Xiphos x Ethereal: Who has the best style?

[carldaman]

Possible that an engines, produced a nice style is more or less random!

Yeah, one of the engines with the most impressive playing styles was Thinker, but Lance Perkins was unaware of this and it's clear the style was completely accidental. The development of the engine was the typical one focused on "ELO chasing" and the height of stylistic choices was a serendipity.

When Kerwin Medina took over the development of the engine, he had no idea about the amazing style, and even created a version to "improve the style" called Active.

I didn't like Active because it was more of a suicidal engine that couldn't hold its own against anything, while Thinker Inert at its peak could be the top freeware engine and beat all the others.

While this is an example of engines with great style achieving it at random, Eelco De Groot's Bluefish showed how you can achieve great playing style while keeping strength, with Bluefish outperforming Stockfish's default in many tournaments, and today it's very easy to apply Bluefish's settings to Stockfish dev, and have an engine with a great style that plays stronger than Stockfish 11, and of course, Xiphos and Ethereal.

So you don't need to sacrifice style for strength, but perhaps it's required that engine developers take a look at the games their engines play, and why, instead of relying on a method that just measures 0.01 elo advantage over default so the change is accepted without caring if the engine had to sacrifice its soul for it.

Even Thinker suffered this, Thinker 5.3B was the "last good Thinker" regarding style, providing the nicest games to watch from all the versions. This was lost on future versions, but because the CCRL lists Thinker 5.4D as the highest elo one, that's the one people download, and they miss out on extraordinary games.

Even I'm guilty of this, the reason I tested Xiphos and Ethereal was because they're at the top of ELO, if Xipho's programmer put attention into keeping some face for Xiphos and sacrificed its strength for it, perhaps it'd have been below RofChade, and I'd have tested it instead. It's like a battle that can't be won because chess itself is all about the games that you can win...

Extremely well put, Ovy!

Unfortunately, not every developer is also an accomplished chess player, to look at the actual games and critique them from a stylistic point of view.

[AndrewGrant]

look at the actual games and critique them from a stylistic point of view.

Do you really think you, or anyone else, is capable of viewing a large enough sample to be able to determine a change in (alleged) style from a change that ultimately impacts an engine by a few elo? Its not like people are running around deleting code with comments like "// This is important for style, but loses elo!".

[AndrewGrant]

look at the actual games and critique them from a stylistic point of view.

Do you really think you, or anyone else, is capable of viewing a large enough sample to be able to determine a change in (alleged) style from a change that ultimately impacts an engine by a few elo? Its not like people are running around deleting code with comments like "// This is important for style, but loses elo!".

That really is not meant to be patronizing, by the way. I want to know whether people genuinely believe that they can identify minor style changes as a result of viewing some games. Sure, I can imagine an extreme case, where you force a playstyle by massively overrating a positional feature, but is there a real, typical case?

It was said here that programmers decide to chase elo at the exchange of style. That means that one can directly code style. I would like to see some examples which don't fall under the extreme case above.

[Dr.Wael Deeb]

look at the actual games and critique them from a stylistic point of view.

Do you really think you, or anyone else, is capable of viewing a large enough sample to be able to determine a change in (alleged) style from a change that ultimately impacts an engine by a few elo? Its not like people are running around deleting code with comments like "// This is important for style, but loses elo!".

That really is not meant to be patronizing, by the way. I want to know whether people genuinely believe that they can identify minor style changes as a result of viewing some games. Sure, I can imagine an extreme case, where you force a playstyle by massively overrating a positional feature, but is there a real, typical case?

It was said here that programmers decide to chase elo at the exchange of style. That means that one can directly code style. I would like to see some examples which don't fall under the extreme case above.

Minor style changes are nearly impossible to detect even if examined by a team of top human grandmasters working closely with the programmer ....

I was away from the computer chess scene for a while now but I returned with even bigger interest and because maybe I am one of the few people here still playing against the chess engines rated from 2400 - 2700 Elo,I can say that you must play at least 20 games against a chess engine to even begin to grasp it's superiority or weak spots let alone it's playing style ....

Playing chess will train your brain cells and keep it's activity beside training in the gym ....

Forever young regards,
Dr.D

[Dann Corbit]

There are lots of fun styles in chess. Pirates over the wall of Tal. Slow, deliberate strangulation of Bottvinik.
When it comes to chess engines, the best style (IMO) is the style that wins the most.
So I guess that is SF and LC0.
But I like other engines and the fact that they play differently is a big bonus.
I have both Xiphos and Ethereal on my system and I use them both. Neither is very good at tactical problem solving, but they play well which is (I suppose) what counts the most.

[Rebel]

look at the actual games and critique them from a stylistic point of view.

Do you really think you, or anyone else, is capable of viewing a large enough sample to be able to determine a change in (alleged) style from a change that ultimately impacts an engine by a few elo? Its not like people are running around deleting code with comments like "// This is important for style, but loses elo!".

I know this is not interesting for you since your focus is different but Benjamin is an example, it's the gambit version of ProDeo, in self-play it loses 28 elo and is lower rated than ProDeo. But against much stronger engines it performs much much better than ProDeo. Style can do that.

[Ovyron]

That means that one can directly code style. I would like to see some examples which don't fall under the extreme case above.

Rybka 3 Dynamic - Vasik Rajlich added code that favored positions with material imbalances. As opposed to having code that likes, say, Knight+Bishop over Rook+Pawn, or viceversa, you make it add a bonus for future positions where those imbalances are on the board, no matter what side you're on. Similar to today's Contempt of Stockfish where it analyses its own side with positive contempt and the opponent's side with negative content, so contempt ON makes the engine analyze white and black differently, here Rybka Dynamic thought the opponent wanted to avoid material imbalances.

It was the engine with the best style of its epoch, at the cost of 30 elo below default. Back then it was afforded because Rybka's advantage over the rest was huge, so Dynamic could prevail as stronger than everyone else. On future versions Vas decided to just allow the users to create their own styles by modifying the pieces's values of each color independently, but by then Houdini and Stockfish (and even Critter) were taking over and this wasn't explored much (codeword Mindbreaker.)

If you added a Dynamic Parameter to Ethereal that added a bonus to leaf positions where material is irregular, I'm sure it'd work as a "Style" setting, as pumping it up would make it more fireworks appear on the board.

Though of course asking Ed Schröder how Benjamin works could bring a more sophisticated style.

[Rebel]

That means that one can directly code style. I would like to see some examples which don't fall under the extreme case above.

Rybka 3 Dynamic - Vasik Rajlich added code that favored positions with material imbalances. As opposed to having code that likes, say, Knight+Bishop over Rook+Pawn, or viceversa, you make it add a bonus for future positions where those imbalances are on the board, no matter what side you're on. Similar to today's Contempt of Stockfish where it analyses its own side with positive contempt and the opponent's side with negative content, so contempt ON makes the engine analyze white and black differently, here Rybka Dynamic thought the opponent wanted to avoid material imbalances.

It was the engine with the best style of its epoch, at the cost of 30 elo below default. Back then it was afforded because Rybka's advantage over the rest was huge, so Dynamic could prevail as stronger than everyone else. On future versions Vas decided to just allow the users to create their own styles by modifying the pieces's values of each color independently, but by then Houdini and Stockfish (and even Critter) were taking over and this wasn't explored much (codeword Mindbreaker.)

If you added a Dynamic Parameter to Ethereal that added a bonus to leaf positions where material is irregular, I'm sure it'd work as a "Style" setting, as pumping it up would make it more fireworks appear on the board.

Though of course asking Ed Schröder how Benjamin works could bring a more sophisticated style.

Works only against much stronger engines, else regression.

[Ovyron]

Works only against much stronger engines, else regression.

That's why I make them play at the same strength before testing, that way the style can be seen without any "stronger" or "weaker" in the formula. 50% performance against an opponent no matter how crazy the engine plays should give very high margin to showoff style.

Anyway, what I meant to say was that you could give instructions so people could program Benjamin's style setting into their engines, though I can understand why you'd want for it to be your engine's exclusive.

[Rebel]

Works only against much stronger engines, else regression.

That's why I make them play at the same strength before testing, that way the style can be seen without any "stronger" or "weaker" in the formula. 50% performance against an opponent no matter how crazy the engine plays should give very high margin to showoff style.

Anyway, what I meant to say was that you could give instructions so people could program Benjamin's style setting into their engines, though I can understand why you'd want for it to be your engine's exclusive.

It's not a secret, a few parameter changes plus some new code, 10-15 minutes work. The art is to tune it with a minimum loss in strength, that took weeks

But no programmer in active competition is interested to put energy in styling, elo is all.