What a fun in those times!

[PK]

When I started adding style parameters to Rodent, I noticed the following: for parameters that are separated correctly usually there is a range of good values rather than just one optimum. Separation is really the point. For example, modifying knight value by 5 centipawns had more consequences than changing all the material values (and I mean literally all: bishop pair, knight pair, imbalance bonuses and penalties etc.) by 5%.

What I really dream about is a set of parameter groups that are not mutually exclusive: for example having bishop pair both in material group and in a group containing pawn weaknesses (because the first bishop is often exchanged for opponent's concession in pawn structure). In a way, such tuning mechanism would resemble a small two-layered neural network.

Now for style: perhaps different playing styles have something to do with different parameter groupings? For example, the less interdependence between king safety and other factors, the more it can be tweaked without putting the entire evaluation off balance (a strange dependence story is shown here: https://sites.google.com/view/cdrill/personality/tuning - engine tuned on Morphy's games has king safety ludicrusly low because threats against enemy pieces seem to have the same effect on gameplay).

[Lyudmil Tsvetkov]

I do not find games matching engines couple of hundreds elo apart interesting at all: it is like a soccer game between Barcelona and a local town club, the mant goals from the stronger side do not make the game more interesting.

False equivalence.

(https://en.wikipedia.org/wiki/False_equivalence)

In my youth, I have been playing for 5-6 hours daily soccer, especially during classes, so I know extremely well what I am talking about.

[Lyudmil Tsvetkov]

I am no programmer, but...

I think sometimes what people have seen as 'style' is little more than how an engine operated with a zone of uncertainty.

That is to say that at one time evaluation functions bumped up against the horizon effect with uncertain results...with the hardware being so powerful these days, this is less of a problem and evaluation functions are likely to be closer between engines...because they see more.

Just my two cents.

There's probably some validity to this, based on what is generally observed, but... I have found that one can tweak current Komodo's settings and get plenty of style, while sacrificing relatively little strength.

(For example, I don't really care if the engine is rated 3100-3200 instead of 3300, but now has a very exciting attacking style with plenty of piece/pawn sacs).

Regards,
CL

Brian makes an interesting point.

And with regard to your comment Carl, you're essentially "weakening" (making Komodo's evaluation less accurate) Komodo by making it more "stylish".

Nothing wrong with that of course, we both know that "weakening" engines is pretty much all I do!

With that being said...

The same argument can be used across the board (no pun intended) with the same result.

In the games of humans we witness the same phenomenon.

At 1300-1500 level, games are so random and full of mistakes, that they are very exciting for other amateurs to watch. All intentions and plans are more or less clear and tactics not very deep. The outcome is ALWAYS uncertain and the "horizon" very low.

At the IM-Average GM level, we see a closer alignment of move choices, but still retain individual prejudices (all that a "style" is, is the material/positional/psychological prejudices we obey when choosing a move or plan) which give our game a unique character.

I could tell an Alekhine game from a Tal game for example, because of this "fingerprint", despite them both being attackers.

at this level, the "horizon" is much higher than the aforementioned amateurs, but still low enough to "speculate" or have your own unique approach.

At the 2780-2800+ level, these guys have already seen all of the ideas, all of the attacking structures, key games and plans.

They all calculate very deeply with extreme accuracy.

For this reason, the "horizon" is even higher than for the strong IM or average GM. They simply know in advance how a certain plan/idea/direction is going to end and choose accordingly.

This is why as Nakamura "matured" toward the 2800 level his style became less and less flashy. At that level, the styles more or less merge together (for the most part). The "fingerprint" is less distinct.

In computer chess, the top "players" improve at a greater speed than do the human players, so at the very top level the styles will become more and more merged. I think Brian is right.

But I don't think this is a bad thing.

I don't think the programs see more because of just hardware improvements.

The evaluation is also a big deal.

If I could calculate 25 moves ahead, I still doubt I could beat Carlsen.

His positional judgement would only require him to see a few moves ahead and still outplay me handily.

Just as Fritz 7 on my Macbook Pro will lose to Stockfish running from my old Android phone.

Anyway, I digress...

I, like a few of the fellas here, am specifically fond of this "weakness" which style basically is, and prefer a 2700-2800 engine like Prodeo/Frenzee/Ktulu/Gandalf/Junior 7/Rodent/Wasp and others, than the powerhouses approaching 3500 ELO.

I save those just for opening analysis or "truth seeking".

Besides, If I run Stockfish 8 against Ktulu (for example), it will stomp Ktulu into the ground with no effort, but the game will be a typical "stockfish win".

Everything is okay until Ktulu spots what SF's deep intention was and starts doing dumb stuff to avoid it. There is then a tactical explosion and the game is over.

It probably wouldnt be as beautiful as this win by Rodent II Henny (which was designed to be weak enough for ME to train against!).

[pgn][Event "Sicilian Masters Blitz Challenge"]
[Site "?"]
[Date "2017.01.08"]
[Round "7"]
[White "Rodent II 0.8.7 Henny"]
[Black "Ktulu 7.1"]
[Result "1-0"]
[ECO "B81"]
[WhiteElo "2160"]
[BlackElo "2364"]
[PlyCount "61"]
[EventDate "2017.01.07"]
[Source "Norman"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 e6 7. g4 h6 8. f4
Nc6 9. Be2 e5 10. Nxc6 bxc6 11. fxe5 Nxg4 12. Bxg4 Qh4+ 13. Bf2 Qxg4 14. exd6
Qg2 15. Rg1 Qxh2 16. Rg3 Bd7 17. Qd4 h5 18. O-O-O h4 19. Rxg7 Bxg7 20. Qxg7
Qf4+ 21. Kb1 Rf8 22. Be3 Qg4 23. Qf6 Qe6 24. Qxh4 f6 25. Bd4 O-O-O 26. Bb6 Rde8
27. Qf2 Qe5 28. Bc7 Bg4 29. d7+ Kxc7 30. dxe8=N+ Qxe8 31. Qg3+ 1-0[/pgn]

For me, this is what computer chess is all about.

Not so much edging closer and closer to 32 man tablebases, but creating chess games that make us fans smile.

If I have to sacrifice some strength to get this style, so be it.

I do not find games matching engines couple of hundreds elo apart interesting at all: it is like a soccer game between Barcelona and a local town club, the mant goals from the stronger side do not make the game more interesting.

I do not agree that it is more important for an engine to have an intriguing playing style than to play stronger. We have witnesses many engines with intriguing playing style, while we still have not witnessed the strongest engine. Of course, new things are much more interesting.

It is not true that engines progress faster than humans. Weak engines, like Komodo, SF, Houdini, etc., average some 50 elo progress per year, but that is only they are weak and have a lot to improve. Top GMs do not have that much room for improvement, as they are much closer to perfect positional play than top engines are.

It is also not true that top engine styles are going to merge more and more as the game advances. I can claim with a very large degree of confidence that Junior and Fritz from 15 years ago are closer in playing style than SF and Komodo of today, and it is very difficult to refute that.

Truth is, search space is at least 5 times larger than what most people actually suppose, and eval parametrising could be made much more detailed and accurate, so expecting quick convergence of playing styles sooner than 30-50 years from now is simply very much off-target.

So what's going on Lyudmil?

You're now stalking me because I humiliated you in two other threads?

Grow up man. You're like 20 years older than me and showing a poor example of how a man should conduct himself.

Rodent II Henny has been weakened so much that it is only about 2400 strength by CCRL measures, but it is also designed to create complete chaos on the board and this paid off against Ktulu 7.1 (which is 200-300 points HIGHER, not lower).

I don't know why I bother addressing the things you say.

"Weak engines, like Komodo, SF, Houdini, etc., average some 50 elo progress per year, but that is only they are weak and have a lot to improve. Top GMs do not have that much room for improvement, as they are much closer to perfect positional play than top engines are."

What a hilarious thing to say.

These "weak" engines you mentioned (which happen to be the strongest chess entities in the universe) do AMAZINGLY well do gain 50 points in a single year.

Top GMs having no room for improvement, despite being much weaker than said engines (which you say do have much room for improvement) is also a weird contradiction.

By the way, have you see any player from the top 5 of humans who has gained 50 points in one year? No room to improve, right?

Lyudmil, you've made you aggressive intentions clear, so I'm going to keep this in mind.

Not to censor myself, nor to flood this great forum with flame posts, but I might deal with our "differences" in more detail elsewhere.

you perceive it as stalking, I perceive it as a creative discussion, one of us will learn something, one way or another.

how can I be 20 years older than you, when I am 22, you should be 2 then small wonder?

my point was that new things are always more interesting, older things are a bit boring, and we have seen a lot of old engines' playing style, while we have seen almost nothing of how the close-to-perfect engines play. that is why dealing with the top is much more interesting than dealing with weaker engines, as a rule. what is so scary and difficult to understand for you here? watching a new film is usually worth it, while watching the same film for the 10th time is as a rule boring? what is so difficult to understand here?

If you want to level threats of unknown nature, please ask the mods to open for you a special threat subthread, where you can pour all of your venom, I will remain unfluttered though.

I can not understand why this can not be just a fruitful discussion, instead of trying to overreact each and every time.

[Lyudmil Tsvetkov]

When I started adding style parameters to Rodent, I noticed the following: for parameters that are separated correctly usually there is a range of good values rather than just one optimum. Separation is really the point. For example, modifying knight value by 5 centipawns had more consequences than changing all the material values (and I mean literally all: bishop pair, knight pair, imbalance bonuses and penalties etc.) by 5%.

What I really dream about is a set of parameter groups that are not mutually exclusive: for example having bishop pair both in material group and in a group containing pawn weaknesses (because the first bishop is often exchanged for opponent's concession in pawn structure). In a way, such tuning mechanism would resemble a small two-layered neural network.

Now for style: perhaps different playing styles have something to do with different parameter groupings? For example, the less interdependence between king safety and other factors, the more it can be tweaked without putting the entire evaluation off balance (a strange dependence story is shown here: https://sites.google.com/view/cdrill/personality/tuning - engine tuned on Morphy's games has king safety ludicrusly low because threats against enemy pieces seem to have the same effect on gameplay).

sometimes probably we have just to tune the rigth the things.

for example, both things are true:

1.) queen value>2*(rook value), and
2.) 2 rooks value>queen value

how can that be possible?

very simple, queen value is more than 2 times higher than a single rook value, but, when you have a pair of rooks, the interacction between them(imbalance) makes so that a single rook is valued more than half a queen.

give imbalance bonus for the rooks vs queen while not changing(increasing) material queen value, and fully not clear how the imbalance rule will work.

[reflectionofpower]

The style is often realized with System immanent methods,
E,g, chess genius had an asymmetric search tree with static exchange evaluator,

In ply 1-3-5-7-9 etc, it computed very selective while in ply 2-4-6-8... , I. The opponent plies, it computed more brute force.
As a result of this chess genius often saw threads against himself.
On the other hand it oversaw moves that could end the game .
This lead to a playing style that was very positional and on the other hand slightly boring.

nice

[leavenfish]

On the human side, as far as Nakamura is concerned, it may be that *he* needed to restrain his flashy impulses to maintain a near 2800 rating. But is the same true of Kasparov, Topalov and Shirov back in their prime? Shirov played his usual brand of 'fire on the board' chess, while being one of the top players on the rating list for many years.

Cheers,
CL

K, T and S...had that style against very human opponents and human opponents 'err' big and small, leading to....MEMORABLE AND INSTRUCTIVE GAMES!!

[leavenfish]

It's actually why those game still have 'life to them'. Off the top of my head I can't think of a single 'engine' vs 'engine' game...

the I suggest for next TCEC 10 final to match Komodo with Skipper_2017 as a wild card.

there will be some extreme attacks and tremendous positional crushes from Komodo.

what is your estimate of the people that will follow it?

Hey, I'm only making a point. But...if those did happen, they actually might be memorable games!

[BrendanJNorman]

how can I be 20 years older than you, when I am 22, you should be 2 then small wonder?

...dealing with the top is much more interesting than dealing with weaker engines, as a rule.

If you want to level threats of unknown nature, please ask the mods to open for you a special threat subthread, where you can pour all of your venom, I will remain unfluttered though.

I can not understand why this can not be just a fruitful discussion, instead of trying to overreact each and every time.

Again, gonna be quick so as not to derail an otherwise interesting thread.

1. How can you be 20 years older than me? The math isn't difficult, bud. "You should be 2 so small wonder" makes no sense. Try to rephrase the statement.

If you need need me to greatly elaborate on that particular point due to some insurmountable language barrier, that's fine.

I was saying "I am in my early thirties and you are (in my estimation) in your early fifties.

I should be learning from YOU how to be polite to others and tactful in communication. I should be learning from YOU how to be a gentleman and not insult others, but no. I am reprimanding YOU for your haughty attitude as others have over the years (from what I've seen on this forum alone)."

2. "Interesting" is a very subjective thing. Things that are interesting to you (insulting people online, deluding yourself into thinking you can beat Stockfish, making excuses/false analogies to prove a false premise, etc) are not interesting to me.

There are numerous chess fans here who don't mind weaker engines, as long as they have a unique playing style. I know a strong IM (2470 ELO) who uses just Prodeo 1.6 for analysis because the moves "make sense".

Who are we to argue? Who are you to impose on them what you think is "interesting"?

3. I don't make threats, I make statements. I'm not an internet bully like you. What I was referring to was a preference not to flood a wonderful computer chess forum with non-useful stuff just to address you. I have my own site, my own platform and thousands of readers who might find it more interesting. It's stupid to have this type of thing polluting the same thread where we can learn interesting things from a master like Pawel (who shared quite interesting views just before).

This is your venom Lyudmil.

You can't troll around the internet being rude and obnoxious to people and then play the victim when you're called on it.

Grow up.

[Lyudmil Tsvetkov]

how can I be 20 years older than you, when I am 22, you should be 2 then small wonder?

...dealing with the top is much more interesting than dealing with weaker engines, as a rule.

If you want to level threats of unknown nature, please ask the mods to open for you a special threat subthread, where you can pour all of your venom, I will remain unfluttered though.

I can not understand why this can not be just a fruitful discussion, instead of trying to overreact each and every time.

Again, gonna be quick so as not to derail an otherwise interesting thread.

1. How can you be 20 years older than me? The math isn't difficult, bud. "You should be 2 so small wonder" makes no sense. Try to rephrase the statement.

If you need need me to greatly elaborate on that particular point due to some insurmountable language barrier, that's fine.

I was saying "I am in my early thirties and you are (in my estimation) in your early fifties.

I should be learning from YOU how to be polite to others and tactful in communication. I should be learning from YOU how to be a gentleman and not insult others, but no. I am reprimanding YOU for your haughty attitude as others have over the years (from what I've seen on this forum alone)."

2. "Interesting" is a very subjective thing. Things that are interesting to you (insulting people online, deluding yourself into thinking you can beat Stockfish, making excuses/false analogies to prove a false premise, etc) are not interesting to me.

There are numerous chess fans here who don't mind weaker engines, as long as they have a unique playing style. I know a strong IM (2470 ELO) who uses just Prodeo 1.6 for analysis because the moves "make sense".

Who are we to argue? Who are you to impose on them what you think is "interesting"?

3. I don't make threats, I make statements. I'm not an internet bully like you. What I was referring to was a preference not to flood a wonderful computer chess forum with non-useful stuff just to address you. I have my own site, my own platform and thousands of readers who might find it more interesting. It's stupid to have this type of thing polluting the same thread where we can learn interesting things from a master like Pawel (who shared quite interesting views just before).

This is your venom Lyudmil.

You can't troll around the internet being rude and obnoxious to people and then play the victim when you're called on it.

Grow up.

just taking selected words from your above post. you call me:

1) a bully
2) an old man (and I am much younger than your estimation, of course)
3) haughty
4) insulting people online
5) deluding yourself into thinking you can beat Stockfish
6) making excuses/false analogies to prove a false premise
7) stupid
8) thread-polluter
9) rude and obnoxious to people
10) troll

that is your language, man, and, yes, you levelled a formal threat too, one post in advance.

in comparison, I have used much less unpleasant words, and no threats at all.

can you figure out who is the rude person and the bully, and who acts much more gentleman-like? especially, taken into account, that you are into a much more convenient mood than me currently. If you feel fine, you should be very condescending; on the other hand, people in difficulty do sometimes have the rigth to be more loose in some of their words/actions, simply because they are pressed.

one thing I can say for certain: I have never until now said I can do something special or praised any of my posts in any conceivable way spontaneously; I have done so, only after my posts have been labelled useless and my contribution annihilated.

so, I have every reason in the world to think I have behaved throughout gentleman-like and up to the moment; on the other hand, I am not so certain of your behaviour, because of the afore-mentioned couple of factors.

[Evert]

how can I be 20 years older than you, when I am 22, you should be 2 then small wonder?

Just how common is your name, then?
Either way, in that case I guess you want to update your CPW entry...?