Thursday, November 20, 2014

Komodo 8 in Action



     Now that I’ve had K8 in use on Lechenicher SchachServer for a while, I can make some preliminary observations. Supposedly K8 leads to sound positional play and the discovery of ideas that are much deeper and more subtle than other leading engines. Also, the claim is that it sees deeper than any other engine. Blah, blah, blah. I don’t trust the ‘claims’ and don’t understand the technical jargon. What I want to know is, how well does it do in (non-blitz) competition?
     Komodo is known for superb positional play and I appreciate the fact that Kaufman and Dailey refused to sacrifice positional play just to score better on tactical problem and in blitz tournaments against other engines.  Positional ability makes Komodo useful for opening analysis and Kaufman has paid close attention to make sure evaluations agree with theory.  And at slow time controls in correspondence chess (where engine use is allowed) opening analysis and accurate positional evaluations are extremely important.
     As I have mentioned in previous posts, I never trust engine evaluations in materially unbalanced positions. This is not because their evaluations went against my gut instincts because I’m not good enough to make that distinction; it’s because I got burned a few times trusting the engines in those situations!
     According to the experts, Komodo is best at opening play and evaluating middlegame positions where tactics are not a factor. Stockfish is best in the endgame and in seeing very deep tactics. Houdini is the best at blitz and at seeing tactics, so they all have their areas of expertise so to speak. For all practical purposes, there’s not much difference between them, so my feeling, as expressed in an earlier post, is go for free! But, how has Komodo been performing on LSS? Here’s a position from one of my recent games (I’m black and it’s my move):


     Who’s better here? White looks to have good chances for an attack while black has a slight material advantage. My feeling is that overall white has the better chances. So, if I thought that, why did I play into this line? I wanted to test it and was willing to risk a couple of rating points in the process; they aren’t worth anything anyway.

Houdini 2: The position is equal (0.22) after 11...Kg7 or 11...d5.

Komodo initially also recommended 11...Kg7 but thought white was much better, 0.83.
In the end, it liked 11...d6 and evaluated the position as almost a Pawn in white’s favor. I didn’t understand its second choice, also evaluated at 0.83, of 11...Nb8.

Stockfish evaluations: 11...Bxf2+ and the position was one Pawn in white’s favor. Its second choice of 11...a5 (1.13) doesn’t make sense to me.

     So, the choices are: 11...Kg7, 11...d5, 11...d6 and 11...Bxf2+. Is the position equal or in white’s favor?
     It looks to me like white is better, so which move makes it the easiest for black to try and equalize? I eliminated 11...d5 because to me it makes no sense to give up a P after 12.exd5 and the N has to go back to b8. 
     As for 11...Bxf2+ the B is immune because if 12.Kxf2 then 12...Nxe4 and black is pretty close to equalizing. But, after 11...Bxf2+ 12.Kf1 white retains his advantage and will play Qf3 next move putting a lot of pressure on black’s position.
     It made sense to play 11...d6 so as to get the B out then after 12.Qf6 Kg7 defending the N is forced. I chose 11...Kg7 first followed by ...d6 and evaluations still favor white by about a Pawn. None of the engines were giving any clear indications of how either side should proceed. In a few moves we reached this position:

Black to move
     I ran a Monte Carlo analysis using the only engine that works with this method, Deep Rybka 4 w32, and after 400 games was surprised at the result: white won 83 lost 152 and drew 165. I didn’t trust those results though because the engine is too weak and the search depth was way too shallow. I don’t even know if this method is being marketed anymore and it may have been discredited. 
     Houdini 2, Stockfish 5 and Komodo 8 were showing the position in white’s favor by only a teeny bit but were not offering anything that looked concrete. Plus, rightly or wrongly, Monte Carlo results not counting, my feeling was the ending was probably going to favor white. I didn’t see any way to threaten his K with my two pieces and was afraid that he would open up files for his R’s on the Q-side leaving me with too much to defend against...I accepted his draw offer.
     In the end, I haven’t found any difference in Komodo 8 and Stockfish 5 in actual play at correspondence time controls.
     Download Stockfish from its own site HERE.  You can also visit the Stockfish Programming Wiki for more information.

 
I was getting pretty bored with engine-assisted CC. An endless round of Najdorf Sicilians, Nimzo-Indians and Slavs…boring! Then I started experimenting with some discredited gambits: Urusov, Rousseau, Bryntse, Sicilian Wing Gambit, Boden-Kieseritsky gambit, etc. and found they can make for some interesting chess.

1 comment:

  1. I use to analyze only the Komodo 8 .on the game most of the "human style"

    ReplyDelete