Experimental testrun of Revenge 1.0, testing my EAS-tool

[Whiskers]

Thanks for the post! It seems I was wrong, and it’s just a coincidence that the two strongest engines are at the top (although I do expect that being able to see winning tactics other engines can’t is an advantage when it comes to EAS).

My testing with Patricia has also so far shown that aggression is not necessarily correlated with strength: I had a version of Patricia 50 elo stronger than Willow 2.4 that was nonetheless completely outclassed in terms of EAS score; I made some dedicated changes to try and change that, and the updated Patricia had 3x the EAS rating it did before at the same strength.

Thanks again!

[pohl4711]

Thanks for the post! It seems I was wrong, and it’s just a coincidence that the two strongest engines are at the top (although I do expect that being able to see winning tactics other engines can’t is an advantage when it comes to EAS).

My testing with Patricia has also so far shown that aggression is not necessarily correlated with strength: I had a version of Patricia 50 elo stronger than Willow 2.4 that was nonetheless completely outclassed in terms of EAS score; I made some dedicated changes to try and change that, and the updated Patricia had 3x the EAS rating it did before at the same strength.

Thanks again!

I do thank you! It is a pleasure to read this post: Not being angry, being proven false about something is so unusual in these times - much appreciated.
And I agree, there is a (small) advantage for the real strong engines in an EAS-evaluation - no way to prevent this. I only wanted to show, that the advantage is smaller as people expected and that a real aggressive engine will always be recognized by the EAS-tool and it will get a high rank in an EAS-ratinglist...

And it is really a shame, that the wonderful Revenge 1.0 engine (by far the strongest really aggressive playing engine besides Stockfish) is so hard to get for the users today... Only way to get it seems to be: Buying Revenge on the revengechess-site, getting Revenge 3.0 and then asking the author for sending Revenge 1.0, too. Worked for me. But it is very difficult. But, if you dont have Revenge 1.0 - I strongly recommend to buy it. This engine is just playing wonderful. Dont miss it!

[Graham Banks]

Stefan - why not Revenge 3.0?

[pohl4711]

Stefan - why not Revenge 3.0?

Because I already tested Revenge 3.0...

In the full ratinglist you will find it:
https://www.sp-cc.de/files/uho_full_list.txt

Scroll down to see the EAS-full-ratinglist. There you see, Revenge 3.0 is only on rank 11 with an EAS-score of 121885. This is not a bad score, but Revenge 1.0 has an EAS-score 197919 and Revenge 1.0 is on rank 2, close behind Stockfish 16.
So, sadly Revenge 3.0 lost a lot of aggressiveness compared to Revenge 1.0. Especially the sacrifices: Revenge 1.0 won 31.18% of its games with a sacrifice. Revenge 3.0 won only 13.18% of its games playing a sacrifice...
But I have to be clear here: This is not only a problem of Revenge. Most engines lose more and more of their aggressiveness, when increasing their strength.

[Frank Quisinsky]

Stefan,

this is wrong!
You need the lost games...

For example:

Spark 1.0 = 2700 Elo

Spark 1.0 ... against 20 engines with 2700 Elo and 10.000 games played
500 won, 0 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3000 Elo and 10.000 games played
100 won, 100 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3300 Elo and 10.000 games played
0 won, 250 lost (quick wins below 50 moves)

With the final result that no other engine with 2700 Elo can beat against 3300 Elo only 250 lost games and it is possible that Spark 1.0 with 2700 Elo is more aggressive than Stockfish 16 (with 0 wins and only 250 lost games) if you compare engines with different Elo for the aggressivess.

If your tools work without quantity of lost games, we can do nothing with it.
To avoid a fast loss is 50% important for such statements.

And all that with quick wins ...
25 moves to mate
30 moves to mate
35 moves to mate
40 moves to mate
45 moves to mate

Looking for a sacrifice:
If 20 engines with 2700 Elo can make an average against 20 engines with 2700 Elo after 10.000 games from 500 sacrifice games, the same group of 20 engines with 2700 Elo can make an average against 20 engines with 3000 Elo after 10.000 games from 50 sacrifice games. Now you can evalute how strong for engine with 2700 Elo vs. 3000 Elo are 100 sacrifice games.

For the sacrifice games you need a table with points when you evaluate something.

At the end of the day you have
- number of short wins after x moves
- the number of sacrifice moves

somewhere like this ...

This is not enough to say that engine x is more aggressive than engine y.

The pawn structure is much more important as number of short wins after x moves or the number of sacrifice moves!!!!!

For an example:
- how many aggressive pawn moves against the king... most attacks come with earlier pawn attacks against the opponent's king.

And please note:
The results for 1+1 games can be completely different from the results for 5+3 games. For example, Minic lost many games very quickly with a 4+2 time control. You can search for a game that Minic lost very fast with a time control of 66+6 and you will never find it. Revenge is after my impressions from version 1.0 to 3.0 clearly improved, also with aggressivess. Such results you have written you get if your tools are not looking on the loses.

Much more interesting as 1:0 and 0:1 are today all the draws!
In how many attacking games the opponent are able to make 3-fold.

If an engine are strong in aggressivess and are able to win many games very fast, the same engine should be strong in the defence, menas will never lose many games very fast. But this is one point only, not more not less.

Best
Frank

[Frank Quisinsky]

A tool for the early middlegame, the middlegame, the early endgame.
In the end it makes more sense!

If I want to give the information to others ...
Engine A is completely bad, because Engine A produces a lot of fast losses is wrong.

Engine A has other strengths.
Many available engines are very strong in the late midgame or early endgame.

The information is not complete if we keep talking about the first phase of the game.

So let us look at Stromphrax!
In the late mid-game and early endgame the engine is in the TOP-10.

I think that if I create the stats by hand (a lot of work).

And the final information is more interesting for others:
Stromphrax has problems in the early midgame, but is very, very strong in the late midgame and early endgame.

Best
Frank

[carldaman]

Stefan,

this is wrong!
You need the lost games...

For example:

Spark 1.0 = 2700 Elo

Spark 1.0 ... against 20 engines with 2700 Elo and 10.000 games played
500 won, 0 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3000 Elo and 10.000 games played
100 won, 100 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3300 Elo and 10.000 games played
0 won, 250 lost (quick wins below 50 moves)

With the final result that no other engine with 2700 Elo can beat against 3300 Elo only 250 lost games and it is possible that Spark 1.0 with 2700 Elo is more aggressive than Stockfish 16 (with 0 wins and only 250 lost games) if you compare engines with different Elo for the aggressivess.

If your tools work without quantity of lost games, we can do nothing with it.
To avoid a fast loss is 50% important for such statements.

And all that with quick wins ...
25 moves to mate
30 moves to mate
35 moves to mate
40 moves to mate
45 moves to mate

Looking for a sacrifice:
If 20 engines with 2700 Elo can make an average against 20 engines with 2700 Elo after 10.000 games from 500 sacrifice games, the same group of 20 engines with 2700 Elo can make an average against 20 engines with 3000 Elo after 10.000 games from 50 sacrifice games. Now you can evalute how strong for engine with 2700 Elo vs. 3000 Elo are 100 sacrifice games.

For the sacrifice games you need a table with points when you evaluate something.

At the end of the day you have
- number of short wins after x moves
- the number of sacrifice moves

somewhere like this ...

This is not enough to say that engine x is more aggressive than engine y.

The pawn structure is much more important as number of short wins after x moves or the number of sacrifice moves!!!!!

For an example:
- how many aggressive pawn moves against the king... most attacks come with earlier pawn attacks against the opponent's king.

And please note:
The results for 1+1 games can be completely different from the results for 5+3 games. For example, Minic lost many games very quickly with a 4+2 time control. You can search for a game that Minic lost very fast with a time control of 66+6 and you will never find it. Revenge is after my impressions from version 1.0 to 3.0 clearly improved, also with aggressivess. Such results you have written you get if your tools are not looking on the loses.

Much more interesting as 1:0 and 0:1 are today all the draws!
In how many attacking games the opponent are able to make 3-fold.

If an engine are strong in aggressivess and are able to win many games very fast, the same engine should be strong in the defence, menas will never lose many games very fast. But this is one point only, not more not less.

Best
Frank

If you compare games vs common opponents, with the same openings, side by side, it becomes noticeable that Revenge 1.0 is more aggressive.
Revenge-v3 is stronger because it plays safer chess.

Some of the top engines play way too safely for their strength, especially with Black. The problem really shows up when facing much weaker opposition (400+ Elo diff).

[BrendanJNorman]

Stefan,

this is wrong!
You need the lost games...

For example:

Spark 1.0 = 2700 Elo

Spark 1.0 ... against 20 engines with 2700 Elo and 10.000 games played
500 won, 0 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3000 Elo and 10.000 games played
100 won, 100 lost (quick wins below 50 moves)

Spark 1.0 ... against 20 engines with 3300 Elo and 10.000 games played
0 won, 250 lost (quick wins below 50 moves)

With the final result that no other engine with 2700 Elo can beat against 3300 Elo only 250 lost games and it is possible that Spark 1.0 with 2700 Elo is more aggressive than Stockfish 16 (with 0 wins and only 250 lost games) if you compare engines with different Elo for the aggressivess.

If your tools work without quantity of lost games, we can do nothing with it.
To avoid a fast loss is 50% important for such statements.

And all that with quick wins ...
25 moves to mate
30 moves to mate
35 moves to mate
40 moves to mate
45 moves to mate

Looking for a sacrifice:
If 20 engines with 2700 Elo can make an average against 20 engines with 2700 Elo after 10.000 games from 500 sacrifice games, the same group of 20 engines with 2700 Elo can make an average against 20 engines with 3000 Elo after 10.000 games from 50 sacrifice games. Now you can evalute how strong for engine with 2700 Elo vs. 3000 Elo are 100 sacrifice games.

For the sacrifice games you need a table with points when you evaluate something.

At the end of the day you have
- number of short wins after x moves
- the number of sacrifice moves

somewhere like this ...

This is not enough to say that engine x is more aggressive than engine y.

The pawn structure is much more important as number of short wins after x moves or the number of sacrifice moves!!!!!

For an example:
- how many aggressive pawn moves against the king... most attacks come with earlier pawn attacks against the opponent's king.

And please note:
The results for 1+1 games can be completely different from the results for 5+3 games. For example, Minic lost many games very quickly with a 4+2 time control. You can search for a game that Minic lost very fast with a time control of 66+6 and you will never find it. Revenge is after my impressions from version 1.0 to 3.0 clearly improved, also with aggressivess. Such results you have written you get if your tools are not looking on the loses.

Much more interesting as 1:0 and 0:1 are today all the draws!
In how many attacking games the opponent are able to make 3-fold.

If an engine are strong in aggressivess and are able to win many games very fast, the same engine should be strong in the defence, menas will never lose many games very fast. But this is one point only, not more not less.

Best
Frank

If you compare games vs common opponents, with the same openings, side by side, it becomes noticeable that Revenge 1.0 is more aggressive.
Revenge-v3 is stronger because it plays safer chess.

Some of the top engines play way too safely for their strength, especially with Black. The problem really shows up when facing much weaker opposition (400+ Elo diff).

Can Revenge 1 still be bought? I wanna give it a test drive.

[Frank Quisinsky]

Hi there,

I can't see such things with Revenge 1.0.
The fact is that Pedone 3.0 / 3.1 / Revenge 1.0 / 2.0 / 3.0 are all very strong in aggressiveness.
But Revenge 3.0 loses significantly less fast games and that is in my opinion important for the questions of aggressiveness to 50%!

Possible with very short time controls and very boring UHU positions, which Stefan likes to do.
The results should be right under these conditions and his tools with the possibilities the tools have.

Assessing whether an engine is able to play aggressively in a balanced way, i.e. whether
the style of play itself does not expose many tactical gaps.

Best
Frank

Note:
Engines like Revenge or Uralochka, SlowChess, Igel ... reduce the gap to number 1 with longer time controls significanty. And the stats are quiet different to very fast time-controls.

[pohl4711]

Can Revenge 1 still be bought? I wanna give it a test drive.

Not directly. You have to buy Revenge 3.0 and write an email to the author and ask for Revenge 1.0.
If you already purchased Revenge, just contact the author. Worked for me.

support@revengechess.com