Thursday, March 22, 2012

More about plan A

The past 3 years my positional play has improved a lot. I have much more knowledge now, both about the openings I play and about the accompanying middlegames. In 90% of my games I have a clear plus in the middlegame. Yet my rating has nosedived the past 3 years from about 1850 to 1700.

Since I have improved in almost all areas of chess, there can be only one explanation for this phenomenon. I haven't solved any serious tactical problems the past 3 years. For the very reason that in the 5 years before I had done a bit too much and I was simply fed up with it. In 3 years my tactical skills have detoriated. Due to neglection of "plan A".

Besides that, I felt that it was necessary to improve the trainingsmethod first. And in that area I have made the biggest progress the past 3 years.

Munich has implemented my suggested improvements of plan A and proved that it has become a viable method. He has made a considerable progress in little time and he has been  so friendly to describe how he precisely implemented the ideas:

  •  No DIY = it is not important that you get a headache and find the solution by yourself. It is much faster to understand the tactic. Think short about it, make your move and pick up the solution, especially if you did it wrong.
  •  Reduce complexity of chess (try to isolate what you want to learn)
  •  Add intelligence (train only what you cant do well)
  • Automate the new gained knowledge (you need to be able to apply it subconsciously)

  • Deliberate practice (part of adding intelligence: you only train puzzles you did slow or wrong)
  • Tag-sorted training (part of No DIY and part of reducing complexity of chess: of course it helps to know you look for a fork. It speeds things up in adding intelligence. At the end you only need to understand the tactic. You know for instance it is a fork before you even start the puzzle, it makes your live more easy. Also, you will see typical patterns more often, which reduces complexity and helps your understanding in finding and applying typical patterns.)
  • Spaced repetition (part of automating new gained knowledge and part of reducing complexity in chess. The later is so, because spaced repetition is the most efficient way to remember what we learned, so it is the fast track to success.)
  • Rating sorted sets (part of no DIY and reducing complexity: if you know what difficulty to expect, it makes live more easy.)
  • Train puzzles that are easy for others but not easy for you. (part of automating and adding intelligence and reducing complexity: you don’t repeat puzzles you did very fast. But you need to do an easy puzzle at least once to check if you were able to do it very fast. If you are, than you just wasted a few seconds to eliminate this puzzle from the learning list. It is a little sacrifice. After you went through the whole set of ie easy fork puzzles, then you only concentrate on puzzles you did wrong or slow. These are the very puzzles that are easy for others but not easy for you. You miss important intelligence.)
  • Statistical relevance (the easy ranged puzzles are puzzles which expert players and master players failed to see during long tournament games. They were easy tactics which are easy, but not easy for the master who just failed to see it and hence lost the game. So even master players should go through easy puzzles and see if they know them all.)
  • Puzzle range (I recommend easy puzzles, but not too easy puzzles. It is a bit of a trade off. If you can solve 90% of all puzzles that means you need to check 10 puzzles to just find 1 puzzle that is not easy for you. If you can solve 98% that means you need to check 50 puzzles to just find 1 puzzle that is not easy for you. And chances are, that it was a mouse slip anyway. So don’t do too easy puzzles. If you do puzzles rated 320 points below your rating, that means you will solve 90% of them. But since you want to solve them very fast, you may go lower than 320 points below your own rating. I suggest a range 450-350 points below your peak all time high CT rating . In temposchluckers case the range should hence probably be a CT Blitz rating range between 1350-1450. I assume his CT Blitz rating peak ATH to be at 1800.)
  • CT Blitz rating as gauge only (The puzzle rating is more stable, because it takes time into account. Since we aim to do puzzles fast we take time into account. Using standard rating is just a bad habbit, if you have CT Blitz rating available, too. A 5 mover takes longer than a 2 mover. )
  • A puzzle is known if you can find the solution without much hesitation. I suggest to aim for solving a puzzle within 9 seconds. A mate-in-2 maybe even faster.

All these leads to following training plan:
  • Tag-sorted (ie all forks, or all mate-in-2)
  •  Rating sorted (reduce complexity, No DIY)
  •  Easy range but not too easy. (add intelligence, the range I suggest to be 450-350 below your CT Blitz all time high rating (your peak rating)).
  •  Repeat failures and repeat slow solved puzzles (add intelligence/ deliberate practice)
  •  Spaced repetition (until you know them very fast. Really, it is beneficial if you  memorize them. Even literally, don’t worry about this. You will recognize the learned puzzles in your own games, where you will compare your knowledge with the actual situation on board. It enables you to calculate faster. Really. GMs draw heavily from there Long term memory and don’t calculate much. They “know” the tactic.)
I couldn't have said it better:)
I have started a trainingplan according to these principles.

Besides this I have to work on plan B, of course. Plan A is about the patterns of the pieces. Plan B is about the patterns of the squares. Exercising according plan A only brings you limited knowledge of plan B. Mr. Z expressed it beautifully:

The problem is we learned chess too materially. If I'd to start over again, I'd say, the squares are exactly (or more to be safe) important as the material. I used to wonder why they came up with the result in those eye following test that GMs looked at the squares dramatically more than amateurs, and this is why. I also answered the question I posed to you some time ago, why is chess so different? One of the reasons for me, because it is played on a deceptively flat map (or terrain). I mean, in starcraft and in real life most people know that holding a higher point like a mountain is an advantage, or being pushed into some narrow cross is bad, but where is the mountain top in chess? It is a flat board for your eyes! At least it looks like that. But really it is the playing field. It isn't flat at all, moreover it changes its highest points turbulently, better pay attention to it. Your pieces can be isolated, walled in, they can be pushed into a swamp of pawns, one knight on a mountaintop can barrage over the whole battlefield. You must go around it, or siege it just like you'd do if you see it in real life (not hidden by a flat board).
 We have to invent a method to make the invisible patterns of the squares visible before we can even think of training the patterns of the squares.


  1. All these leads to following training plan:

    Repeat failures and repeat slow solved puzzles (add intelligence/ deliberate practice)

    Spaced repetition (until you know them very fast. Really, it is beneficial if you memorize them. Even literally, don’t worry about this. You will recognize the learned puzzles in your own games, where you will compare your knowledge with the actual situation on board. It enables you to calculate faster. Really. GMs draw heavily from there Long term memory and don’t calculate much. They “know” the tactic.)

    The spaced repetition at CT will automatically repeat your "failures und to slow solved puzzles" more often then the quick solved problems. There is no need to do something extra anymore.

  2. Spaced repetition: Which is true.
    However, the very first go of your set, I recommend rating sorted. This is not provided by spaced repetition.
    And at the very first time, there is no historic record of slow solved puzzles, because you havent solved any puzzles, yet.

    Then on your seconds walk through, then you may consider spaced repetition. I recommend still rating sorted and filter for puzzles you solved slow and filter for puzzles you did wrong. But if you decide to not do spaced repetition, make sure it doesnt take too long to cover your set. You found out for yourself, that if you have not seen a puzzle within 8 days, then you have completely forgotten about it. So you should cover the first walk through in 4 days or less. At the end of the day (on the same day) you repeat what you did wrong.
    The slow solved puzzles, you may solve after you went through all of them. After day 4 if the set took you 4 days.
    Then maybe again looking at all wrong solved puzzles.

    And only then I would start to put them into a spaced repetition set, and richards algorythm takes over. Richards spaced repetition algorythms are only good if there is a little history available. And doing them rating sorted is simply reducing complexity (no DIY).
    You also see very similar patterns more often clustered together, because similar patterns have similar ratings (= my experience). In that way you will notice patterns which you would not really notice as a pattern, because the pattern dont seem to be so sharp and clear.

  3. If i do a problem wrong in a spaced repetition set, then i will see it again after a few minutes, then again after 1 day and so on. No need for any special action.

    Richard did some modifications. As guicker the problem is solved as longer the space between the repetition, and the space i doubeling each correct repetition.

    I see every! first-time-problem again within something like <2 days then again after <4 days and then after <8 days, so the difficulty of not seeing a problem in 8 days dont exist here. In these 8 days i have already repeated the problem 2-3 times.

    These SRS do a lot by themself. And Richard will do some improvements.

    My set is not sorted for rating, thats to dry form me. I have high bloodpressure, to much Salt from the saltmines is not good for me

  4. Aox, did you understand my point, that you first have to solve a first timer to see if you can do it fast or slow?
    The Spaced Repetition set cant know how long you will take to solve a puzzle before you have solved the first timer.

    As long as there are traces of memory left, the repetition is not too late.
    If after 1 days or 4 days - the longer, the better, but, too long and you forgot all of it or the protein traces in your memory are too destroyed to be repaired.
    Richards algorythm is a "one fits all" algorythm and is repeating a bit too early. Nevertheless, it keeps repeating and the spaces (time gaps) between repetitions are increasing.

    You did not understand my point, that similar patterns have similar ratings, and the patterns sometimes dont have clear sharp patterns you would easily identify as a pattern. I am thinkg of "queen and knight are in harmony if they stand on the same colour and two diagonal squares seperate them" pattern.
    Since there is no history in the first walk through, spaced repetition will do nothing for you. That richard repeats wrong solved puzzles allready after 5 minutes is debatable, but at the end I believe it doesnt matter, because it will also be repeated the next day. Slightly too often repetitions (for my taste) will just result in a little extra work. At least for me. It is one size fits all, and if somebody has the attentions span of a flee, then it might be good that the repetition is already after 5 minutes.

    But the spaced repetition throws away the possibility to have them rating sorted in the first go through.
    I am repeating myself here, but you didnt understand my previous writing, so I try again now.

    Then again, even if you do right from the beginning a spaced repetition set, you will learn the puzzles, too. So it is just about optimizing the training, making the most of your training time. It makes only a little difference in the beginning. Later not anymore, cause I believe, that if you have the puzzles in your memory, then you will also see the similarity of not clear patterns, too. And if the rating range isnt so far appart like mine (I did 1150-1475 at the beginning) then it doesnt matter much, too. The closer the top and bottom of the rating range, the less the rating sorting will have a useful effect.

  5. It is not neccessary to see, if i can solve a problem quick or slow. The algorithm of SRS at CT does the work.
    If i can solve a problem quick then i dont see the problem that often. In my whole (rest-)life i will see such a already known problem max 7 times. You may improve even on things you already can do well.
    I hope Richard can do his improvements on these sets qucik, so we can improve further.

    If i need much time for a problem, then i see the problen quicker=more often. If i cant solve the problem i see it again imediatly.

    The supermemo algorithm is made to learn a set of information (usually sets of several thousands vocabulary) as effecient as possible. You try to find an algorithm who does it even more efficient.

    It make sense to do the training sorted by rating, but i am not able to do it to often, its to boring. Motivation is an important element of sucessful training to. I need my salt spiced.

    The repetitions of the SRS at CT are not to often, they are a little to seldom for me. The average space increase factor is 2 at CT, 2.5 at SM2 but i need something like 1.8 ( like Empirical Rabbit). SM2 is created to keep information in the memory, we want to learn (improve) by SRS while we do the repetitions so we need more repetitions. The SRS at CT are repeating to soon( after 5 min and/or 1 day).

    I do a SRS with ~11000 mating problems rated from 0-1500. Every day i do my repetitions of the problems i already "learned" so far these are the sheduled problems. Then i do ~100 new problems. So my set of "learned" problems is getting bigger every day. at my speed at the moment it will take me 110 days till i see the last problem of my set, then i will do only repetitions.

    My analysis did show that a training of a pattern need to be continued for a long time, 110++ days should be a good start.

    This big sets enables me to watch how my performance on firsttimer develops. If i do a training like you suggest then i cant watch my improvement on firsttimer.

    This captcha goes on my nerves, 3 .repetition now

  6. no, you cant see how you do on first timers. If it is of interest besides the fide elo rating estimate, then there is nothing wrong with starting from the very first beginning a SRS.
    I believe it is not optimal, but it is not a huge difference. At the end of a SRS you should know most puzzles. That is the aim in both our approaches.
    Motivation is very important, because I very well can understand how dry these salt mines are. Very dry indeed.
    I try to get my motivation in solving them very quickly. I want to get it done as quickly as possible. Any "hint" will help me to solve a puzzle quick. and the hint that I know where the puzzle is rated is making my live more easy.
    With such a super size set like you do, then there is of course no other option then to do a SRS right from the beginning. I was thinking of set sizes below 1000 puzzles and a motivation to do 300 and more per day.

  7. Tempo, I was looking at your profile in CT. While I said I recommend 350-450 below your all time high, I was thinking of your active ATH, or a reasonable ATH. The upper bound of the fluctuations. You reached once 2100, but that is far too high to be a serious ATH.
    The name of your set is "non spaced" which isnt such a ideal name. Why not calling it what it is, rather than what it is not? I would call it "fork" if it is a fork set.
    Though I believe you dont do yourself a favor at the range you have chosen, at the end all that counts is probably that you memorize them.
    If it is not a special tag set such as fork or "mate in x" or "discovery" or whatever, then I wonder why you dont do spaced repetition. Then you dont need to go through puzzles so often you solved in reasonable time. Not doing spaced repetition set is only an argument if you want to do them rating sorted. But a set from 1800-1810 is allready rating clustered. They are all the same, so the benefit of rating sorted is zero. Here I would strongly recommend to do them with a spaced repetition. Simply create a new set with "Problems I tried" in the range of 1800-1810 in a spaced repetition set, and no harm is done and the time you invested is not wasted. Instead the spaced repetition algorythm is looking at your previous solving times and serves you your trouble puzzles.
    I am curious if you will soon notice your learned puzzles in your real games. Chances are a bit lower then with me, because the puzzles in the range you chose wont occur so often in your games, while you will miss a lot of easy tactics in your games that could have been possible. You will see...
    But in any case I believe you should become stronger, no matter what range you do.

    Good Luck.

  8. You are a tactics fundamentalist.

    If I'm going to do the same as you, who is going to invent new methods?

    The set is called NON SPACED since it is, well, non spaced. To distinquise it from the sets that are called, emm, SPACED.

    The set is not called FORKS since, well, it doesn't contain solely forks.

    You adjust your sets to your ATH, I adjust my ATH to my sets;)

  9. But then your training differs very little from people who do simply CT Blitz on hard mode. The only exception seems to be, that you repeat them. But do you actually only repeat blunders and slow solved puzzles? looking at your history, it looks as if you repeat everything.

    Doing a different rating range is the point I am most likely to be not so strict with my views. It could work, too.
    But not doing tag-sets is against reducing complexity, too.
    Doing all puzzles, even the ones you can allready do is against adding intelligence (against deliberate practice), and since your range will likely result in solving times well over a minute, it will take probably too long to cover the set? Then the repetition sets in too late you would be much better of by putting your set into the spaced repetition algorythm. No puzzle would get lost, it would still be the same set, but in between the repetitions starts when they are due. If they are not due, then there is no difference with a normal set.

    As long as you will memorize your puzzles it should still work. The only question is: will you be able to memorize them if you dont take help, such as tag assorted sets? (no DIY!)
    For me it easier to memorize a poem that rhymes, then to memorize a poem that rhymes not and is written in japanese. Both is possible though. However, the success rate of most people who do intensivley CT Blitz on "hard" mode seems to be close to zero for matured adults. If your method is hardly different, the outcome will likely be the same.
    with my method I increased about 200 chesscube rating points within 11 month. My CT blitz rating rose considerably, and my felt strength is telling me, that indeed I jumped from A-Class level into expert level (as long as it is not an endgame).

  10. @Munich,
    you were supposed to say now:"Interesting! With what experiment are you busy at the moment?" instead of slapping me with my own advice:)