Sunday, August 12, 2012

Deliberate practice

Mr. Z. provided the following text:

".... At one point, not long after I started training, my memory stopped improving. No
matter how much I practiced, I couldn’t memorize playing cards any faster than 1 every
10 seconds. I was stuck in a rut, and I couldn’t figure out why. “My card times have
hit a plateau,” I lamented.

“I would recommend you check out the literature on speed typing,” he replied.

When people first learn to use a keyboard, they improve very quickly from sloppy
single-finger pecking to careful two-handed typing, until eventually the fingers move
effortlessly and the whole process becomes unconscious. At this point, most people’s
typing skills stop progressing. They reach a plateau. If you think about it, it’s
strange. We’ve always been told that practice makes perfect, and yet many people sit
behind a keyboard for hours a day. So why don’t they just keeping getting better and
better?


In the 1960s, the psychologists Paul Fitts and Michael Posner tried to answer this
question by describing the three stages of acquiring a new skill. During the first
phase, known as the cognitive phase, we intellectualize the task and discover new
strategies to accomplish it more proficiently. During the second, the associative
phase, we concentrate less, making fewer major errors, and become more efficient.
Finally we reach what Fitts and Posner called the autonomous phase, when we’re as good
as we need to be at the task and we basically run on autopilot. Most of the time that’s
a good thing. The less we have to focus on the repetitive tasks of everyday life, the
more we can concentrate on the stuff that really matters. You can actually see this
phase shift take place in f.M.R.I.’s of subjects as they learn new tasks: the parts of
the brain involved in conscious reasoning become less active, and other parts of the
brain take over. You could call it the O.K. plateau.


Psychologists used to think that O.K. plateaus marked the upper bounds of innate
ability. In his 1869 book “Hereditary Genius,” Sir Francis Galton argued that a person
could improve at mental and physical activities until he hit a wall, which “he cannot
by any education or exertion overpass.” In other words, the best we can do is simply
the best we can do. But Ericsson and his colleagues have found over and over again
that with the right kind of effort, that’s rarely the case. They believe that Galton’s
wall often has much less to do with our innate limits than with what we consider an
acceptable level of performance. They’ve found that top achievers typically follow the
same general pattern. They develop strategies for keeping out of the autonomous stage
by doing three things: focusing on their technique, staying goal-oriented and getting
immediate feedback on their performance. Amateur musicians, for example, tend to spend
their practice time playing music, whereas pros tend to work through tedious exercises
or focus on difficult parts of pieces. Similarly, the best ice skaters spend more of
their practice time trying jumps that they land less often, while lesser skaters work
more on jumps they’ve already mastered. In other words, regular practice simply isn’t
enough. For all of our griping over our failing memories — the misplaced keys, the
forgotten name, the factoid stuck on the tip of the tongue — our biggest failing may be
that we forget how rarely we forget. To improve, we have to be constantly pushing
ourselves beyond where we think our limits lie and then pay attention to how and why we
fail. That’s what I needed to do if I was going to improve my memory.


With typing, it’s relatively easy to get past the O.K. plateau. Psychologists have
discovered that the most efficient method is to force yourself to type 10 to 20 percent
faster than your comfort pace and to allow yourself to make mistakes. Only by watching
yourself mistype at that faster speed can you figure out the obstacles that are slowing
you down and overcome them. Ericsson suggested that I try the same thing with cards. He
told me to find a metronome and to try to memorize a card every time it clicked. Once I
figured out my limits, he instructed me to set the metronome 10 to 20 percent faster
and keep trying at the quicker pace until I stopped making mistakes. Whenever I came
across a card that was particularly troublesome, I was supposed to make a note of it
and see if I could figure out why it was giving me cognitive hiccups. The technique
worked, and within a couple days I was off the O.K. plateau, and my card times began
falling again at a steady clip. Before long, I was committing entire decks to memory in
just a few minutes....

I invite you to consider which of the following methods is more in the spirit of deliberate practice in the light of this text:

Doing easy tactics over and over again untill you have memorized them all. Focussing on assimilation of low level patterns in huge quantities.

Doing difficult problems and focus on where and why you go astray and trying to prevent the same mistakes in the future. Changing strategy when matters are executed on autopilot.

27 comments:

  1. This very piece of text was one of my key findings. I remember it very well.

    My interpretation was (together with other research papers) that time pressure is good: That speaks for CT blitz mode instead of CT standard mode. But CT Blitz mode does nothing for deliberate practice. But fails you did under CT Blitz mode are very good candidates to be trained with deliberate practice.
    (Then again, due to limited available puzzles, we better dont CT Blitz too much, otherwise we have too many duplicates later and rating progress measurement is increasingly influenced by the duplicate penalty. But Standard mode is a "no go" for sure!)

    So once we have our collection of fails, we can do deliberate practice with them. I dont only repeat them with spaced repetition, but I also want to solve them fast.
    My aim is to solve them within 10 seconds. (Which time is best? I am not sure. Our STM lasts 20 seconds, and hence I feel we should aim for a time limit below these 20 seconds).

    In type writing, the deliberate practice is probably best spend on words that have statistical relevance. I dont know how to find out which words are causing me trouble to write, but let's assume I found out that amongst other 2 words are causing me trouble:
    a) the word: "don't"
    b) the word: "desoxyrybonuleic".

    I guess I can train both, but probably it is of more value to train "don't" since it will be needed much more often.

    Transfered to chess I cant help but try to persude tempo, that the puzzles we need to train by deliberate practice are the puzzles that are actually "easy" but we failed them.
    They are easy for others, but they are not easy for me. I found 900 failed puzzles in the range of 1300-1475. And currently I am at 1900!
    So actually a 1362 puzzles should be a piece of cake for me.

    But I dont think we should train puzzles that are way too easy. (for me these are puzzles below 1200).
    I fail them hardly. And if I fail them, then it is more likely due to a misclick with the mouse because I rushed through these puzzles so quickly. My thoughts go astray, I dont concentrate as well while I rush through ultra easy puzzles.
    Probably I can master these ultra easy puzzles below 1200.
    While the puzzles in the range of 1300-1475 are still challenging for me if I try to do them fast (solving them in 20 seconds). If I cant solve them within a minute, I press the "give up" button. Because then this puzzle is a fail and is worth to be trained: --> it caused me trouble, right?

    It is probably also o.k. to filter for slow solved "easy" puzzles.
    The main idea is to train puzzles, that are easy for others but not easy for oneself. I call them "trouble puzzle". And an easy puzzle solved way too slow indicates similar trouble, like the word "don't" does not really need to be written wrong, but it is enough to justify a deliberate type writing practice to train words that slow me down in type writing.

    For CT Puzzles it is not so easy to say what can be considered as "too slow". A "fail" is obviously a trouble puzzle. But a puzzle solved in 30 seconds? It depends on the average solving time. If it is 10 seconds, then 30 seconds is too slow.

    ReplyDelete
  2. @Munich,
    of course it is deliberate practice what you do. Without you noticing it, it is aiming at another goal though. It's goal is no longer to become a good chess player, but to become good at low rated chess puzzles.

    But being good at low rated chess puzzles is just a side effect of becoming a good chess player.

    And as I have explained in extenso the past posts, something essential is lacking in those low rated problems.

    I failed when I did my 100K+ problems at CTS. We found the culprit: too less attention for feedback. So it was actually no deliberate practice. I'm sure that when I will add that feedback, I will become a respectable low level problemsolver. Maybe a blitzplayer, even.

    But that is not by goal. So I turn myself to deliberate practice that is in accordance with my goal: to become a good chess player.

    Thus I was wrong, in the past. You gave me no chance to forget that. But I claim the right to err.

    At the same time, this obliges me to grant you the right to err too. So if you ever, somewhere in the very distant future for instance, want to change opinion, there is no longer the need to cling on to an obsolete system ;)

    ReplyDelete
  3. you said: "We found the culprit: too less attention for feedback."

    But I do have feedback.
    I always look up why I failed a puzzle.

    Also I dont think way too easy puzzles are any good. (see my previous comment about puzzles below 1200, starting with: "I fail them hardly...My thoughts go astray..."

    you did 100K, but you did many things different to me (in my opinion very wrong). I explained here why tempos 100K in easy puzzles was no good (and how you should do it right):

    http://empiricalrabbit.blogspot.com/2011/12/important-discovery.html?showComment=1323088767350#c8350270995488441245

    (If I may say so, I explained it in that comment pretty well.)

    ReplyDelete
  4. I don't think we will reach a verdict this way.

    ReplyDelete
  5. I like that:
    "Psychologists have discovered that the most efficient method is to force yourself to type 10 to 20 percent faster than your comfort pace and to allow yourself to make mistakes. Only by watching
    yourself mistype at that faster speed can you figure out the obstacles that are slowing you down and overcome them"

    Solving chessproblems at high speed is uncomfortable. But it works. Especially if you do it on puzzles where you are weak ( which are automatically selecet more often by the spaced repetition algorithm of CT).
    This text describes perfectly the high speed tactics method of munich exept that munich does focus at easy "texts".

    I wonder how that this text can be used to supply your :"make it right and then make it quick" method? It says "make it quick and then make it right"!

    I doubt that anyone can have a (much) better rating- performance at high rated problems than on low rated problems. At least i have the same rating performance on high rated problems as on low rated problems, on problems with low av_time and on high av_time and so on.

    ReplyDelete
  6. I don't think we will reach a verdict this way.

    ReplyDelete
  7. Its easy, get better and we will all follow your path. Till now your results are not convincing. I hope you will have a positive result. Thats the reason why i read every post of yours. I dont like that high speed training. ( But i did start it again, my alternative experiments did not work )

    ReplyDelete
  8. Fair enough. Let's postpone the discussion untill I can show results.

    ReplyDelete
  9. Aoxomoxoa, just to avoid a misunderstanding for other silent readers:
    you said: 'This text describes perfectly the high speed tactics method of munich except that munich does focus at easy "texts"'

    I dont do an "exception" to the text.

    I dont focus on easy "text". It might be easy for others, but I failed these "easy" puzzle because they were indeed too difficult for me.
    These Puzzles are my weakness. I dont do s.th. that the text about deliberate practice suggests otherwise.

    But I could even follow the rule of deliberate practice with puzzles I did not fail: to solve a 1400 rated puzzle might be easy. To solve it within 10 seconds - that is difficult! It is a speed that is a bit too fast for me.

    ReplyDelete
  10. The first step, making a list of the tactical elements, is in fact very akin the systematic application of Heisman's seeds of tactical destruction.

    ReplyDelete
  11. You try to find tactical element quick. That is strong related to munichs method. Your "tactical elements" are just "simple tactics".
    The question is: what method is more effective. Thanks spaced repetition sets at CT i would think its munichs method.
    You would need to create a special chess server to get a optimal workout. Present a problem and detect the "tactical elements" in high speed and Spaced repetition on elements detected to slow. But you would need a special problemgenerator or a community to generate these problems.

    ReplyDelete
  12. Actually, I think it is fair to say, that I already proved it is working with deliberate practice of failed "easy" (not easy for me) puzzles.
    Within a year (21.04.2011-25.04.2012) I did this kind of training.
    I reached my new all-time-high with 1936. Before this, my old all-time-high (=ATH) was just at 1854. Besides, this old ATH was achieved with a higher RD (= higher uncertainty), while my 1936 was achieved with the top RD of 35. (best RD you can get at CT).
    My chesscube rating went up from old ATH @1981 to new ATH 2122 (somewhere close to the date when I reached my 1936 CT rating).
    Currently my chesscube rating is at a new ATH at 2124 (haven't continued to played since July 2012).

    Some people do think I plateaued since then, but this is not even visible in my CT rating graph. Back in March/April 2012 I hit relatively lows at around 1825. Since I started CT blitz again at the end of July 2012, I never hit an low that deep anymore, even though I did approximately 1500 Blitz puzzles since July. My average rating over the last 1000 puzzles is 1897. If you think that it is now much harder to keep up my CT Blitz rating due to the duplicate penalties, this sound pretty promising.

    But I changed my training (which I recommend here all the time) since 25th of April. Maybe that is why I dont improve anymore that much?
    The reason is, that I covered all my fails and slow solved puzzles within the CT Blitz range of 1300-1475.
    I felt that to do them even faster wont yield much more rating, since I had already an average time solving of 11.1 seconds for this range.

    So I started to deliberate practice all my fails from 1475-2000.
    I am still busy in learning them, but I have allready covered 1350 out of 1500. So soon I will have learned them all in an SRS-set. And then I will continue to do them for a while.
    Maybe covering these higher rated puzzles are not as valuable like my fails in low rated puzzles? That could explain, why my CT Blitz rating does not show such dramatic improvement like before (though I think I did improve a little bit, and a new all-time-high is well in reach soon).

    After I will have covered all my fails in the range 1475-2000 then I dont know how to continue.
    I might go back to my 1300-1475 range and try to make them even faster.
    And after that? What to do then? Well, I will look at all my slow solved puzzles. And then? Then I hope I will already be well above 2000 in CT blitz rating. Because otherwise I would not know how to continue. I will be running out of puzzles to train.

    But all the "follow up"-trainings I consider to be less effective than my training of the failed low rated puzzles. I unfortunately covered this range.

    ReplyDelete
  13. @Aox,
    I'm convinced that I can speed that up, one way or another. I already improved from 40 minutes per problem to 10 minutes. That is the least of my worries.

    The speed of recognizing all elements has no influence on the succesrate.

    The crucial point is that tiny factors play a major role. The tactical elements interfere with each other and I must learn to mould the combination myself. This is the heart of the matter.

    ReplyDelete
  14. I don't think that rote learning of patterns has anything to do with pattern recognition. That's not how it works.

    I don't think that rote learning has anything to do with deliberate practice.

    ReplyDelete
  15. What is the difference in learning a pattern with deliberate practice, and rote learning a pattern I failed before?

    Reminds me of the following question of Dan Heisman:
    "Do you know your name - or did you just memorize it?"

    ReplyDelete
  16. The pattern is a cognitive construct.

    Probably you get mixed up with "rote learning the moves", and not learning the pattern.
    But "rote learning the pattern" does not make much sense to me.

    For pattern, I am willing to agree that it can just as well be a "guidance/ logical reasoning proscedure".

    On times I dont explicitly name all 3 (pattern/guidance/logical reasoning), but call it pattern.
    After learning quite a lot of them, I cant really tell the difference between the 3 terms, because there is often no difference.

    Let's look at a simple back rank mate:
    A "pattern" would be:
    black 3 pawns h7/g7/f7 and a black king on g8, white rook gives check mate on a8#.

    Another "pattern" would be:
    black 3 pawns h7/g7/f7 and a black king on g8, white rook gives check mate on b8#.

    and so on...

    a "guidance" would be:
    A black king on the back rank and 2 or 3 pawns that block his escaping squares. For instance a king on h8 and 2 pawn h7/g7 is similarly "trapped" like a king on e8 with pawns on d7/e7/f7.
    Such a king is prone to queen or rook side checks, giving check mate at the end.

    So a guidance includes hundreds of patterns.

    "logical reasoning" would be then:
    if there are no queens or rooks on the board, then it is useless to look for back rank mates - unless a pawn could get promoted to a rook or a queen.
    With logical reasoning I am able to exclude some things.
    Sometimes the king is on a white square, so I need a piece that can give checks on white squares.

    The differnce between logical guidance and logical reasoning is basically, that logical reasoning is made of several guidance parts or patterns.

    I try an analogy: But the pure basic element, the atoms of chess, that is the pattern.
    Guidance are certain molecules, and logical reasoning is the reaction between molecules.

    Back to rote learning of puzzles that cause you trouble, versus deliberate practice of puzzles that cause you trouble - well, it is actually the same.

    Example of a pattern where you dont think much, because you rote learned the pattern:
    http://chesstempo.com/chess-problems/50570

    The cognitive process in solving the above puzzle simply died.
    But back in the past, you had to derive it with a cognitive process. But meanwhile it is so deep rote learned, that you see it very quickly, without realy thinking much.





    ReplyDelete
  17. I don't think that rote learning of patterns has anything to do with pattern recognition.

    ReplyDelete
  18. You mean such as memorizing your name would not lead to recognise it?

    ReplyDelete
  19. Pattern recognition is about similar yet different patterns. Rote learning is about the exact same pattern. This are different animals.

    ReplyDelete
  20. I guess, either you or me still dont understand.

    Here I give you two examples of exactly the "same" pattern (sample "A" and Sample "B":

    http://chesstempo.com/chess-problems/56218

    http://chesstempo.com/chess-problems/77497

    And now an example of a pattern, that is similar to the above pattern, but a little bit different (Sample "C"):
    http://chesstempo.com/chess-problems/49178

    So you are saying, that if I "learned" A, but not pattern C, then I wont recognise B, even though A and B is the same pattern?

    ReplyDelete
  21. Your text describes my learning process well.

    In a game, I have about 3 minutes per move. The time that I have to look for tactics is a fraction of this. With a problem, I have the advantage of knowing that there is a simple tactic to be found.

    I want to find my faults in the simplest possible circumstance, to provide a clear diagnosis and make effective use of time. It makes sense to train with the easiest problems that give me trouble with an average time per problem in the region of one minute, and try to fix any faults in my thought processes. Fixing my faults temporarily does not help, so I need to repeat over time.

    The problems that I got wrong are the harder ones, so just training those is equivalent to training on a harder problem set, which is not what I want. Problem difficulty also depends on context, and problems that are easy on one occasion are not always easy on another. Revisiting simple problems that cause me difficulty more often than the others is appealing nonetheless.

    ReplyDelete
  22. @Munich,
    No, what I say is that pattern recognition is something different than rote learning.

    Allthough pattern recognition accidently might appear as a side effect of rote learning, by no means does that implicate that rote learning is a way to improve pattern recognition.

    ReplyDelete
  23. @BK,
    The text "Amateur musicians, for example, tend to spend
    their practice time playing music, whereas pros tend to work through tedious exercises
    or focus on difficult parts of pieces."
    points in another direction.

    ReplyDelete
  24. Sorry, I still dont understand.

    Puzzle A and Puzzle B have the same "pattern".
    I rote learn this pattern (which is somehow a conflicting sentence, because the "pattern" is a certain smother idea, and requires understanding, which rote learning does not).

    Anyway, I "rote learn this pattern" and even though I know now this pattern, I would not be able to recognise it? So after I extracted the pattern from A and B, I would not neccessarily extract it from Puzzle "D" (so you say)?
    So I would not neccessarily recognise the same pattern in puzzle D?

    Puzzle "D":
    http://chesstempo.com/chess-problems/53672

    And why is that? Do you have any idea, how I can learn the smother pattern, without recognising it in puzzle D? Because I have trouble in imagine how I can "rote learn" a "pattern" (a principle that requires understanding).
    I could only imagine rote learning as learning the move sequence. I mean "the exact move sequence" like puzzle A and B have. (they really have the same annotation for the moves!). Then of course puzzle "D" would not be recognised. But rote learning the move sequence would not mean learning the pattern, since a pattern is an idea, a principle, a set up with certain key actors (here queen, knight, cornered king, and 2 pieces blocking the escaping squares of the cornered king).

    Actually, the definition of rote learning does not fit to learning a pattern (which requires understanding).

    I guess (like before) you still get very mixed up with learning the move sequence and learning the pattern (which is not the same of course). Or it might be me who does not understand, but you are not really talkative with 1 or 2 sentences, so I have little chance to decode what your point is.

    ReplyDelete
  25. P.S. you just quoted about "pros concentrate on difficult parts".
    What is difficult?
    For a musician, it probably is the part of a certain music piece, where he stumbles and hits the wrong key.
    What is difficult for him, might be easy for other professional piano players. "Difficult" is, what you cant do (well). Translated to tactics: Difficult puzzles are the ones, that you fail.

    ReplyDelete
  26. Guys,
    these dicussions lead us nowhere. I'm going to end it. We are apparently bishops on a different color. We live in a different universe. It's going to cost me too much time and energy. You will only be convinced by results, so you will have to wait.

    ReplyDelete