If there were one engine I could bring back and covert

[Ovyron]

4) Deep Junior (it is not UCI) => it is UCI since long

But not the version we want. Deep Junior UCI is but a shadow of the style we used to love, just like it happened to Hiarcs, Shredder, and Fritz 11. It's just called "Junior", and it's stronger than Junior, but it's something else.

So I guess I'll throw in my vote for Fritz 10.1, because I actually liked its style better than Junior's, and I find unbelievable that people managed to crack the CTG opening book's code so such books can be used anywhere, yet nobody has been able to crack old fritz GUI engine protocol (which they don't even use anymore...) If they did all those engines could be used with some FritzGUI2UCI adapter.

Though, nowadays we've been flooded with freeware engines that perhaps would play stronger and with a better style than Fritz 10.1, but they're so many it's hard to know. Those engines from the past are remembered fondly because that's all we had, and I even wonder if the style of some engines doesn't hold up to today's standards.

[mclane]

The engines in the past are remembered because they were different.

Today's engines are almost all similar.

The only thing the engines today have originally is the name.

That was different in old days.

Junior, wchess, mchess, rebel, CSTal, the king, genius, Mephisto III, hiarcs, virtual chess, nimzo, Gandalf .... Chess tiger, Colossus chess, Sargon, philidor chess,

...were all very unique different approaches.

That made computer chess unique and interesting. Today the engines are all similar.


It makes not much sense to test those very similar engines anymore.

Mchess or the king or rebel or Mephisto III were so amazingly different that the games were unbelievably interesting.

Today's programs play too much similar.

[Sergio Martinez]

If I could choose one it would be Ferret.

I know it was a private engine and compatible with Winboard but it was never released publicly.

It would be nice to be able to use it today

[Ovyron]

Well, apparently speaking about engine similarity has become taboo nowadays, with censorship and moved posts and topics running amok. But what about Ethereal 11.50 which shows some 22% similarity with other engines and even with earlier versions of itself? According to this if you're on a position with many playable moves Ethereal would play an original move, different from others, 4 out of five times.

What this elo difference have given us is the ability to sacrifice it to make the engines play differently. If Stockfish is near 3500 ELO, then you could sacrifice 500 ELO and make a personality that plays in the style of CSTal, or MChess or Mephisto or whathave you but at a much higher level (say, instead of having to wait for 1 hour modified Stockfish would play that move within seconds.) Heck, we could make some engine that sacrifices 800 elo and plays a Magnus Carlsen move in positions 9 out of 10 times.

The reason this hasn't happened yet is that eval tuning has always been done to increase ELO. Programmers don't care about move choices (except for PK) and just test changes and if results improve they commit them, and as it turns out in most chess positions there's moves better than others and stronger engines are more prone to switch to them, their similarity increases as they become stronger (except for Ethereal.)

If instead, there was a set of positions that represented the style of a player or engine, and the tuning was done to approximate the move choices, we wouldn't need ancient engines to be ported to UCI, and we could extract key positions from Paul Morphy's games and have engines play like him.

Because in the end, it doesn't matter what method is used to arrive to those moves. People always clamour about their favorite engines from the past, but they need to show what made them special, the positions where they shined, with original move choices that today's approaches ignore. Only then someone will be able to build something that plays those moves in those positions.

[mclane]

In older days the engines were kind of twin brother of the programmer.
They showed the personality of the programmer.

Today in times of strong open source codes, the programs are no twins of the programmers anymore.
They are kind of clone of the public available sources.

That’s making in uninteresting.

The old chess programmes played very unique and completely different.
Believe me.

I am not exaggerating.

They used completely different approaches that had to do with the programmers experience.

[Guenther]

Well, apparently speaking about engine similarity has become taboo nowadays, with censorship and moved posts and topics running amok.

...

You never get around something? You can warm this up now for years, if it makes you happy.

But what about Ethereal 11.50 which shows some 22% similarity with other engines and even with earlier versions of itself? According to this if you're on a position with many playable moves Ethereal would play an original move, different from others, 4 out of five times.

...

You read something somewhere, but you don't understand it...

The very low similarity percentage for Ethereal (and probably some others) happens only at very low depths
and results in quite weak play, which doesn't matter, as those depths are gone in millisec.

It simply does pruning from the root until real search kicks in at higher depths. (simplified for you)

[carldaman]

What this elo difference have given us is the ability to sacrifice it to make the engines play differently. If Stockfish is near 3500 ELO, then you could sacrifice 500 ELO and make a personality that plays in the style of CSTal, or MChess or Mephisto or whathave you but at a much higher level (say, instead of having to wait for 1 hour modified Stockfish would play that move within seconds.) Heck, we could make some engine that sacrifices 800 elo and plays a Magnus Carlsen move in positions 9 out of 10 times.

The reason this hasn't happened yet is that eval tuning has always been done to increase ELO. Programmers don't care about move choices (except for PK) and just test changes and if results improve they commit them, and as it turns out in most chess positions there's moves better than others and stronger engines are more prone to switch to them, their similarity increases as they become stronger (except for Ethereal.)

Yes, I totally agree here. I call it "trading Elo/strength for style". It is a great trade-off and I've been a big proponent of this over the last few years. The huge strength of today's engines can be used as some sort of 'currency' to convert back into chess style.

Pawel Koziol deserves a lot of credit blazing this path with Rodent/OpenTal, and now we're seeing commercial Komodo introducing engine personalities, even tweakable ones. It looks very promising.
I hope that they don't turn away from it.

[Ovyron]

Pawel Koziol deserves a lot of credit blazing this path with Rodent/OpenTal, and now we're seeing commercial Komodo introducing engine personalities, even tweakable ones. It looks very promising.
I hope that they don't turn away from it.

Yes, what I've been missing is an objective way that allows one to define it, what makes a style a style?

Like, I've seen people praise Rodent's Karpov personality, about how it feels very Karpovish, and about how they couldn't tell the difference between a Karpov's game and a Rodent Karpov's game. What is missing is some positions shown where Karpov played some unique moves and the personality plays them too, otherwise, if someone else can come up with another one that looks like Karpov's but plays completely different moves we have achieved nothing.

While chess engine personalities feels like a small niche inside a small niche, it's clear all the people in this thread wanting to "bring back" "classic" chess engines just want to see the chess moves they remember again, so we couldn't just build chess engine personalities that play like famous human GMs but also ones that play like those engines, and I wonder if they would be happy with some Rodent Junior (that plays like Junior 10) or Rodent CSTal personality that showcase the same style, and if producing them would even be possible, or if they'd forcefully want the real thing, since in such case trying to imitate the style of old engines would be useless.

[PeterO]

Friends,

you are ABSOLETELY right!!! Lets trade Elo against style!
For me is RODENT IV a new engine generation. Lets create facinating unique personalities which we can PLAY with!

I REALLY hope somebody can compile Rodent IV for android. So we can use it even with the wooden ChessGenius Exclusive Board - and Aart Biks - Chess for Android - app. Its just a dream to play the engines on the big wooden board!!!

I think the next years are the REAL GOLDEN AGE for computerchess!!

[tmokonen]

So I guess I'll throw in my vote for Fritz 10.1, because I actually liked its style better than Junior's, and I find unbelievable that people managed to crack the CTG opening book's code so such books can be used anywhere, yet nobody has been able to crack old fritz GUI engine protocol (which they don't even use anymore...) If they did all those engines could be used with some FritzGUI2UCI adapter.

I can't imagine an adapter of that sort would be hard to figure out for someone who is decent at x86 assembler. I am terrible at assembly language (I am a worthless loser at everything), but even I was able to figure out what some of the functions inside an .eng file were doing and the parameters being passed to them when I disassembled a few .eng files with IDA.