Saturday, April 5, 2014

Training in chess: A scientific approach

http://people.brunel.ac.uk/~hsstffg/preprints/Training_in_chess.PDF

Published in 2005 by one of DeGroot's PhD students.  I came across this while looking at an old Knights Errant improvement blog.  He summarized some of the findings and I just wanted to do a quick post going a little deeper in detail.

Explicit knowledge:
The example they give is the queen's gambit declined exchange variation.  A master knows that one of the plans in this opening is the minority attack on the queenside.  This is something that you learn explicitly.  The authors think it's best to have a small opening repertoire and study the methods in these positions thoroughly (this gets brought up again later).  Since this requires pattern recognition, you should use repetition to learn it.  They suggest going over the positions with different learning objectives in mind IE learn about the strategic aspects, then later go through looking for tactical ideas.

The opening:
As mentioned above, focus on a small repertoire of openings and expand later.  Rote learning is necessary.  Study your openings from different points of view that can be linked together.  Learn the typical middle-game and endgame positions that arise from that opening.  They recommend a "decomposition method" of taking a position from one of your openings then removing everything but kings and pawns.  Then gradually reintroduce more pieces to develop proficiency in the endgames that arise from your repertoire.

There's a series of books that Shereshevky wrote that are out of print called Mastering the Endgame Volume I and Volume II.  The idea was to cover the endings that arise from certain openings/pawn structures.  These books seem to fit that type of training and I don't think there are many like it.  Score another one for the "Russian Chess School". Mednis has From the Opening into the Endgame which is similar.  I haven't read any of those books, but the Shereshevky book came highly recommended from a native Russian speaker.

The middlegame:
Rote learning is less useful here because of the intersection of learning specific facts and general principles.  Specialize in the middle games that arise from your opening repertoire.

The endgame:
Start simple and acquire basic knowledge of all endgame piece configurations, paying special emphasis to pawn and rook endgames.  Study typical positions, and avoid arcane knowledge (Soltis mentions this in one of his books).  At later repetitions, add more complex endgames and also ensure your previous basic knowledge is still there.

They specifically recommended 3 books for this purpose.
Pachman - Les Finales/Chess Endings for the Practical Player
Averbakh - Lehrbuch der Schachendspiele (Band 1)
Averbakh - Lehrbuch der Schachendspiele (Band 2)

I assume Averbakh's Chess Endings: Essential Knowledge is a later compilation of those earlier German textbooks.

Study well annotated endgames by strong players for practical application of the theoretical endgame books.  They recommend 3 books for that purpose.

Shereshevsky - Endgame Strategy
Mednis - Practical Endgame Lessons
Euwe - Die Endspiellehre und ihre praktische Anwendung

I assume Euwe's A Guide to Chess Endings is the english version of the above.

Tactics:
A lot of what the chess improvement blogosphere already knows.  practice, repetition, chunking, etc.
They do cover some misconceptions about how to improve tactical skills.


  1. Calculate a branch of the search tree only once (Kotov).   
  2. Play blindfold chess.  They believe it's useless at best and possible harmful.  Blindfold chess skill is a side effect of being a good chess player, learning how to play blindfold doesn't make you better.
  3. Train with artistic chess problems.  Probably not good for competitive chess as the skills are different.  I assume these are things like Triple Loyds and other fun puzzles like switcheroos. Endgame studies are okay if they are practical in nature.

Strategy:
Don't go overboard with lots of strategy books as there's a big diminishing returns.  Study a couple and move on.  They recommend 2-3 classics plus a recent book on strategy.

Euwe - Judgment and Planning in Chess
Kmoch - Pawn Power in Chess
Nimzowitsch - My System

Strategic education is best gained by building your opening repertoire.  Example books using this method are:
Nunn - Benoni for the Tournament Player
Watson - Queen's Gambit: Chigorin Defense

There's a new series of "Move by Move" opening books published by Everyman Chess which might be good for that purpose, but I haven't read them.  Caro-Kann Move by Move, Ruy Lopez Move by Move, etc.

Other methods not recommended:
  1. Practicing visual representation.
  2. Practicing short-term memory
  3. Practicing deep calculation.
    1. They recommend that acquiring more knowledge is better than trying to think further ahead.  Pattern recognition narrows the list of candidate moves in a given position. They even suggest that knowledge is more efficient at improving your ability to look ahead.  This makes sense as someone with more knowledge will have better initial candidate moves and will focus less on bad variations which increases ability to look ahead more moves.

TLDR:
Start with a small opening repertoire and master the strategy, tactics and endgames that arise from them.
Study lots of tactics to build your pattern recognition.
Read annotated games meant for instruction.
Get a coach.

Sounds like the advice you'll hear from most strong players.




5 comments:

  1. Very interesting paper ! Thanks for sharing.

    ReplyDelete
  2. "I assume Averbakh's Chess Endings: Essential Knowledge is a later compilation of those earlier German textbooks."
    I have just checked that, its true. You may find the PGN for these books here : http://www.gambitchess.com/semi/db6.htm
    All books of Euwe are very good but after i have heared that these old endgame-books have many tablebase-checked errors ( exept Averbakh )
    I make my endgame training with Chessimo-Endgame and Convekta-software. You may test your endgame-skill at Chesstempo , they have endgame-puzzles too
    I think exercising ( plus reading) is much better than only reading

    I think the ability to play blindfold is useful: you will be able to calculate (long-) lines more easy and that will save mental energy

    And i studied a lot ! of tactics, even more than temposchlucker. There's a big diminishing of returns too ( after a few thousand attempts ).

    ReplyDelete
  3. Averbach's Essential Knowledge is a good 'one-volume' endgame work. But as you already own both Silman's and Howell's books, I think Averbach's isn't really necessary for you.

    ReplyDelete
  4. Thanks for the summary of the research paper.

    I think training in pattern recognition is very important but so also is calculation and evaluation of positions without moving the pieces. No, one doesn't often have to calculate 12+ ply deep very often, but it is very common to have to calculate 6+ ply deep and then be able to evaluate the position. This requires that you are properly visualizing the resulting position. This skill takes practice and/or a lot of time playing slow time control games.

    ReplyDelete
  5. Do you have the Judgement and planning in chess in pgn format? TIA.

    ReplyDelete