Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / chess engines playing styles
- - By noor (*) Date 2012-08-11 13:38
hi every one

what is the difference between the playing styles of the chess engines and how we can know if the engine is :

7-Human-like ( human style )

and as you think put the chess engines you knows with the appropriate playing style for the following as a start :

Stockfish :
Komodo :
Houdini :
Shredder :
Zappa :
and allll other chess engines

Parent - - By mindbreaker (****) [us] Date 2012-08-12 09:22 Edited 2012-08-12 09:25
It is hard to generalize because it often depends on the opposition, openings selected, and stage of the game.  Also what one player might think is aggressive another may not.  I think aggressive is advancing the pawns forcing pieces to move and the disruptive carnage to ensue. Just going right up the middle with piece support.  Someone else might think being aggressive is just going after the king and bringing everything and the kitchen sink to take him out.  Another may think it is playing speculatively, yet another might think it is just making a position ungodly sharp. And then there is boa style taking away squares strangling an opponent to death.  Or maybe someone thinks it is taking the initiative and holding control regardless of the investment in material required. Some might even think it is out and out greed, sucking up pawns, exchanges and defending those gains come what may. Some might even think aggressive is going ahead with plans while permitting the opponent to go on with his, barely defending at all. Just counterattacking. And there is any number of combinations of descriptions. And descriptions of aggressive include perhaps other things unrelated to the possibilities I have listed.  Maybe a higher probability of quick kills?  Sky's the limit. It could even be the exact opposite dragging the guy around 100, 200 moves whatever is necessary relentlessly until they finally get the win.  It could be as simple as avoiding drawish positions even if there is a bit of risk. Most engines will make peace though if they think their position is slightly worse unless there is some contempt.

If a strong engine is playing a weaker one it is hard for it not to look aggressive as it pounds opponents into the dirt.  Similarly, it is hard to look aggressive when playing a superior opponent and being forced to deal with superior moves.

And there are many other characteristics of engine style/personality than you list.  Some engines like to hold on to certain pieces, some like to move certain pieces more, some like open games, simi-open, or closed games more. Some like hypermodern positions more, others like classical, some will shoot to endgame if they get the slightest advantage others hang on to stuff for no good reason (often the safer course if you don't really know what you should be holding on to; which is common for engines).  Some are cowards with their kings, others boldly enter the game. And there are lots more I haven't mentioned.

And how do we define Human-like? Weak? Tactically inept?  Sure there are things humans do better than the machines some only make a difference in the length of a game rather than changing the outcome and then there are other things none of the computers can do.  For the most part "human" means unimaginative, blunder prone, arbitrarily rule bound, cowardly, easily worn out, and psychologically fragile.  Some of those things diminish at postal level but don't disappear. Some of the things computers don't do well become very well developed and powerful skills in the postal GM.  Computer: ingenious, beautiful, solid, objective (to a greater degree), flexible with changes in fortune, patient, energetic, unintimidated, opportunistic, never distracted by what could have been, loud music, or inevitable verbal abuse.

My take:

Stockfish: Constantly finds things that are incredibly profound only to discover that they are nonsense and has to settle for less than it could have had, had it had a little better intuition. Capable of being at least 50 Elo stronger if this were fixed.  Could use some style shifting for different phases of the game and more endgame knowledge.  Coding perhaps too simplified to go for the top spot.  Of course other engines may sit around with bugs in them for years without the author noticing because of the complexity.

Komodo: Please don't take offense. Boring! Unimaginative, zero profundity. Reasonably solid in all phases of the game.  Maybe the mulicore if it ever arrives can be more inspiring. Just one person's opinion.

Houdini: Ugly, poor anti-human, poor understanding of initiative/tempo, dang strong in endgame and leading up to endgame. Very tactically solid. Tip toes fine paths...finds the save, holds the win, seeing those narrow paths deeply.  Sometimes flops around a little when the path is very broad. Still, plenty of room to grow.

Shredder: Very refined, understanding of tempo especially in early opening, greater understanding of many types of positions especially endgame but other engines are so darn fast it can get swindled. Very effective anti-human. Good multiline.  Choices seem to get looked at well.  Doesn't have a heavy style bias.

Fritz: Has quirks.  It likes to do certain things the same way just like some humans like to achieve certain things.  If looked at exclusively you get the impression that you need to develop certain ways or die,  but the other engines show there are many strong ways of doing things. Sometimes heads for sharp positions it should avoid vs the big guys. Very polished.  More objective about types of openings. Can be predictable...well...until things get sharp then as a human I can't predict entirely accurately.

Zappa? Unfortunately even though I have seen thousands of games they have been fast and against stronger opposition, so I just see it smashed repeatedly.  It is hard to judge style when it has no control of the game.  Never updated.  There was supposed to be a new version with a new name but never happened.

Critter: My favorite of the elite.  Tactically strong, excellent at openings, excellent at seeing reasonable alternatives.  Somewhat biased toward making pawn moves in the opening, somewhat ignorant of hypermodern strategy, good at middlegame planning, reasonably good endgame. On single line it tends to get convinced one move is better than the others and stay fixated on it. In multiline it is wide open to seeing options and improvements.

Rybka: Awkward very early opening without book.  No comprehension of early pawn structures or tempo. Easily fixated in analysis.  My impression is that it is looking for there to be something wrong with its choice rather than looking for an alternative to be slightly more right.  Slow ply progression (hurts when another engine can get another ply or two after half the turn time is used and Rybka is at the same ply it was at the first half of the turn), great at middlegame planning, very decent endgame.  Great flexibility in altering personality.  Some personalities are vastly different than anything else but very good at the same time.  It is like you have a hundred 3200 choices.  Some early positions with many pieces on the board it can hit the search horizon hard and just not progress while other engines breeze deeply. It tends to get killed in those situations tactically. Dang thing looses on time on the last move of the time control often, like it is blind. Excellent anti-human especially with a little contempt thrown in. Even though it is good at endgame it gets run over by Houdini in endgame. Too many features just never worked right but those features were/are bold in their conception. Scrutinized more than any engine in history.

Junior: Has quirks like Fritz has, but quirks are not limited to opening.  Can wield a queen with amazing resourcefulness often fooling opponents into thinking it is a perpetual but is some deep fork or something.  Too bold for its own good against the heavyweights.  Too sacrificial to end up with good endgames.  Sometimes gambits for absolutely no good reason.  Understands hypermodern openings better than most engines. 
Don't have latest version but one I had would get to like depth 34 in a half an hour and just stop.  Not too useful for analysis. Too aggressive in pruning in long games.

Naum: Only engine to have a reasonable grasp of fortress positions. Other than that similar to Fruit or Toga.  Well, better...faster. A little nicer style.  Nothing new for a while.  Enjoyable to watch.

Spike: Aesthetically enjoyable games.  Does not gain much strength with extra time.  Probably some unwarranted generalizations in positional eval or something.

Loop: Can do some funky tactics in middlegame but can boomerang too.

Sjeng: Interesting, different, but being single threaded gets crushed.  Sure would be nice to have an up to date mulithread to see what it really can do.

Thinker: Hard to say much about it, its thinking lines are hidden. Single thread. Absolutely useless for analysis. Available versions get smashed.  A strong version does exist but has not been released, and maybe never will be.

Don't have Vitruvius, Chiron, or Onno.

There are engines that are nowhere near modern engines in tactics but nonetheless are very interesting against humans:

Chess Tiger (gambit mode): nothing develops and organizes an attack on a king with more single-minded determination often ingeniously.  Unfortunately it burns its bridges by doubling pawns to get rooks out and such never trying later to correct the positional weaknesses crippling its chances in endgame if the opponent survives the onslaught.

WChess: plays very much like a good human grandmaster.  Games are virtually indistinguishable.

Chessmaster: can play close to human, adjustable parameters, best graphics for actually playing.

The rest? Well, it is hard for me to see their style as I have just seen games where they just get crushed repeatedly. Wildcat is interesting and the Baron.  They both have potential.  And Comet and LambChop can be entertaining.

I wonder what other peoples' perceptions are.  I have not been looking at most of the engines lately and my memory can get fuzzy.
Parent - By Uly (Gold) [mx] Date 2012-08-12 11:19
Nice post! I always enjoy talking about engine's playing style, I think I agree with you in general.

>Zappa? Unfortunately even though I have seen thousands of games they have been fast and against stronger opposition, so I just see it smashed repeatedly.

Zappa has a very solid and "feets on earth" playing style, I actually could say it's the best "Insanity Check" engine for seeing if dubious ideas work or not. It also has unique evaluations, so you have have all the bunch of chess engines saying a position is won, while Zappa will stay calm and show the drawing lines, or others where all other engines think it's draw but actually, a piece is stuck and Zappa knows it and gives high eval and shows the winning lines.

Unfortunately, it takes time for this and Zappa generally plays poor moves at shallow depth, and is very poor unattended, which is why you will see it having low elo on rating lists, something that does not reflect to its great analysis strength.

I'd say Zappa Mexico II's style is the anti-Stockfish, in that Zappa is super-stable (even with MP uncertainty, you may give a position to Zappa and see how it ends with the same PV and score of last time!, while Stockfish will always be all over the place) and consistent in its style.
Parent - By tomgdrums (****) Date 2012-08-12 12:00
Nice post!

For what it is worth:

Komodo:  (K4 only)  I actually find this engine to be pretty interesting at times when doing post-mortem analysis of my OTB games.  It isn't super sacrificial but I am often surprised by what it will do.  (when doing some work on Spielmann's, "Art of Sacrifice in Chess" book I was surprised how many times K4 liked the sacrifice more than other engines did!

Houdini:  Often comes up with the best defense.  And that is why I like it.

Junior:  I have had an up and down relationship with Junior depending on the version.  With regards to the most recent version, I am finding it to be the best analysis partner with Houdini for my OTB post-mortems.    As much as an engine can, I think this version of Junior gets initiative and tempo.  I often prefer the lines it shows.  I always double check it in case Junior is drunk that day.  :)  But it really does often show the most troublesome path (for the opponent) with regards to OTB play. 

Vitruvius:  maybe I haven't used it enough but I don't think it is aggressive or sacrificial at all!  (one of the few engines where I felt like have actually wasted money)  I am open to being wrong about this but I think K4 will "go for interesting chess" more than Vitruvius does.  I actually think it plays fairly conservatively.

Chiron:  Still early but this is turning into one of my favorite alternative engines as well.  Will present a move that is just a bit different.

After a lot of experimentation by main engines for post-mortem analysis help are Houdini (tactics and defense), Junior (initiative and boldness) Komodo 4 and Chiron (in case either one finds something really interesting).   For basic evaluations I like Fritz 13 because it is so middle of the road.
Parent - By MarshallArts (***) [us] Date 2013-08-26 01:10
RE: Thinker

(I, too, wonder what happened to the unreleased/private version that was supposedly stronger.)

To do justice to Thinker you must pit it against weaker, or maybe nearly equal, opposition.
Then its strength will really shine through. Try its Active personality. Few other engines can
match its elegant play.

The latest available Thinker version is v5.4D, and it is quite strong, especially on multiple cores
(close to Zappa M-II in strength).
- - By Tiranasta (**) Date 2013-08-25 16:18
Anyone else have any thoughts on this? (yes, I know this is an old thread, but I figured it's better to bump this than to start a new one, since I'd only be asking the same thing and the insight already here is valuable)
Parent - - By Nelson Hernandez (Gold) [us] Date 2013-08-25 16:54
What strikes me in the question and the responses is that everything is so subjective.  A linguistics expert could explain this better than I could, but let me try to explain the problem: here the question seeks to somehow translate words such as "tactical" and "positional", which represent symbols of different ideas, into concrete evaluations of how programs compare with each other. 

How does one construct an objective test to measure this, or is the answer purely intuitive?

We can develop 1,000 tactical positions that have one "correct" move and have many engines run the gauntlet.  In so doing we can see which engines are more effective at finding the correct move.  But that doesn't answer which engines were trying to be tactical but failing versus which were not being tactical at all and were looking at the position more strategically.

We can develop 1,000 positions that have no clear-cut tactical shots but do have one move that, twenty moves later and far beyond the "tactical" range, turns out to be a critical move.  Lots of problems with this idea, though.  How do you really know a move was "positional" at all?  Isn't the whole design and spirit of computer chess "tactical" whether you're looking six moves or twenty moves ahead?
Parent - By Ozymandias (****) [es] Date 2013-08-25 17:30
Food for thought, you woke up today feeling philosophical.
Parent - - By Carl Bicknell (*****) [gb] Date 2013-08-25 17:30

> What strikes me in the question and the responses is that everything is so subjective.


Style is nonsense IMO. It's a bias where one shouldn't exist. The best chess is to just play the position. If an attack is needed - attack. If simplification is needed - play the endgame. But to try and be 'aggressive' in all positions is a weakness and engine authors should avoid this.
Parent - By Master Om (Bronze) [in] Date 2013-08-25 18:22

>Style is nonsense IMO.

Thats why all say the Romantic days are gone. Style is important. Tal also mated opponents so do Petrosian. But how its done is the beauty of chess.
Hence Tal is more popular than Petrosian. I was motivated to play chess by looking at the game of Paul Morphy vs Brunsiwick And Count Isoward. Had he not played in neo romantic style I would not have motivated to play. Style is very important in chess.

>If an attack is needed - attack. If simplification is needed - play the endgame.

That is decided by the temperament of the player even both have win in sight. You cant expect Capablanca to play like Alekhine.
I can give you a fine example also.

[Event "New York"]
[Site "New York"]
[Date "1857.??.??"]
[Round "?"]
[White "Paulsen, L."]
[Black "Morphy, P."]
[Result "0-1"]
[ECO "C48"]
[PlyCount "56"]
[EventDate "1857.??.??"]

{Paul Morphy competed in only one tournament in his short career, the 1st
American Chess Congress in 1857. In the final round of this knock-out event,
he defeated German master Louis Paulsen by a score of +5, =2, -1. In this game,
he demonstrates both his better grasp of positional play - Black's control of
the center files makes a marked contrast to White's flailing on the flanks -
and his combinative ability, as he finishes the game with a startling and
brilliant Queen sacrifice.} 1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. Bb5 Bc5 5. O-O
O-O 6. Nxe5 Re8 {Rather than permit the "fork trick" 6...Nxe5 7.d4, Black
sacrifices a Pawn for rapid development.} 7. Nxc6 dxc6 8. Bc4 b5 9. Be2 {The
seemingly more logical 9.Bb3 fails to 9...Bg4 10.Qe1 [or 10.Ne2 Rxe4 winning
the pinned Knight] 10...b4, and if 11.Na4 Rxe4 traps the White Queen.} Nxe4 10.
Nxe4 Rxe4 11. Bf3 Re6 12. c3 {If White were able to follow up with d2-d4, this
would be a good move, but he can't. He should reconcile himself to 12.d3.} Qd3
13. b4 Bb6 14. a4 bxa4 15. Qxa4 Bd7 16. Ra2 Rae8 {Threatens mate with 17...
Qxf1+. White's reply defends against this sacrifice, but allows another, which
Paulsen can hardly be blamed for missing. Relatively best was 17.Qd1} 17. Qa6
Qxf3 {Morphy took 12 minutes to decide on 17...Qxf3, an unusually long time
for him. Paulsen, a notoriously slow player, thought for over an hour before
capturing the Queen.} 18. gxf3 Rg6+ 19. Kh1 Bh3 20. Rd1 {Black threatened 20...
Bg2+ 21.Kg1 Bxf3++, and 20.Rg1 fails to 20...Rxg1+ 21.Kxg1 Re1+. The key line,
which Paulsen probably missed at move 17, is 20.Qd3 [hoping to return the
Queen with 20...Qxg6] 20...f5!, and White is helpless.} Bg2+ 21. Kg1 Bxf3+ 22.
Kf1 Bg2+ $2 (22... Rg2 $3 {This move would have sealed the game immideately
which was later found by Dr E Lasker. But even geniouses miss too!!.} 23. Qd3
Rxf2+ 24. Kg1 Rg2+ 25. Kh1 Rg1#) 23. Kg1 Bh3+ $2 (23... Be4+ $1 24. Kf1 Bf5 25.
Qe2 Bh3+ 26. Ke1 Rg1#) 24. Kh1 Bxf2 $6 25. Qf1 Bxf1 26. Rxf1 Re2 27. Ra1 Rh6
28. d4 Be3 {White Resigns.} 0-1

Look At 17th move by P Morphy. 17...Qxf3!!. He could also have won after QxRf8!!, a quite move which leads to a Calm win in endgame. But opted for a neo romantic move Qxf3!! which too wins.
Parent - - By leavenfish (***) [us] Date 2013-08-25 19:32
"Tactics" are the easy part of the game for an engine. In general, either a good tactic is there or it is not...given how deep they look these days, it's really not so important for pure evaluation of a position...playing at different time limits of course throws a spanner into things, that's why I mention the depth of search.

"1,000 positions that have no clear-cut tactical shots..."

Now here is the only real interest I have in computer chess - evaluation of a given position at a reasonable depth. That is where you might argue that 'style' comes into play: how this and that is weighted when it comes into deciding on a move. One could even argue that 1.d4 games are by nature more positional in nature, 1. c4 even more so...than 1.e4. The programmer might therefore make his engine lean on one or the other more heavily depending on it's 'strenghth'...and even play 'quieter' lines when playing 1.e4 and meeting...say an open Sicilian, for example.  For such reasons I take all these computer matches with a grain of salt. Why for example would you force Komodo to play 1. e4 or play aggressive lines against a Dragon...if where it excel's is in other areas?

I would argue those who write specific opening books tailor those to the 'style' of the engine based on thousands of games and as little as a little as 1% difference in how said engine fairs in more open or more closed positions. In that sense, yes, a well designed gauntlet of starting positions might give an indication of which engine is better in a varied environment...but really, that is still like trying to make Tal play like Petrosian and Petrosian play like Tal. In an actual match, Tal and Petrosian would both play to accentuate their own strengths and dampen the others. That is just hard to do with chess engines because even when they use their own tailored opening book...humans are making those and one human may do a better job than another at putting his engine on the best footing for battle.
Parent - By Nelson Hernandez (Gold) [us] Date 2013-08-25 20:39
The thing is, every chess program has it's own unique "style".  What annoys us is that their styles are so dissimilar from human styles that we can scarcely understand their choices in some positions.  Moreover we tend to focus on games where a program lost and can easily see where there was a misjudgment.  What we fail to see are the many times the engine played brilliantly, because that is now the expectation.

If you look at the game as if you had a 32-man EGTB you'd see with great clarity that chess is as deterministic as tic-tac-toe.  (No need to point out the vast differences.  All I said was that it is deterministic.  There is a certain outcome to every position with optimal play.)  Assuming you had such a tool what would you like to see?  More moves that turned winning positions into losing ones, but looked pretty?
Up Topic The Rybka Lounge / Computer Chess / chess engines playing styles

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill