More On Chess Engine Evaluations

      Tuesday's post, how engines evaluate a position and make blunders, showed you can't ALWAYS religiously trust an engine's evaluation. As demonstrated, even at depth, an engine's first choice may not be the best move, so this is a reminder on some things that have to be taken into consideration when looking at the evaluation score.

      First, a lot depends on what you are trying to accomplish. If you are looking for best lines against human opponent's then your goal may be to pose as many difficulties as possible and so you may want to choose a move that an engine would not consider best.

Equal (=): 0 to 0.26

Small advantage for White (+/=): 0.27 - 0.70

Large advantage for White (+/-): 0.71 – 1.49

Decisive advantage for White (+-): over 1.5

      One major problem with an engine's evaluation is that even in some drawn positions they might be showing an advantage for one side. If there are 6 pieces or less, then you want to refer to endgame databases.

      When it comes to tactical evaluations engines don't miss much, but when long term planning and positional evaluations are the major consideration you may find engines will be evaluating one side better for a long time then all of a sudden, later in the game, the evaluation makes a drastic change! Believe me, I have seen this happen many times when playing on Lechenicher SchachServer.

      This factor is especially important when evaluating openings and positions where there are several moves with nearly equal evaluations. It's very important to understand the ideas and plans for both sides which is, for most if us, beyond our abilities. In these cases I have found the best thing to do is use my Fritz 12 GUI and run some Shootouts. Shootouts allow you to use one or more engines to play out the game at specified ply depths.

Consider the following position:

Black to move

      This position is from the 1909 Lasker vs. Janowsky World Championship Match. I let Stockfish 6 evaluate the position for 15 minutes, reaching a depth of 34 ply, and its evaluation score for 12...Ne7 was a tiny bit in black's favor…-0.11. Next I used Komodo 8 and in 15 minutes it reached the same conclusion except it had a very tiny preference for 12...Re8 and its evaluation was -0.08. Neither engine was correct. 

Notice the two main features:

     Both sides have doubled Pawns. White's are on the same wing as black's P-majority and they are even doubled. Doubled Pawns are a weakness when mobility is a factor, but here they are very effective because they are capable of stopping the creation of a passed Pawn. 

     On the K-side things are different. White's P-majority is capable of eventually creating a passed Pawn which would give him an advantage in the ending. Another factor is the value of the minor pieces. White's B is active and in this rather open position, it is superior to the black N.

     As a result, white has a clear positional advantage that in the long run should give him excellent winning chances. This was confirmed in a series of Shootouts using both engines at a depths of 7 to 25 plies. White scored +8 -3 =17 (59 percent), showing the validity of the conclusion.  This information can be a valuable piece of knowledge to possess!

      In the actual game Lasker did go on to win with only a couple of minor mistakes by Janowsky.