Patricia 2.0 - likely the most aggressive chess engine ever made

[PeterO]

Hello,

I ABSOLUTELY agree with BrendanJNorman: „ Polish style/features (levels, multiPV etc) - Elo be damned!“

Nobody is interested in higher Elo! We want to see a fascinating playing style!

The MOST important feature for ME: I want to set a Elo Number in Patricia - for example Elo 1800. And play against it. AND - Elo 1800 should REALLY mean - HUMAN Elo 1800. I think a combination of your idea:
1. Random-„Human Blunders“ + 2. (Pohls Suggestion) node-limit would be a DREAM!

Thank you for your engine!! Peter

[Whiskers]

Hello,

I ABSOLUTELY agree with BrendanJNorman: „ Polish style/features (levels, multiPV etc) - Elo be damned!“

Nobody is interested in higher Elo! We want to see a fascinating playing style!

The MOST important feature for ME: I want to set a Elo Number in Patricia - for example Elo 1800. And play against it. AND - Elo 1800 should REALLY mean - HUMAN Elo 1800. I think a combination of your idea:
1. Random-„Human Blunders“ + 2. (Pohls Suggestion) node-limit would be a DREAM!

Thank you for your engine!! Peter

Dev Patricia already has some skill level options, set to CCRL elo - 50 at 60+0.6 and weaker at longer time controls that a human would likely play at. I feel it is a decent approximation of human elo at blitz-rapid TCs, and it’s all based on node and depth limits in order to not be artificially weak.

[pohl4711]

I did a testrun of Patricia 2.0 vs. 10 engines (strength around Rybka 4.1), which fits the strength of Patricia 2.0 very well, for an EAS-calculation (short wins move-limit fixed to 60 moves).
10000 games, 3min+1sec, singlethread. Balanced openings, engines start thinking from move 9.

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

Download the played games and 874 spectacular wins of Patricia 2.0 (with at least a 2 pawn-units sac) (filtered by my Interesting Wins Search Tool) directly:
https://www.sp-cc.de/files/patricia_2.0_test.7z

Code: Select all

                                 bad  avg.win 
Rank  EAS-Score  sacs   shorts  draws  moves  Engine/player 
-------------------------------------------------------------------
   1    337941  44.01%  36.20%  04.84%   67   Patricia 2.0  
   2    134485  07.65%  54.34%  33.84%   61   Critter 1.6a  
   3    130651  02.60%  53.76%  35.44%   60   Rybka 4.1  
   4    125405  03.57%  55.61%  34.45%   61   Andscacs 0.88  
   5    121305  04.28%  56.03%  32.84%   61   Laser 1.5  
   6    106943  03.61%  45.10%  30.06%   64   Komodo 5  
   7    100754  04.08%  47.96%  36.93%   64   Houdini 1.5a  
   8    100360  05.03%  46.15%  35.51%   63   Texel 1.7  
   9     97459  03.85%  46.15%  36.69%   63   Nirvanachess 2.4  
  10     84082  04.61%  43.80%  37.65%   65   Hannibal 1.7  
  11     67447  01.51%  29.06%  24.70%   72   Princhess 0.16  
-------------------------------------------------------------------
*** Average length of all won games:     65 moves

An EAS-Score of 337941 is very impressive. No other engine (except the weak OpenTal engine) has ever achieved more than 300000 EAS-points in my testings !!!

[BrendanJNorman]

No other engine (except the weak OpenTal engine) has ever achieved more than 300000 EAS-points in my testings !!!

The "weak OpenTal" engine was created so that an even weaker (2200ish) human player (me) could have an aggressive GM-level sparring partner.

Not to battle with superhuman chess engines.

It is still comfortably above 2500 Elo level though, which was enough for me.

[pohl4711]

No other engine (except the weak OpenTal engine) has ever achieved more than 300000 EAS-points in my testings !!!

The "weak OpenTal" engine was created so that an even weaker (2200ish) human player (me) could have an aggressive GM-level sparring partner.

Not to battle with superhuman chess engines.

Correct. I just wanted to point out, that OpenTal exists and plays also very aggressive. But OpenTal has no neural-net (it is a Rodent-modification) and so, it often plays sacrifices, which definitly not fit the position on the board... Patricia is not only much stronger (+700 Elo?!), but has a neural-net and plays its aggressive style with much more understanding of the board-position.

So, Patricia 2.0 is, right now, the most aggressive playing engine, which is on superhuman Elo-level and has a neural-net for much better positional understanding.

[BrendanJNorman]

No other engine (except the weak OpenTal engine) has ever achieved more than 300000 EAS-points in my testings !!!

The "weak OpenTal" engine was created so that an even weaker (2200ish) human player (me) could have an aggressive GM-level sparring partner.

Not to battle with superhuman chess engines.

Correct. I just wanted to point out, that OpenTal exists and plays also very aggressive. But OpenTal has no neural-net (it is a Rodent-modification) and so, it often plays sacrifices, which definitly not fit the position on the board... Patricia is not only much stronger (+700 Elo?!), but has a neural-net and plays its aggressive style with much more understanding of the board-position.

So, Patricia 2.0 is, right now, the most aggressive playing engine, which is on superhuman Elo-level and has a neural-net for much better positional understanding.

Everything you say is absolutely correct.

OpenTal is an aggressive beast, but with all the usual HCE positional weaknesses/lack of knowledge.

Patricia takes the aggression to a far more human-like attacking level.

Can't wait to mess around and play blitz vs a weakened version.

[Whiskers]

I did a testrun of Patricia 2.0 vs. 10 engines (strength around Rybka 4.1), which fits the strength of Patricia 2.0 very well, for an EAS-calculation (short wins move-limit fixed to 60 moves).
10000 games, 3min+1sec, singlethread. Balanced openings, engines start thinking from move 9.

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

Download the played games and 874 spectacular wins of Patricia 2.0 (with at least a 2 pawn-units sac) (filtered by my Interesting Wins Search Tool) directly:
https://www.sp-cc.de/files/patricia_2.0_test.7z

Code: Select all

                                 bad  avg.win 
Rank  EAS-Score  sacs   shorts  draws  moves  Engine/player 
-------------------------------------------------------------------
   1    337941  44.01%  36.20%  04.84%   67   Patricia 2.0  
   2    134485  07.65%  54.34%  33.84%   61   Critter 1.6a  
   3    130651  02.60%  53.76%  35.44%   60   Rybka 4.1  
   4    125405  03.57%  55.61%  34.45%   61   Andscacs 0.88  
   5    121305  04.28%  56.03%  32.84%   61   Laser 1.5  
   6    106943  03.61%  45.10%  30.06%   64   Komodo 5  
   7    100754  04.08%  47.96%  36.93%   64   Houdini 1.5a  
   8    100360  05.03%  46.15%  35.51%   63   Texel 1.7  
   9     97459  03.85%  46.15%  36.69%   63   Nirvanachess 2.4  
  10     84082  04.61%  43.80%  37.65%   65   Hannibal 1.7  
  11     67447  01.51%  29.06%  24.70%   72   Princhess 0.16  
-------------------------------------------------------------------
*** Average length of all won games:     65 moves

An EAS-Score of 337941 is very impressive. No other engine (except the weak OpenTal engine) has ever achieved more than 300000 EAS-points in my testings !!!

Thank you very much! I really appreciate the test you did, as well as the page about Patricia on your website, and I hope I will find good use for all the games that were played.

[Werewolf]

Is this a search bug or just part of the way it plays? Patricia was White. Black was a 2000 Elo bot.
Game in 15.
Black wins.

[pgn]1. e4 c5 2. b4 cxb4 3. a3 Qa5 4. Nf3 e5 5. d4 Nf6 6. Nxe5 Nxe4 7. Bd3 bxa3+ 8. c3 Nxc3 9. Qd2 Bb4 10. O-O Nxb1 11. Qg5 Rf8 12. Qxg7 a2 13. Bg5 h6 14. Bf6 d5 15. Re1?? Bxe1 16. Rxa2?? Qxa2 17. Nxf7 Rf7
0-1[/pgn]

0-1

[towforce]

Is this a search bug or just part of the way it plays? Patricia was White. Black was a 2000 Elo bot.
Game in 15.
Black wins.

[pgn]1. e4 c5 2. b4 cxb4 3. a3 Qa5 4. Nf3 e5 5. d4 Nf6 6. Nxe5 Nxe4 7. Bd3 bxa3+ 8. c3 Nxc3 9. Qd2 Bb4 10. O-O Nxb1 11. Qg5 Rf8 12. Qxg7 a2 13. Bg5 h6 14. Bf6 d5 15. Re1?? Bxe1 16. Rxa2?? Qxa2 17. Nxf7 Rf7
0-1[/pgn]

0-1

Final position is tantalisingly close: black gets mate just before Patricia does!

[BrendanJNorman]

Is this a search bug or just part of the way it plays? Patricia was White. Black was a 2000 Elo bot.
Game in 15.
Black wins.

[pgn]1. e4 c5 2. b4 cxb4 3. a3 Qa5 4. Nf3 e5 5. d4 Nf6 6. Nxe5 Nxe4 7. Bd3 bxa3+ 8. c3 Nxc3 9. Qd2 Bb4 10. O-O Nxb1 11. Qg5 Rf8 12. Qxg7 a2 13. Bg5 h6 14. Bf6 d5 15. Re1?? Bxe1 16. Rxa2?? Qxa2 17. Nxf7 Rf7
0-1[/pgn]

0-1

White was completely winning before 15.Re1.

15.Nxf7!! wins, but Patricia's eval rewards bigger sacrifices, so I guess it "speculated" on this 15.Re1 move.

In other words, this is how it plays. Who cares if another engine was able to refute something?