BoCC -- Beauty of Computer Chess

[pohl4711]

Whoops, web page sloppiness, clicking on the download button gives the wrong version, clicking on the text link is the good one, dated August 15, 6:08

Meanwhile I put the executable on my Google account, please re-download in case you are not sure.

Yes, this is the one, I downloaded 2.5 hours ago. Shows Rebel Extreme 1.1 (not CSTal 21.) and Copyright 2025. August 15, 6:09

[peter]

So I re-downloaded and repeated the match against Patricia5, I had posted with the wrong one Rebel-Extreme-1.1.exe here

, this time with new binary:

Score of Patricia5.0 vs RebelExtreme1.1: 288 - 32 - 180 [0.756]
Elo difference: 196.5 +/- 25.4, LOS: 100.0 %, DrawRatio: 36.0 %
500 of 500 games finished.

Still (and even 2 more) games for the normally necessary 50 wins missing on Rebel's side, yet to have an EAS- score again at all anyhow:

Code: Select all

                                 bad  avg.win 
Rank  EAS-Score  sacs   shorts  draws  moves  Engine/player 
-------------------------------------------------------------------
   1    123774  40.63%  03.13%  22.78%   78   RebelExtreme1.1   XXXXX WARNING: Not enough games, EAS-score not reliable [50+ wins and 30+ draws needed] XXXXX  
   2     63622  10.42%  03.82%  20.00%   62   Patricia5.0  

[pohl4711]

Ed, I have good news:

I had an idea, which I am convinced, should make it possible to detect early attacks without using any evals or anything new in my EAS-Tool.
How? Because I am an idiot, I totally forgot, that I used pgn-extract to build my UHO-openings and for doing this, I used the command --plylimit many times to cut off the games in the megabase.pgn file, because I only needed the first 16 plies for my work.
If I use this command (--plylimit) to cut off all won games at a threshhold movenumber (35 moves or so (testing needed here, perhaps it canbe calculated out of the overall lentgh of all won games - this is always better than a hardcoded gamelength)) by an engine and then just doing my (superfast) sac-search again on this cutted games. All sacs, which are still found, now, must be happening early in the game! Totally simple...and fast.
This makes it possible to build an "early-attack" EAS-point-bonus for each engine, very similar to your king-attack detection, when looking into the results, hopefully. And your new shortie-detector can be used here, too, to add more short attacking games - I will definitly try it. The more short attacking games are found, the better. Finally all double-games will of course be deleted.

The bad news is: This needs a complete re-build of the scoring-system of the EAS-tool. And additionally, I want to make a third list in the output, containing the EAS-points of the single stats, not percents, as you wanted. And, I want the EAS-Tool to output the found (interesting) games in just one pgn-file, well sorted, as in my InterestingWinsSearch-Tool.

So, this is not a quick hack. It will be a huge update. Hopefully finished in 2025. No promises here. It is impossible to predict, how long the re-build of the scoring-system will take. Can be a week or 6 months... combining several exponential scorings into just one number, is a nightmare...especially preventing extreme results, skyrocketing to absurd numbers.

[Rebel]

https://rebel7775.wixsite.com/rebel/downloads

August 18, 2025
How-Rebel-Extreme-1.1-plays-chess, a collection from the 105.000 games it played divided in Aggressive King-Attacks, Short wins and material [Knight, Bishop, Rook and Queen] sacrifices.

[Rebel]

Ed, I have good news:

I had an idea, which I am convinced, should make it possible to detect early attacks without using any evals or anything new in my EAS-Tool.
How? Because I am an idiot, I totally forgot, that I used pgn-extract to build my UHO-openings and for doing this, I used the command --plylimit many times to cut off the games in the megabase.pgn file, because I only needed the first 16 plies for my work.
If I use this command (--plylimit) to cut off all won games at a threshhold movenumber (35 moves or so (testing needed here, perhaps it canbe calculated out of the overall lentgh of all won games - this is always better than a hardcoded gamelength)) by an engine and then just doing my (superfast) sac-search again on this cutted games. All sacs, which are still found, now, must be happening early in the game! Totally simple...and fast.
This makes it possible to build an "early-attack" EAS-point-bonus for each engine, very similar to your king-attack detection, when looking into the results, hopefully. And your new shortie-detector can be used here, too, to add more short attacking games - I will definitly try it. The more short attacking games are found, the better. Finally all double-games will of course be deleted.

The bad news is: This needs a complete re-build of the scoring-system of the EAS-tool. And additionally, I want to make a third list in the output, containing the EAS-points of the single stats, not percents, as you wanted. And, I want the EAS-Tool to output the found (interesting) games in just one pgn-file, well sorted, as in my InterestingWinsSearch-Tool.

So, this is not a quick hack. It will be a huge update. Hopefully finished in 2025. No promises here. It is impossible to predict, how long the re-build of the scoring-system will take. Can be a week or 6 months... combining several exponential scorings into just one number, is a nightmare...especially preventing extreme results, skyrocketing to absurd numbers.

Great plans, lot's of work, especially because of the limitations of the script language you are using, but getting the most out of it, well done. In programming perfection is cruel.

I am toying with the idea to focus on human games only, King-Attacks and the simpleness of the fast PGN shortie evaluator, both need no comments.

But first I take a break.

[pohl4711]

Ed, I have good news:

I had an idea, which I am convinced, should make it possible to detect early attacks without using any evals or anything new in my EAS-Tool.
How? Because I am an idiot, I totally forgot, that I used pgn-extract to build my UHO-openings and for doing this, I used the command --plylimit many times to cut off the games in the megabase.pgn file, because I only needed the first 16 plies for my work.
If I use this command (--plylimit) to cut off all won games at a threshhold movenumber (35 moves or so (testing needed here, perhaps it canbe calculated out of the overall lentgh of all won games - this is always better than a hardcoded gamelength)) by an engine and then just doing my (superfast) sac-search again on this cutted games. All sacs, which are still found, now, must be happening early in the game! Totally simple...and fast.
This makes it possible to build an "early-attack" EAS-point-bonus for each engine, very similar to your king-attack detection, when looking into the results, hopefully. And your new shortie-detector can be used here, too, to add more short attacking games - I will definitly try it. The more short attacking games are found, the better. Finally all double-games will of course be deleted.

The bad news is: This needs a complete re-build of the scoring-system of the EAS-tool. And additionally, I want to make a third list in the output, containing the EAS-points of the single stats, not percents, as you wanted. And, I want the EAS-Tool to output the found (interesting) games in just one pgn-file, well sorted, as in my InterestingWinsSearch-Tool.

So, this is not a quick hack. It will be a huge update. Hopefully finished in 2025. No promises here. It is impossible to predict, how long the re-build of the scoring-system will take. Can be a week or 6 months... combining several exponential scorings into just one number, is a nightmare...especially preventing extreme results, skyrocketing to absurd numbers.

Great plans, lot's of work, especially because of the limitations of the script language you are using, but getting the most out of it, well done. In programming perfection is cruel.

I am toying with the idea to focus on human games only, King-Attacks and the simpleness of the fast PGN shortie evaluator, both need no comments.

But first I take a break.

Great idea going forward. Especially because my EAS-Tool does not work very reliable on human games, because of the main concept of my sac-search.

[mclane]

Many chess engines have huge problems to successfully play a gambit opening. If you force that gambit opening line, the engine often tries to get back material and to get back material what completely ruins the idea of the gambit.

Humans have not so much problems to follow the idea of a gambit opening.

[jefk]

Many chess engines have huge problems to successfully play a gambit opening

used to be true, but not anymore since Alfa-zero; see the book Gamechanger by GM M. Sadler.
After that came Leelazero which also doesn't care so much about a pawn with its Mcts, and
with the advent of the nnue method the gambit play for the alfabeta machines also is improved
i think; with some attacking engines P5 CST21 and Rebelextr i've seen some interesting games ,
starting from (good) gambits (but possibly the SF beast could do that as well especially against
much lower ranked engines of course).

The example you give of a game by the late IM Diemer is not indicative of high level play.
Nowadays also the Blackmar Diemer gambit is not considered very sound btw (ie one of the
best gambits) anymore as some BDG fans used to think; although not many people know
how to 'refute' it (i'.e. what's the best defense against it).

[Rebel]

ChessBase 18 can make player profiles among it a graphical about its playing style. It more or less is in sync with EAS and BoCC evaluations but more detailed. Consider the below screenshots.

[pohl4711]

Yeah, I saw this already on my machines. Problem here: It is done by Chessbase. Means: Looks pretty, but works like garbage.

Proof:

Risk-value of Igel (considering your screenshots) = 295. Highest value of all engines here. Higher than CSTal 2.1 EAS (215), Patricia 3.1 (228) = nonsense

Positional play: Patricia 3.1=120 , Stockfish 17.1 = 119 (consider, the Patricia net is more than 100x smaller than Stockfish) = nonsense

Endgame: Patricia 3.1 = 209, Stockfish 17.1 =189 (consider, the Patricia net is more than 100x smaller than Stockfish) = nonsense

So, some values are good, but some are obvious nonsense. So, nobody should rely on this tool by Chessbase... we can do this way better, Ed.