16 January 2013

Auto-analytics for Chess

I recently had some time on a plane to catch up on my journal reading and ran across an article on auto-analytics from the Harvard Business Review.  Auto-analytics is defined as the practice of voluntarily collecting and analyzing data about oneself in order to improve.  As the article notes, "athletes have long used visual and advanced statistical analysis to ratchet up their performance" and then goes on to discuss its applications to the workplace.  Here, of course, I'm more interested in its usefulness for the chess player.

Chess is a natural fit for the discipline, especially in the modern era of database software, as each player can easily store their own games in a personal database for subsequent review and analysis.  Analyzing your own games, which I believe should be a central practice of the improving player, should probably be considered as part of the analytics process, since it reveals in-depth your strengths, weaknesses, and specific patterns of errors.  (One recent example of the last category was the repeated miscalculation of the ...Ne4 move that I uncovered, first seen in a Colle System structure from Annotated Game #78.)

Auto-analytics is more generally applicable to examining patterns of personal data to see what they reveal about your performance and behavior.  Simply arranging and presenting the data can often be useful, as they will almost always highlight areas of particular importance that you were not aware of.

Let's look at a simple set of categories, using my own (200+) games database as an example, accompanied by my best explanations and observations regarding the results.  A couple of minor surprises appear, along with some clear indications for where I should best concentrate my study and improvement efforts to have the most impact on my overall performance.  Any player using computer chess resources should similarly be able to generate their own set of results.

Cumulative Score 
Wins: 37.1%
Draws: 28.2%
Losses: 34.7%
Average rating of opponents: +22 Elo higher than me
It's nice to have a small overall plus for my career, but the most revealing statistic for me is the 28% draw rate.  This is in fact below my original impression - I might have guessed at least 35% - and I would consider it as reasonable, not worryingly high.  (A high draw rate at the Class level can be a problem for the improving player, often a sign of over-emphasizing safety over winning chances.)  I usually prefer to play in higher-rated sections or in open tournaments, which is reflected in the fact that my opponents have on average been rated slightly higher.
As White

Wins: 38.4%
Draws: 29.3%
Losses: 32.3%
A significant plus that is greater than the average expected plus for White.  Reflects well on my choice of the English Opening (although see below).
As Black

Wins: 35.9%
Draws: 27.2%
Losses: 36.9%
A small overall minus, indicating that I should pay greater attention to my openings and overall play as Black, if I want to have a more significant improvement of my winning percentage and therefore my overall performance.
Game length
Mode: 29 moves (White: 36 and 39, Black: 29)
Mean: 39 moves (White: 40, Black: 38)
Range: 11-76 moves
I am almost never "busted" in the opening phase of the game, so unless my opponent makes an early mistake, the game is likely to be around 40 moves.  (The mode shows that a shorter game will occur with some frequency, however.)  In any case, I should not expect quick results and should have the patience to settle in for a long middlegame and possibly endgame struggle.
Openings highlights (and lowlights) by ECO code
As White
A16: 50% (7 games) - notable for its frequency, if not its result.
A17: 62% (4 games) - closely allied to A16, with Nimzo-English or Queen's Indian type setups. Overall, a strong score.
A28: 38% (9 games) - the opening (English Four Knights) is normally played well, but I have often stumbled in the subsequent middlegame.

As Black
A00: 100% (2 games) - I do well when faced with irregular openings; they do not throw me simply because they are out of my personal book.
B18: 41% (12 games) - a disappointing result in the Caro-Kann Classical, although this is mostly due to middlegame problems rather than weak opening play.
C02: 0% (2 games) - my particularly bad losses in the Advance Caro-Kann are classified officially as an Advance French opening, which is what they transposed into (a tempo down).
D10: 58% (6 games) - the basic Slav Defense is a real winner for me.

1 comment:

  1. This post is inspiring !
    I never know which opening to play, and this kind of data could help me choose and stick to a repertoire.

    ReplyDelete