Best Style/Strength Ratio?

[BrendanJNorman]

Yeah I agree I put OpenTal 1.0 extermly aggressive and if u want more strength then use latest stockfish dev contempt 100.

If you dont care if the engine gets too strong (i.e you arent using it for training games), there is Fizbo 2, Vitruvius, Spark 1.0, Critter, Andscacs (earlier versions) and yep...Stockfish dev with contempt (and can also increase kingsafety, aggressiveness in older versions of SF).

All really strong and aggressive, but not so useful for training games.

[jmartus]

And put the engines vs another engines that 400 elo weaker hard to tell style when facing harder opponent

[BrendanJNorman]

And put the engines vs another engines that 400 elo weaker hard to tell style when facing harder opponent

Right. I call them "cannon fodder" engines.

Guys who play a reasonable game, human-like strategy, but only about 2300-2400.

Engines in my "cannon fodder" list include:

- WChess 1.06
- Kiwi 0.6
- BigLion
- Adroitchess

and some others.

Together with a "stylish" engine, they produce great games.

Not just the typical "tick, tack, toe, oh shit I've been outsearched - throw pieces in the way before I get mated!"

And those engines are also nice opponents in their own right.

[Ovyron]

Come on man, you know OpenTal is designed for play against humans!

Playing against Stockfish is like Mike Tyson against Peewee Herman.

Yeah, but I don't care about game results here, I remember that in the ProDeo 1.2 days some personalities would throw away everything in some awesome attack, and go on to lose, but who cares if it's the attack that mattered. Zappa DA was the only one I saw playing such attacks and winning or at least being able to hold the draw after it didn't work out, I even had discussion with Richard Vida where he claimed Zappa DA was playing losing moves and it wasn't being punished because his Critter was announcing some +2.00 scores, but Vempele and me kept showing moves he kept missing that... saved the game (and I just always wished for those positions to be won, but alas, the only time I've seen such flips have been in the Alpha Zero games.)

Is it really that OpenTal's style only holds against humans? I just can't believe it, an engine's style shouldn't depend on opponent, lest ProDeo's high Confusion setting might be the way on here, but it'd just make the engine weaker otherwise.

To answer your questions though, since I'm not really a fan of engines above around 2900 (I like to play against them), here is my list.

Thanks for the list, I always hear about some new stylish engine I didn't know existed. Though, again, I'd want to test them against some other stylish engine

Anyway, have you tried limited depth matches? They're very powerful, and there will almost always be some depth that plays below your level and another that plays above your level, then you can enjoy the style of the top engines without worrying about losing to them.

The only problem is the instant playing thing, I have had great performance against players of higher rating at reverse handicap games. Yes, that is, I have 10 minutes on the clock, they have 20 minutes and an increment of 15 seconds, and I play better than when they have 10 minutes, because I ponder greatly, and the longer they think, the longer I think, and my game improves. The most difficult times are against those that play quick and don't give me ponder time, and engines that play instantly work like this, we need a GUI that delays the engine's moves (LucasChess?)

[carldaman]

Come on man, you know OpenTal is designed for play against humans!

Playing against Stockfish is like Mike Tyson against Peewee Herman.

Yeah, but I don't care about game results here, I remember that in the ProDeo 1.2 days some personalities would throw away everything in some awesome attack, and go on to lose, but who cares if it's the attack that mattered. Zappa DA was the only one I saw playing such attacks and winning or at least being able to hold the draw after it didn't work out, I even had discussion with Richard Vida where he claimed Zappa DA was playing losing moves and it wasn't being punished because his Critter was announcing some +2.00 scores, but Vempele and me kept showing moves he kept missing that... saved the game (and I just always wished for those positions to be won, but alas, the only time I've seen such flips have been in the Alpha Zero games.)

Is it really that OpenTal's style only holds against humans? I just can't believe it, an engine's style shouldn't depend on opponent, lest ProDeo's high Confusion setting might be the way on here, but it'd just make the engine weaker otherwise.

To answer your questions though, since I'm not really a fan of engines above around 2900 (I like to play against them), here is my list.

Thanks for the list, I always hear about some new stylish engine I didn't know existed. Though, again, I'd want to test them against some other stylish engine

Anyway, have you tried limited depth matches? They're very powerful, and there will almost always be some depth that plays below your level and another that plays above your level, then you can enjoy the style of the top engines without worrying about losing to them.

The only problem is the instant playing thing, I have had great performance against players of higher rating at reverse handicap games. Yes, that is, I have 10 minutes on the clock, they have 20 minutes and an increment of 15 seconds, and I play better than when they have 10 minutes, because I ponder greatly, and the longer they think, the longer I think, and my game improves. The most difficult times are against those that play quick and don't give me ponder time, and engines that play instantly work like this, we need a GUI that delays the engine's moves (LucasChess?)

In my tests, OpenTal holds up well against engines up to around 2600 CCRL, meaning it usually wins more than it loses.

Btw, I've suggested a move delay feature for GUIs in the past, too. It is very 'off-putting' to face an opponent that keeps moving instantly.

[Ras]

It's also possible to add in virtual CPU throttling to engines. Then they take e.g. 3 minutes, but only really calculate for, say, 3 seconds. The upside is that this works with all GUIs; the downside is that this has to be implemented per engine.

My engine can do this via percent CPU or, in the upcoming release, directly in kNPS. Both using UCI options accessible via GUI.

[yanquis1972]

at some point surely someone will just program a "Tal" switch into a top engine, i'd think we're already close. in the OpenTal - SF game posted i had the sides wrong in my mind & playing through the game i really liked blacks style in the opening.

something like high contempt (favor highly complex positions), substantially overweight opponent king safety, favor speculative/spectacular moves. maybe tactical search too depending what that limits.

[Ovyron]

something like high contempt (favor highly complex positions), substantially overweight opponent king safety, favor speculative/spectacular moves. maybe tactical search too depending what that limits.

I think Stockfish has all that hidden in the code, and someone could take a look and make UCI options so those things are adjustable. Talfish might be within reach...

Note stockfish_14051722 allowed modifying Mobility, Pawn Structure, Passed Pawns, Space, Aggressiveness, and Cowardice, so perhaps the best try for a TalFish would be to open up 14051722's code, and implement Stockfish 9's Contempt, then you use Contempt=100+Aggressiveness=200+Cowardice=0 for some assassining personality o_o

[BrendanJNorman]

Is it really that OpenTal's style only holds against humans? I just can't believe it, an engine's style shouldn't depend on opponent, lest ProDeo's high Confusion setting might be the way on here, but it'd just make the engine weaker otherwise.

I think you've made a false deduction here buddy.

I didn't say that OpenTal's style only holds for humans.

I have written about OpenTal smashing 2600 CCRL level engines (probably 2700 Elo) in a spectacular way.

http://chessncognac.com/opental-chess-engine/

But just as the REAL Tal wouldn't be able to play his dazzling style against Stockfish, neither can OpenTal.

The real Mikhail Tal, even at his peak, would have been quickly smothered and crushed by Stockfish - and would have been prevented from showing any of his strengths.

So rather than the mistaken "only holds against humans", I was asking "WHO can look spectacular against Stockfish?".

Naturally Stockfish (3450) is going to look brilliant against OT (2600), just as Mike Tyson would look brilliant against somebody half his weight.

This is a case of simply using OpenTal as the "cannon fodder" I mentioned elsewhere.

It's rather strange to say that Elo doesn't matter in one thread, and then jump on the Stockfish train in another.

Yep, Stockfish is the strongest engine, that's why I keep it to analyze the games of mine and interesting engines.

Note: I still use Fizbo 1.1, Alfil 11, Frenzee 3.5.19, Wasp 1.01, Zarkov 6.55, ProDeo 1.2 (except the 2.6 personalities I created) and other "old engines".

They're enough for me - with OpenTal.

BTW...the real Tal personality for Rodent is about 2745* (as apposed to OT's 2600 approx), so if you want the same style, but a bit stronger, this one is more "conservative".

OT was designed to also play the unsound sacrifices that Tal played in his career, and in move-matching tests it plays 99% of them.

That's why it comes across as "weak" against strong engines.

*Based on 21/44 vs Av CCRL Rating of 2759

[Ovyron]

It's rather strange to say that Elo doesn't matter in one thread, and then jump on the Stockfish train in another.

Ah, but the big thing happening here is that the latest implementation of Stockfish Contempt is like some built-in style modifier.

Think of ProDeo's "Attractiveness" setting, except some values make Stockfish play both more stylish AND stronger!

So the question is not what engine is strong enough to show its style against Stockfish...

The question is... WHAT IF Stockfish is already playing with a better style than OpenTal, but hundreds of elo stronger?

How would we know? Is it that Stockfish is so strong that its strength obscures any possible style it has? What if there's some optimal value of Contempt that increases the Style/Strength Ratio to the max and it's between 40 and 100, say Contempt=66 plays with better style than OpenTal, but 67 or 65 doesn't?

I wouldn't have made this thread if the event where the engine with the best style was also the strongest one didn't seem so close, but I want to be ready because we're so focused on sub2900 elo engines that we might miss it.

I wonder if you'd have included Stockfish with Contempt=40 in your list if it had some 2600 elo, and it seems like some discrimination against strong engines...

Yeah, strength doesn't matter, but it goes both ways! I jumped on the Stockfish train because of the awesome playing style it is showing, not its strenhth! john dalhem even thought the sides were swapped and OpenTal was making Stockfish moves!

So here's an idea:

1. Find whatever depth makes Stockfish play at OpenTal's level. We can't just let them play until OT wins a game, because it'd take forever, we weaken SF.

2. Play a match between them, this match will have some 50% performance so both sides have the same number of wins, loses and draws. Play some extra games and throw away the ones you don't need to achieve this.

3. Shuffle the games AND REMOVE THE NAMES OF THE PLAYERS, like in a turing test.

4. Have a judge look at the games and, on a game per game basis, decide what engine played with the better style. This is better than having to go to specific positions to give awards to hung pieces or queen gambits (where the queen is given away) due to speed, you just do 1-0 or 0-1 results depending on what engine showed the better style, regardless of game result.

If at the end of such a thing, Stockfish has more "style" points, my point would have been proven. If OpenTal's style becomes really apparent and it leaves Stockfish with 0 point, it has proven the soundness of the method, win-win!

So what do you say Brendan? Are you willing to be the judge and take a look at some games played and award points to the side with best style to see if this thing works at all?

The hardest part will be to produce the games, but I already have an idea of how to achieve same level quickly...