Saturday, July 2, 2011

More Advice on Analyzing with an Engine

Robin Smith examined this position in his book, Modern Chess Analysis and made some interesting observations.

      Back in Smith’s day he observed that many engines saw the position as roughly equal or as White having a slight advantage while the more popular ones, Shredder, Hiarcs and Crafty evaluated it as a slight advantage for Black and none saw the position as White having a winning advantage even though White has a strategically won position. 
      According to Smith the problem the programs have is that the weakness of the P’s on e6 and h7 are difficult to write into the program because they aren’t doubled, isolated, backward or on an open file.  As a result the engines did not regard them as particularly weak.  Additionally, the P’s are well defended and will remain so for a long time.  Only in the ending will the engine realize their weakness.
      I ran my own test using the Fire 1.5 xTreme x64 engine and after 10 minutes evaluation it determined White had a slight advantage (0.37) and gave the following line as best:

1.Rd3 Rfc8 2.Nd4 Na5 3.Nce2 Nc6 4.Nxc6 Bxc6 5.Nd4 Kf7 6.b3 a5 7.Rhh3 b5 8.Rc3 Bd7 9.Rxc5 Rxc5 10.Kb2 b4 11.Nf3 Kg8 12.Ng5 a4 13.c3 axb3 14.cxb4 Rc2+

      Believing this analysis without further investigation leads to a whole different conclusion.  One needs to jump ahead to position at the end of the line:


      The engine immediately shows an evaluation of 4.75 in White’s favor.  What happened?  As Smith points out, never believe analysis all the way to the end.  When the engine has generated what it thinks is the best move, it is absolutely essential that you go to the end of the variation and step backwards.  When the engine sees a large advantage for one side, back up a move and often you will find it will vary from the main analysis and then you must do the same thing with side variations.  Following this technique in the above position we see a drastic change at move 7 where the engine thinks if Black plays either 7…Be8 or 7…Bb5 White’s advantage is only a half P.  So, you would look for an improvement for Black at this point, checking both moves.
      Going back to the initial position, Smith gives the following analysis which initself is instructive: 1.Rd3 (keeping the N from going to e3 and then g4 from where it would tie White to the defense of his h-Pawn.) Rfc8 2.g3 (Inching towards an ending.  This move clears the 2nd rank so the weak square c2 can be guarded by placing the R on h2. This is an interesting note because it shows a part of the game most of us are oblivious to but plays a very important role in high level chess…weak squares. Of its top choices the Fire engine only places this move at about 5 or 6 with an almost completely equal position.  Fire recommends as better 2.Nd4 and 12 moves later shows an evaluation of 0.33.  Going to the end of the analysis reveals it thinks White’s advantage is not 1/3 of a P, but nearly 4 P’s!  So much for 2.Nd4 being better than 2.g3!)  2…a5 3.Rh2 b5 4.Ne2 (The two ideal squares for the N’s are d4 and g5, so White prepares to occupy both.) b4 5.Ned4 (Even at this stage, the engine fails to recognize that Black is totally lost, evaluating the position at 0.54 in White’s favor.) e8 (Smith calls this a mistake because the passive nature of the R here will be permanent.  He recommends 5…Rb8 and …Rb6 from where the R guards e6 from a more active square.  Fire wants to play 5…a4 but then meets 6.Ng5 with …Re8 and still believes White is only a half-P better.) 6.Ng5 Rcc8 7.c3 Re7 8.Rc2 Ra8 9.cxb4 axb4 10.b3 Na5 11.Rc7 and at this point the engine finally realizes White’s advantage is 3 P’s.
       What’s the point?  Besides (1) Smith’s interesting observation about why the initial position is strategically won for White even though engines, even today, think the advantage is only minimal, (2) it shows that evaluations by very strong engines can be misleading, (3) it shows why the backward-stepping analysis is essential, (4) it shows that correspondence play at high levels is beyond most of us simply because we would look at the engine evaluation and believe it.  Even if we didn’t believe it, how many of us would look at the initial position and evaluate it as strategically won for White? I guess that’s why we are such bad players…lack of understanding. (5) Finally, the real danger is you can easily get lead down a false path by following moves that the engine says lead to an OK position, but eventually you see its evaluation jump from your having a relatively meaningless half P deficit to it showing your position is so bad you can resign. 

No comments:

Post a Comment