Friday, March 4, 2011

Correspondence Play and Analyzing with Engines

Computers can be a useful tool in opening research and checking out opponents.


      If you are going to play CC these days you will probably use an engine. Not necessarily to select your moves because on most sites their use is illegal. However you may be keeping track of your games using a chess program or researching openings.

      When it comes to openings, Stan Vaughn, a well-known correspondence player, advises that you should never take the evaluations given in books for granted. I can attest to this because over the years, before anybody ever heard of computers, I lost more than one game by blindly following the “book.” Vaughn advises that you should go to the end of the variation, set up a position and let the engine check it out. Personally, I would check out the whole line not just the final position. There may be a bad move that occurs earlier.
      Also be aware that engines may give a deceptive positional evaluation. For example, assuming the game does not take a tactical turn, as you approach the ending, that Q-side P-majority the engine did not give a lot of importance to may be the deciding factor. I have also reached positions where I “knew” a sacrifice would give me a winning advantage but when analyzed with an engine, it was skeptical and, as I’ve said previously, sometimes engines don’t evaluate positions with unbalanced material very well. Play over GM games with an engine and you’ll find that at times their opinion disagrees with the engine. Unless it’s a tactical situation, trust the GM.
      Remember, game databases are museums because they only tell you what was popular in the past and so you have to keep up on the latest theory.
      It is also advisable to look up your opponent’s games on whatever server you are playing on. It may give you a good idea of what openings he likes to play. You may also want to play through some of his games to see if you can spot any strengths or weaknesses.
      If he likes to play the King’s Gambit and you prefer to avoid it, then you may want to meet his 1.e4 with something other than 1…e5. If he’s a lower rated player you may notice he makes a lot of tactical mistakes, or as I have found in a lot of online blitz games, some players have a penchant for sacrifices on f7 (or f2), sound or not. In such cases you can often bait them into situations where you can be reasonably sure they won’t play well. Or, maybe you see your opponent plays endings poorly. If that’s the case, why not avoid tactical lines and play for the ending? Maybe he plays the same openings in every game. You may want to research them and search out the latest improvements. Maybe he won’t be aware of them.

Another question is how to use an engine for effective analysis of a position.

      Using engines to analyze sharp variations has made a big impact on openings these days. Every GM and other strong players use them to one degree or another. Given enough time, really strong players will see the same tactical shots the computer sees but good analysis requires looking at a lot of variations and sub-variations, so this is where an engine will save an enormous amount of time. The final result will be analysis that’s largely free of serious tactical errors. That’s something that cannot be said for much old pre-computer analysis where writers often made their analysis quickly or carelessly.
      The truth is, reliance on engine evaluations decreases as one’s tactical ability increases. Thus GM’s and world class CC players won’t get much benefit from knowing Fritz or Rybka’s evaluation of the position is 0.95; they form their own evaluation.
      It is no use having the engine do the analysis if you don’t understand it. Unless you are running a powerful computer and letting the engine run for a couple of days, when allowing it to evaluate a position, the deeper the evaluation goes, the less likely it is to be totally accurate. Good analysis must be first, broad. Once you have looked at a lot of plausible moves, then it’s time to go deeper into the ones that look most promising. This way you will also notice themes and motifs that recur in one form or another in the different variations. Try to avoid playing the next move and the next and the next until you are far too deep into the variation.
      Very often engines will offer as best moves of a sort that only a computer would select and not moves that a human would play. That’s because humans play moves based on principles humans understand rather than algorithms.
      Another temptation is to trust an engine’s evaluation completely but those evaluations are often worthless. The scoring function of the engine should be regarded as mostly an approximation based on the material balance, nothing more. Chess engines lack imagination. They will see tactics that can be calculated to the end within their horizon, but they will miss ideas that cannot be calculated within their horizon. Long-range positional sacrifices are beyond their ability to calculate.
      This is the realm where CC GM’s excel and it is the reason they routinely defeat players who rely solely on engine outputs. The CC GM can nudge the play in the direction where the engine evaluation is wrong. One interesting observation by a strong CC player I read was, “If you are down material and the machine says the score is about even, you have the advantage.”
      Engines are materialists, so in the final analysis, when it comes to positional judgment, you have to rely on your own, not the engine’s!

No comments:

Post a Comment