Looking for great computer analysis

Looking for great computer analysis

Mar 30, 2018, 8:04 PM |

I would like to review my games with a super grandmaster. Since this is not in my budget, what is the next best thing? In particular, how much mileage can I get from computer analysis?

In order to compare computer analysis outputs, I picked as a sample game a 5|5 blitz played by my friend Armorix. It is a good game, but it contains enough mistakes for the computer to shine. It is also a trick to make sure that at least one person reads this post.

So here is the game with my comments, helped by Houdini 4 and chess.com computer analysis (stockfish.js 8):

Here are the raw comments of chess.com computer analysis (stockfish.js 8):

Chess.com provides good starting points. It is up to to the student to start the chess engine and explore the variations. Since there is no verbal comment, the computer analysis cannot point to higher level concepts, such as strategical imbalances. It is up to the student to draw conclusions from the experience. It seems extremely difficult to do without the assistance of a stronger player.

Chess.com provides a summary of the game:

Strength White Black
Excellent 29 24
Good 4 5
Inaccuracy (?!) 5 2
Mistake (?) 1 2
Blunder (??) 0 1
Forced 0 4
Best Move 51.4% 60.6%
CAPS  85.52 94.27
Avg. Diff 0.78 0.73

 If it was a boxing fight determined by decision, Armorix would have won regardless of the scoring system:

  • 61% best move vs. 51%.
  • CAPS 2290 ELO for Armorix vs. 1770 ELO for amir0504.
  • Average difference per move: 0.73 pawn vs. 0.78 pawn.

As a sidenote, the CAPS idea is great, but I feel that it is a prototype. In this example, the estimates are too high, and the ratings difference is overestimated. In my experience, the absolute numbers do not mean much. It is easy to get a CAPS ELO above 2400 in a slow symmetrical English, but it is hard to get above 2000 in a wild King's Gambit for example. But I feel that the difference between the CAPS estimates of the two players offers an interesting and successful measure of the quality of play in a given game.