Sunday, January 31, 2016

Pushing the envelope

Back to the basics.
Having done a lot of mate in 1-hard (M1-h) lately, it becomes clear that it is important to have your chess atoms in top condition. Under chess atoms (the chess basics) I comprehend the four basic actions of Radovic:
  • Attack
  • Restrict
  • Block
  • Protect
 In a way 15 seconds per M1-h is not so shabby, when you consider the workload you process in those 15 seconds. Yet I feel it should be possible to do it much faster. But in order to do so, I have to become much faster at the basics. I don't think it is a waste of time to work on M1 solely. When I do chess problems at CT, it is evident that I suffer from the same weakness, being slow with the basics. I expect that improvement of the basics acquired doing M1, fully transfers to the other regions of tactical problems.

Comparing mate in 1-easy (M1-e) to mate in 1-hard (M1-h).
 I compared the M1-e with the M1-h. The main difference that I found is that with the M1-e the king is often at the rim of the board, while with M1-h the king is more often in the middle of the board, hence you have to check the aura of your pieces to see whether the cage around the king is complete.
After a few hundred M1-e I reached a speed of 12 mates (e) per minute, while after 6 days M1-h my maximum speed is 5 mates (h) per minute. So the difference in speed between M1-e and M1-h is mainly caused by slow aura vision (chess atom = restrict).



Conscious versus automatic learning.
There are two methods of learning you can try when you want to improve your speed at M1. One method is based on conscious learning, the other on unconscious automatic learning. With conscious learning, you visualize-II the solution for a while, in the hope that somehow you remember the visual characteristics of the position, so you will recognize the same type of position faster in the future. With automatic learning, you don't consciously look at the solution for a while, you just try work as fast as you can. My experiments with Troyis have proved that it is possible to improve automaticly and unconsciously. Just do a lot of exercises as fast as you can.

It seems that Thomasz and Aox tried to improve at M1-e the "Troyis-way". Just pushing the speed limits in an effort to become faster. This method seems subject to plateauing after a week or so of intensive training, just like Troyis. It has proven to be very difficult, if not impossible, to go beyond a certain threshold, despite a lot of working in the salt mines.

Improving with the automatic learning method.
If you plateau with automatic learning, it means that the subject you try to learn is too complex. So you need to divide the complex task in several simple subtasks, and train them separately. When you speed up the subtasks, it follows that the main task will be performed faster too. It looks logical to split up the database with M1-e problems. If I'm not mistaken the program of Aox makes use of a simple javascript array with FEN positions. I assume that it suffices to just replace this array.

The problem is, how do you select the appropriate positions? Scid has some possibilities to select positions based on their used material. Maybe that can be put to use.

Improving with the conscious learning method.
The method of conscious visualization-II has not been proven yet. This is the method I want to try first, since the other guys already have tried their hand with automatic training.

Either way, the journey will be very interesting.


9 comments:

  1. Tomasz had an improvement at m1 of something like 10+% which is a lot! if you consider how fast he already was, so we dont realy know for shure that m1-e is not improvable for every Adult.

    BV find all checks supose to help to improve one subskill to the m1-e problem
    With some hard thinking we may find 10?, 20? other useful subskills...
    Of course its not necessary to write a specialised trainingprogram for these, a thoughtprocess might be enough ( you just have to forget about measuring the progress )


    ReplyDelete
    Replies
    1. I am not sure, but you are not correct my friend. I have started testing the M1-e puzzles about 16-18 months ago. My first "ceiling" was about 37-38 MPM. At my All Time High (the best score) was 46-47 MPM (not all the time, but at least a few times at the set of few dozens of puzzle). It means not "something like 10+%", but even up to 20% (!) (38 +7,6 = 45,6). And it is the score without perfect (optimal) condition. It means even 50 MPM is possible (if I could do it - I would publish a screenshot with such a score).

      And if I could make a progress of about 40% (from 37 up to 50) we could say "it is significant" - do you agree on that?

      Delete
    2. it would be an signifficant improvement
      The problem is that you have already the skill tempo and i would would like to get, you have vision, but therefore you cant calculate , your k is to small.
      Therefore Tempos and mine k is much higher.
      You should work at your calculation skills and not at vision, with only a little better calculation your tactics would improve dramatically.

      Delete
  2. I am still testing M1-e, but due to some technical problems - I cannot check out how much I could improve.

    At present I have just done 700 puzzles with the speed of 40-42 MPM. If I could have a database of the positions - I could reach 50 MPM withing about a months of hard training.

    What hampers me the most is the screen - I cannot see the pieces being at the bottom of the board (as I have to see the King's mate square at thet top). And whenever I shrink the board - it is to small to recognize all the pieces. I regret I cannot write (or obtain) such a program to extract the PGN (or FEN) mate puzzles. I am EXTREMALLY curious how much gain can be made with extensive training with optimal conditions. If I could break 50 MPM - it would mean that progress at this task is possible. And my dream is to solve the puzzles with the speed of 60 MPM. If I could reach this (for at least 100-200 puzzles set) - I would be more than happy! :)

    I think it is a good idea to create the database of all the possible mate patterns. After that it should be memorized - and it may be another element that helps us speed up the task of solving M1 puzzles!

    ReplyDelete
  3. As for now - I solved 2500-3000 M1-e puzzles. It seems that 45 MPM is my actual "ceiling". Now I will have to work out the method how to practice the elements you listed above. The most important ones are these: box squares (around the King) pins, attacking squares.

    What is your idea about the training box squares AFTER the move has been made? Sometimes you can see that all the squares around the King are attacked, but AFTER you move your piece - you release some important squares (and there is no mate). Any ideas?

    I am sure that practicing the elements mentioned above is worth at least 10-15% of progress! Breaking 45 MPM just "practicing hard and long enough" is not reasonable idea as I have solved about 20-30K puzzles with this "method" and it does not work anymore (as I have achieved to high level of speed).

    ReplyDelete
  4. In case it helps you guys, for any mate in 1 puzzle I always look first at what color square the target king is on. Then I search for the same colored bishop, Knights on the opposite color, and then the queen. Queens have fewer checks if on the opposite color. Finally I search for rooks and nearby pawns. In doing that I also see their checks and start understanding the position (pins, blocked squares, defended squares, discoveries, etc).

    Give it a shot -- most of the time I just say the color in my mind intentionally and do the rest automatically

    ReplyDelete
  5. PART I:

    @Dan:

    That's an interesting approach. Elimination of possibilities (ergo, concurrent reduction in cognitive load) is included implicitly in your process: Bishops of opposite color, Knights on the same color, Queen on the same color are eliminated from consideration (at least initially).

    Temposclucker previously emphasized the importance of training over speed, i.e., that speed should be a byproduct of training and not the primary goal. I concur with that idea and would expand on it somewhat.

    For learning to be long-term, a certain period of time must elapse between the solution of the "problems." During that time, reflection as to the specifics of what is being learned should be reviewed, by putting the concept into your own words and "hooking" it to other (somewhat) similar problems by categorization and analogy. This strengthens the neural connections and enables the mind to "see" similarities/differences between previously seen problems and the current problem. (Note that "differences" may be of more importance than "similarities" in terms of "vision," i.e., discrimination.) At a relatively high rate of speed, the practice becomes more and more like "massed practice" (also known as "cramming"). Although it "feels" like massed practice produces immediate results (and it actually does do that), research shows that the long-term retention of what is "learned" drops off sharply in a very short period of time. More about this below in PART II.

    I am NOT against the idea of timed testing. Testing itself has shown to be very effective on two fronts, one of which is widely known. Obviously, timed testing reveals weaknesses. However, the less well-known phenomenon is that testing/quizzing actually enhances long-term retention. In order for testing to do this, there must be some mental effort required to recall the context and concepts that represent the "essence" of the test problem. There also must be a time interval between "learning" and "testing" in order for a certain amount of "forgetting" to occur. This "forgetting" then requires mental effort to recall the appropriate essence, thereby strengthening the long-term retention.

    If we view the solution of mate-in-1 as a "test," then the missing part is the "learning" (reflection, abstraction and conceptualization) that is required for long-term retention of the appropriate relationships. The attentional focus in solving mates-in-1 may (or may not, in most cases, if speed becomes the primary objective) provide long-term benefit in other aspects of playing better chess.

    The ideas above are not mine. They are to be found in make it stick: The Science of Successful Learning by Brown, Roediger III and McDaniel, The Belknap Press of Harvard University Press, © 2014.

    ReplyDelete
  6. PART II:

    There are many more excellent ideas contained in the book, almost all of which are applicable to learning to play better chess. Highly recommended!

    Here is an excerpt from Chapter 3, Mix Up Your Practice.

    "Most of us believe that learning is better when you go at somethng with single-minded purpose: the practice-practice-practice that's supposed to burn a skill into memory. Faith in focused, repetitive practice of one thing at a time until we've got it nailed is pervasive among classroom teachers, athletes, corporate trainers, and students. Researchers call this kind of practice "massed," and our faith rests in large part on the simple fact that when we do it, we can see it making a difference. Nevertheless, despite what our eyes tell us, this faith is misplaced."

    "If learning can be defined as picking up new knowledge or skills and being able to apply them later, [emphasis added] then how quickly you pick something up is only part of the story. Is it still there when you need to use it out in the everyday world? While practicing is vital to learning and memory, studies have shown that practice is far more effective when it's broken into separate periods of training that are spaced out. The rapid gains produced by massed practice are often evident, but THE RAPID FORGETTING THAT FOLLOWS IS NOT. [Emphasis added.] Practice that's spaced out, interleaved with other learning, and varies produces better mastery, longer retention, and more versatility. But these benefits come at a price: when practice is spaced, interleaved, and varied, it requires more effort. You feel the increased effort, but not the benefits the effort produces. Learning feels slower from this kind of practice, and you don't get the rapid improvements [MDLM, anyone?!?] and affirmations you're accustomed to seeing from massed practice. Even in studies where the participants have shown superior results from spaced learning, they don't perceive the improvement; they believe they learned better on the material where practice is massed."

    ReplyDelete
    Replies
    1. I think that my learning from books on tactics like martin w book, how to beat your dad at chess and tactics in the context of a complete classic game have been helpful in supplementing 7 circle type training. Though it is difficult to attribute how one knows what one knows.

      Delete