.40a avx2
   4 35885 18.34% 84 Rebel EAS avx2
   5 35090 16.07% 69 Torch 1 popavx2
   6 27445 13.91% 74 Revenge 3.0 avx2
   7 25705 16.73% 75 RubiChess 230918 avx2
   8 25240 16.25% 72 KomodoDragon 3.3 avx2
   9 24410 11.64% 82 CSTal 2.0 avx2
  10 22180 14.30% 87 Ethereal 14.25 nnue
  11 20190 08.49% 78 Clover 6 avx2
  12 17430 11.45% 79 Caissa 1.14 avx2
  13 17155 09.84% 80 Berserk 12 avx2
  14 15935 06.85% 86 Koivisto 9.2 avx2
  15 11400 07.62% 75 Seer 2.7.0 avx2
  16 9825 09.78% 79 RofChade 3.1 avx2

Rebel EAS is fourth, but I can not comprehend that with the data in the 6 sac pgns.

Can you shed your light on that?

Posts : 132 Join date : 2022-03-01 Location : Berlin

Admin wrote:

@Sfefan, I have been working on the sac part of the EAS util and so far get total different results than you.

https://drive.google.com/file/d/1Wv9wJXRalIOyRVHcMPtqIpj9x1fcH4v7/view?usp=drive_link

It's an older rating list, 120.000 games.

Code:: Rank EAS-Score sacs shorts draws moves Engine/player
-----------------------------------------------------------
   1 56520 21.92% 71 Stockfish 231107 avx2
   2 52760 21.16% 71 Stockfish 16 230630
   3 37975 14.61% 76 Uralochka 3.40a avx2
   4 35885 18.34% 84 Rebel EAS avx2
   5 35090 16.07% 69 Torch 1 popavx2
   6 27445 13.91% 74 Revenge 3.0 avx2
   7 25705 16.73% 75 RubiChess 230918 avx2
   8 25240 16.25% 72 KomodoDragon 3.3 avx2
   9 24410 11.64% 82 CSTal 2.0 avx2
  10 22180 14.30% 87 Ethereal 14.25 nnue
  11 20190 08.49% 78 Clover 6 avx2
  12 17430 11.45% 79 Caissa 1.14 avx2
  13 17155 09.84% 80 Berserk 12 avx2
  14 15935 06.85% 86 Koivisto 9.2 avx2
  15 11400 07.62% 75 Seer 2.7.0 avx2
  16 9825 09.78% 79 RofChade 3.1 avx2

Rebel EAS is fourth, but I can not comprehend that with the data in the 6 sac pgns.

Can you shed your light on that?

First of all: Why not using my new UHO-Top15 Ratinglist games, for a comparison to my EAS-results?

https://drive.google.com/file/d/1R_5ko7gOws4JbQzDqhT2PRNFfDXheZBW/view?usp=sharing

Sencond: The results seem not so different, when I look at Stockfish (always around 21% sacs, same as here). Sacs of Ethereal fit also well (14%). KomodoDragon is around 15-16%. Torch 1 has 15.6% in my full EAS-ratinglist and 16% here. A real comparison would need you to calculate my actual UHO-Top15 gamebase, not a complete outdated gamebase, IMHO.

As I mentioned before: My sac-search depends on a lot of different piece-distribution-patterns (152 for one pawn sacs !!!) in pgn-extract. When one of this patterns matches, but in the 8 ply range there is another capture, this sac can be
a) missed completely (because the ply-counter is resetted to 0 after a capture)
b) sac is (wrongly) recognized being a lower or higher sac, than it is in reality (= too less or too much EAS-points)

Not good, but this error will hit all engines at the same range, when enough games are played...

For my tool, this is not avoidable. But for your tool it is, when just counting pawn-units for each color, instead looking for many different piece-distribution-patterns.

And I would say, you should not try to get the same EAS-scores than my EAS-tool. This will never really happen (see above). Better build you own tool. Without pgn-extract and in a real programming-language, a copy & paste of my work will definitly fail. Just copy & paste my ideas and make a better coding of these ideas. This will definitly lead to different EAS-scores. But I am confident, the ranking of engines in the EAS-lists of your tool and my tool should not be very different. Fingers crossed...

Subject: Re: Enough is Enough Sat Sep 28, 2024 10:24 am

Thanks Stefan for all the information, I made up my mind, it's my opinion your sac code is more than good, there are some exceptions but they are rare and thus because of the volume won't make a difference and we never get it 100% right, me neither, I had a case of a queen sacrifice with a mate announcement of 12 moves (or so), is this really a queen sac? Opinions may differ. Let's say 95% of your cases are right. Good enough. The same applies for the shorties, I think I have something better, see below, but it ain't much. Something else are the bad draws, I think we will never agree Very Happy

Some data, the shorties first, my code left, your code on the right side.

Code:: |Engine Shorties List [game decided margin = 300] |
| |
|PGN database : pgn\uho_ratinglist_games.pgn | Engine Shorties List [Stefan]
|Calculate EAS : 61.800 |
| | Rank EAS shorts
|Engine Won Loss Perc EAS | ----------------------------------------
|Stockfish 231107 av 7439 614 92.4% 15588 | 1 54468 Torch 1 popavx2
|Stockfish 16 230630 7199 701 91.1% 15039 | 2 47939 Stockfish 16 230630
|Uralochka 3.40a avx 2122 5770 26.9% 12346 | 3 46014 Stockfish 231107 avx2
|Rebel EAS avx2 2263 5371 29.6% 11600 | 4 38673 KomodoDragon 3.3 avx2
|Torch 1 popavx2 6527 1368 82.7% 11505 | 5 35802 Revenge 3.0 avx2
|KomodoDragon 3.3 av 5846 1924 75.2% 11001 | 6 32352 Uralochka 3.40a avx2
|Revenge 3.0 avx2 1941 5767 25.2% 10888 | 7 23921 RubiChess 230918 avx2
|CSTal 2.0 avx2 2939 4564 39.2% 10300 | 8 19767 Seer 2.7.0 avx2
|Caissa 1.14 avx2 3153 4442 41.5% 9954 | 9 12230 Clover 6 avx2
|Seer 2.7.0 avx2 2245 5616 28.6% 9634 | 10 10752 Caissa 1.14 avx2
|Koivisto 9.2 avx2 2599 5006 34.2% 9264 | 11 10572 RofChade 3.1 avx2
|Clover 6 avx2 2239 5336 29.6% 9002 | 12 8623 CSTal 2.0 avx2
|Ethereal 14.25 nnue 4112 3367 55.0% 7588 | 13 6789 Berserk 12 avx2
|RubiChess 230918 av 3514 4009 46.7% 7313 | 14 6223 Rebel EAS avx2
|RofChade 3.1 avx2 2291 5583 29.1% 7282 | 15 436 Ethereal 14.25 nnue
|Berserk 12 avx2 5377 2368 69.4% 6354 | 16 0 Koivisto 9.2 avx2

Difference are not much except for Uralochka and Rebel that profit more.

BTW, speaking of Rebel, that's why I (for the moment) keep using the older version because they contain Cstal and Rebel EAS because I happen to know something about these two, what they can and what they can't. And secondly, Cstal is about 30 elo stronger than Rebel EAS, also an important fact for my judgements.

Oh, and then the bad-draws.......

Code:: |Bad draw List [game decided margin = 300] |
| | Bad draw List [Stefan]
|PGN database : pgn\uho_ratinglist_games.pgn |
|Count Bad Draws : 58.100 | bad
| | Rank EAS-Score draws Engine/player
|Engine Won Draw Loss Perc BAD | ---------------------------------------------------
|RofChade 3.1 avx2 2291 7126 5583 39.0% 33 | 1 90601 08.46% Stockfish 231107 avx2
|CSTal 2.0 avx2 2939 7497 4564 44.6% 23 | 2 84681 09.65% Stockfish 16 230630
|Rebel EAS avx2 2263 7366 5371 39.6% 16 | 3 64516 13.78% Torch 1 popavx2
|Seer 2.7.0 avx2 2245 7139 5616 38.8% 16 | 4 56169 15.78% Uralochka 3.40a avx2
|RubiChess 230918 av 3514 7477 4009 48.3% 15 | 5 56169 15.72% Revenge 3.0 avx2
|Revenge 3.0 avx2 1941 7292 5767 37.2% 14 | 6 56169 15.01% KomodoDragon 3.3 avx2
|Uralochka 3.40a avx 2122 7108 5770 37.8% 12 | 7 51984 16.49% Ethereal 14.25 nnue
|Koivisto 9.2 avx2 2599 7395 5006 42.0% 10 | 8 51984 16.06% RubiChess 230918 avx2
|Caissa 1.14 avx2 3153 7405 4442 45.7% 8 | 9 48400 17.85% CSTal 2.0 avx2
|Ethereal 14.25 nnue 4112 7521 3367 52.5% 7 | 10 44944 18.65% Berserk 12 avx2
|Clover 6 avx2 2239 7425 5336 39.7% 6 | 11 41616 19.03% Rebel EAS avx2
|Torch 1 popavx2 6527 7105 1368 67.2% 2 | 12 38809 20.66% Koivisto 9.2 avx2
|Berserk 12 avx2 5377 7255 2368 60.0% 1 | 13 38809 20.31% Caissa 1.14 avx2
|Stockfish 231107 av 7439 6947 614 72.7% 0 | 14 35721 21.24% Clover 6 avx2
|Stockfish 16 230630 7199 7100 701 71.7% 0 | 15 30625 23.35% RofChade 3.1 avx2
|KomodoDragon 3.3 av 5846 7230 1924 63.1% 0 | 16 26244 25.98% Seer 2.7.0 avx2
| |
|Counted 163 bad draws, press a key to continue... |

Differences.

1. I only counted 163 bad draws, you thousands and thousands, even Stockfish 16 has almost 10%.
2. I my case S16 has zero bad draws, because once it has a won position, it simply wins and never draws.
3. And that's what I measure, the moment an engine has a won position.

Where does this leave us is a good question, I don't know.

PS, I will create a similar overview with your current database.

Posts : 132 Join date : 2022-03-01 Location : Berlin

I think, we just mean something different, when using the term bad draw. But this is totally fine (but perhaps little bit confusing for others...). Would be boring, if you just reprogram my way of doing an EAS Tool. 2 different tools, measuring aggressiveness, are better than just one, built in 2 different programming languages... So, I cant wait to see your first release of your EAS Tool.
I have no doubt, your Tool will be much better than mine... My consolation will be, that I was the guy, who had the idea of an EAS Tool and made the first one, which already works quite good and fast. Which is the same for UHO Openings: Stockfish devs made their own, much bigger UHO set, but they made it, following my ideas and concept of UHO/gamepairs.

Please make a LINUX build, when you are done. In Stockfish Discord Vondele already asked for a Linux-version of my EAS Tool, which is of course impossible to do. But your tool could take over here and let the Stockfish devs perhaps tune Stockfish to more aggressive play or add an additional, smaller nnue-net, which is playing aggressive. It could be switched on/off with an UCI-option by the users. That would be just awesome, IMHO.