Friday, September 26, 2014
Komodo 8 Engine
You can download the multi-processor version 5.1 for free on their site, but I have recently purchased Komodo 8 to use in my games at LSS. According to their website the Komodo engine has the approval of GM Boris Avrukh who says he is “going to test Komodo in my future work, especially in very positional play…” and GM Roman Dzindzichashvili who said its analysis produced “absolutely flawless positional chess…” I am skeptical of statements like this which are actually advertising and we all know about advertisements and testimonials, so before spending $60 I researched what other independent reviewers had to say before spending my money. People sometimes accuse me of being a tightwad, but that's not true; I am thrifty. Here's a synopsis of what I found.
What is important is the differences between Houdini, Komodo and Stockfish. One major difference is their search and evaluation functions. According to Larry Kaufman the differences are:
Komodo is best at evaluating middlegame positions accurately once the tactics are resolved.
Stockfish is best in the endgame and in seeing very deep tactics.
Houdini is the best at blitz and at seeing tactics quickly.
Rybka is obsolete.
Fritz is too far below these top engines to be useful.
Another important observation by Kaufman is that Houdini and Stockfish overvalue the queen. This something I have suspected for a while because I have reached positions where I have had the opportunity to reach positions involving a Q against two Rooks that have been evaluated equal or favoring the Q, but my feeling was that the evaluation was possibly not correct so I always avoided those positions. Of course, not being a GM, I could have been wrong but that was my feeling. Also, according to Kaufman, Komodo is the best choice for playing the opening when out of book early.
Stockfish is aggressive in its pruning of the analysis tree so its search is deep but narrow and this is supposed to be advantageous in some tactical positions and the endgame. Houdini is a very strong and fast in tactical positions. Its numerical value is also somewhat different than other engines in that they are supposed to be calculated to give predictable results. A one Pawn advantage gives a 80% chance of winning at blitz time control. A two Pawn advantage will win 95% of the time, and a three Pawn advantage 99% of the time. With a half a Pawn advantage the chances are about 50-50. This is important to know when it comes to correspondence chess because 1) you aren’t playing at blitz time controls and 2) tactical situations are rare; long positional games and endings are the norm.
Because Komodo is the most accurate in positional evaluations (according to Kaufman), it makes sense that it would be a valuable addition to the correspondence player's arsenal. Assuming, that is, that you are playing on a site that allows their use.
Komodo’s numerical evaluations are close to Houdini’s. Stockfish’s evaluations are usually quite high and as a rule of thumb you should use about 2/3 of SF’s score to get a reasonable comparison. I suspect that one reason for Komodo’s more accurate evaluation is that Kaufman is a GM and when it comes to engine evaluations vs. a GM’s evaluation, I always go with the GM. In analyzing games in books you will often see statements claiming one side has a positionally won game or some such and the engine will be showing it equal. If a world champion says one side is positionally won then it’s probably so. Komodo is a little slower than Stockfish and Houdini so it will need longer analysis time, but in correspondence play that is not an issue.
When it comes to answering the question as to which engine is best for use on LSS, well, it will depend. I will be testing Komodo out in a new tournament. I plan to use Komodo after leaving the book and in the middlegame and Stockfish 5 in the ending with Houdini 2 as the kibitzer.
Will it outperform Stockfish 5? Maybe, maybe not because these days I am prone to trying out various gambits in CC play rather than super-solid stuff like the Catalan, Nimzo-Indian and other highly theoretical lines the heavy weights of Advanced Chess play. I have been experimenting with non-mainline openings and defenses lately. Stuff like the Urusov Gambit, Evans Gambit, Sicilian Wing Gambit Deferred, QGD Chigorin Defense, Ruy Lopez Bird Defense and such. At my level even with engines this stuff seems playable.
Komodo is compatible with Chessbase 8 through 12, Fritz 8 through 14, Aquarium and Chess Assistant 10 through 14 (I am not sure about some of the free GUI's like SCID) and is available in download versions for Windows, Linux, Mac and Android. Remember...you are purchasing the engine only; you have to use your own GUI.