Sunday, December 04, 2011

Context and guidance

I don't have the feeling that my idea about guidance comes across very well. Let me try another angle of attack.

First I want to make one point very clear. I don't have anything against speedtraining. As a matter of fact, at this very moment I do speedtraining of the overworked piece tactic with a low rated problemset at Chess Tempo at moments when I'm too lazy to think for myself. It works and it should be trained.

But don't let the results of this interfere with the results of guidance, since that is quite a different animal. First you should read this little story.

What did I learn from this? The mind is protected against cognitive overload by context restraints. You will only find those patterns which match the actual context. I looked for two hours at the position for a positional clue (I had not invented "No DIY" yet) without noticing the mate in 3. After the context was changed, I recognised the mate within 20 seconds. So the patterns for the mate were familiar enough.

The 2200-2300 rated problems with 1-4 moves are only difficult for us due to a limited context. The context limits the patterns we can recognize. No matter how well the patterns are automated. 

Remember the words of Munich:

Still, in the very fist puzzle (with the distraction theme) I only considered the check ...Rf8+, but not the check ...Bg3+. Nevertheless it is good to be able to find 35 checks/ minute instead of 6 or 7 checks per minute (the value I started with when I did the check-training).

This is an example of context-restraint pattern recognition.

Now what am I trying to accomplish with that guidance stuff? A checklist with logical questions guides you through the right context. It protects you from cognitive overload by restraining the patterns to be recognized by their relevance to the context. The mind is beautyfully organized!

Did you notice that whenever the context is changed by a hint that you slap your head and recognize the relevant patterns at once? Not only are you able to recognize the patterns, but you are able to visualize them too.

Take for instance this position from a video of GM Rowson (see diagram).

White to move and win (that's a context!)
Rowson couldn't solve this problem within 20 minutes. He complained that he suffered from a cognitive overload. There are just way too many possibilities, even with only 4 pieces!

You can formulate a few concepts that will limit the amount of possibilities. For instance:
  • A check is always safe.
  • Pinning the pawn is always safe.
  • Covering the promotion square is safe when the king doesn't protect that square too.
  • If you can put your queen on c1 there is time to approach with your king
  • The same is true when you can force the black king towards c1
  • Beware of stalemate with the king on a1 and the queen taking c2
  • Mate enters the equation when you can win tempo's for advancing the white king towards the 3rd file.
  • Some mates allow promotion.
  • Beware of minor promotion to knight.
All these concepts narrow down the amount of moves to consider. Hence they diminish the cognitive overload. Yet that is not enough in this position. You must be aware that there is another concept hidden that plays a role in this endgame: Sometimes it is possible to play a skewer. For instance:
1.Qd5 Ke3 2.Qg2 c1=Q 3.Qg5 wins the queen.

This gives you a bit of an idea how guidance by concepts works. How it diminishes the cognitive overload. How it enhances the visualization. And that even grandmasters are lost in a position with only 4 pieces due to cognitive overload if they lack a concept.

You might argue "yeah, but I remembered the problem xyz so I could solve problem abc. So there is no difference between a concept and a pattern in the sense that they both provide the cues for remembering how to solve the problem." Well, that's true. But don't nihilize the importance of context and its restraining effects on pattern recognition. You need these restraints to prevent you from cognitive overload but at the same time they must not exclude relevant patterns. You can't steer that by pattern recognition alone. You need logical reasoning for that.

My checklist has now 136 questions of which about the half are about tactics and half about positional play. I'm going to test the tactical questions against a high rated problemset.


  1. Munich (drop the "me"): Thanks you for the explanation. This endgame puzzle is indeed a good example, and it convinces me that there is that co-exist next to pattern recognition. Somehow I cant tell the difference between these two. But then there are puzzles like this endgame here. There are patterns in place, but the other thing too: a guideline how to tackle this problem.
    On first thought one might think it is only existent in endgames. But on a second thought - the guidance is just more obviously present in endgames, simply because of the absence of pattern.
    If you had not used the word "guidance" but the word "plans" I might have understood it a bit sooner. (But I admit that "guidance" is a more precise word).

    I suddenly remember a lot of advises from books:
    - start learning endgames
    - treat the middle game as if it was the endgame
    - don't just learn the openings but try to understand the plans behind them.

    These advises are too general to make us understand what their intention is:
    learning these "plans" that you call guidance.
    Patterns are the building blocks of tactics.
    Guidance-rules are the building blocks of strategy. They are the tiny parts of the whole plan.
    There is interference between these two types, too.
    For the very distraction puzzle my guidance elements would be this:
    "If many opponent pieces are clustered together check every check especially carefully - it can lead to distraction or overloading."
    I believe that such a clue could have been enough for me to find the Bg3+ move.
    My son Joey (6) is deriving these logical reasoning, too. "Next time I need to take that rook with the bishop".
    What sounds funny is actually pretty good. Joe is lacking the ability to put down in words what he means. But if he reads his little guideline again, than he will remember the situation: the bishop had to do a backward move to take the rook behind him. If Joe reads his notice where a knight takes a queen with the move Nf3xQg1 that is a backward move, too. And even though his guide tells him about bishops and rooks, I could imagine that this "silly" clue would enable him to see the knight-backward-move, too. So it isn't silly at all, but only I was silly to think so. His notice is simply a rule that shall remind him not to always try to find at the front-line, but look back, too.

  2. @Munich,

    I dropped the "Me".
    The term "guidance" originated from this old post from Blue Devil.
    Semantics can be a bitch.

    I had of course a bit more time to think about the consequences. Imagine that you can only find patterns within the context you are looking! Had you ever come up with a skewer in this endgame position yourself, no matter how familiar it is? I didn't. You don't find it when you don't look for it.

    And how logical is it, once you get the hint!

    And how remarkable is it, that you can immediately visualize it once you get the hint!

  3. For the endgame puzzle - I did now this endgame, so I am disqualifed.
    But nevertheless it made me finally understand.

    I think endings often need context/guidance stuff.
    But not only in endings. In endings their existance is just more obvious.

    Since guidance and patterns often go hand in hand it is sometimes a bit confusing.
    But in general guidance is some sort of an element in strategy, while patterns are an element of tactic.
    Both should be possible to be learned.

    I will continue a while with my pattern training, but I think your guidance training is just as important. I will come back to it later, but need to continue my pattern training, otherwise I would not remember them.

    Does writing down during the the solving attempt not take too much time? You could start to put down in words after you had a look at your solution. I mean, you will see what you missed, no need to have written prove for yourself. After you failed the puzzle you can put the hint more precisely.
    Like my son "Next time I need to take that rook with a bishop".
    (of course I would advice you to find a more precise description).

    In that way you save time and be able to work on one hard puzzle about every 2 minutes. (~1 minute for trying to solve, and after failing another minute to write down a hint that you believe would have helped you to solve the puzzle).
    In that way you cover 20 puzzles per hour (actually 1+1 minute makes 30 puzzles per our, but lets be realistic: some puzzles take longer and we have little breaks in between).

    These difficult puzzles need repetition, too. But it is not easy to say when to start with reviewing them. Since you worked on each of them, they stick longer in your memory, and therefore the space between repetitions seem to be much wider.

    One thing for sure: all the discussed difficult puzzles stick in my mind now. The hints that would have helped me to solve the puzzle as well as the patterns.

  4. You are waaaay too concerned with the big numbers. Besides that you don't quite grasp the difference between the logical building blocks of tactics and tactical patterns. After a few hundred of difficult problems I have identified exact 40 logical building blocks. Guides. Little plans, if you like. Maybe I missed a few. But probably there are AT MAX 50 of such little logical tactical plans.

    Say that you have learned 25,000 tactical patterns during your chess life. Say that these patterns are spread evenly about the 50 logical building blocks. That is 500 patterns per block.

    When you are busy with one block, only those 500 patterns are retrievable within that very context!
    The remaining 24,500 patterns are blocked out by your very own mind. To prevent you from overload. So only 1/50 of the total sum of required patterns are avalable at any moment. Only those 500 patterns are exceeding the threshold of relevance within the current context.

    So what are you going to worry about. About the acquisition of pattern 25,001 or about applying those 50 building blocks well?

    I formulate it pretty black and white, but I hope you get the idea.

  5. Well, it could be that there are much less context guidelines to be found than patterns. (A Pattern is by my definition the whole idea of a tactic (such as the typical "smother", and not just a certain structure of 4x4 squares).

    The amount of patterns are just guestimates. Who did ever count them?
    After I did around 8000 differnt puzzles in CT (which is 20% of all CT puzzles) I would say that some patterns are quite repititive.
    I would say 8000 puzzles give me 2000 or 3000 puzzles. Assuming this is not only valid for the range 1150-1475 that equates than to 8000-12000 patterns that CT has to offer. But actually it should be less, because many higher rated puzzles consist of several easy rated patterns found in easy rated puzzles. (but I found out, that there are patterns on their own, which just represent a simple idea but is hardly known and thus results in ratings of 1700 or higher).
    Anyway, 8000-12000 is rather a number which is still too high. By making a guestimate I would say we shall better look in the range of 5000 patterns to be learned of which 3000-4000 are known by an average player. (missing the difficult puzzles is not so often due to not knowing the patterns as you found out yourself!).

    For the guidance context ("hints") - just alone your endgame with KQ vs KP gives me several guidelines.
    If I think about all the 4-5 stone endgames I know, I believe that they alone make for sure more than 100 sentences (guides) that are useful to solve them.
    As a guestimate (which is much harder for me to do since I havent started your method) I would guestimate that there are much more than 500 sentences I would need to know. I would not be astonished if they equate at the end to as much as 5000.

    Why? Because if I think of a teacher who asks me to do following home-work: "tell me as much typical tactics as you can" - I would come up with maybe 100 I could consiously name? One of them would be the typical smother with queen-sac and suffocating knight-check, another would be the shepards check mate, and a third a typical back rank mate.
    But if he would ask me about rules (guidelines) for endgames, I would also come up with 100 sentences.
    Just think of the easy K+P vs K endgames:
    1) a and h pawns will lead to draw if the weaker side manages to get to the promotion square.
    2) If the stronger side reaches the 3rd/6th rank while the pawn is not on the 3rd/6th rank it will always be won.
    3) If the weaker side manages to get into the square of the pawn (square rule) he can reach the promotional square (=stop the pawn if it has no guide). It does not matter if it is his turn that makes him reach the square or of it is the turn of the opponent - as long as he is in the square he can hold the pawn (unless the opponent's king is in the way).
    4) if you are the weaker side and need to stop the opponents pawn - once you are force going backwards by zugzwang: always move backward straight back. Dont move diagonal.

    These alone are 4 important guides I would say, but if you like you can argue the whole endgame is 1 guideline. Still - just 50 blocks would not be enough to even cover the end games?
    Are there so few guidelines in tactical situations?

    I am absolutely with you that the way you train makes sense and that you will find a lot of context/guidance in your quest. But just 50? making them very general - then 50 is enough. Actually I can make it so general I could make it 3 guides only: "Check-capture-threat"

    But that would not be of much help in the example of the endgame K+Q vs K+P.

  6. I just count the classes, you count al the different instances of a class.

  7. BTW, I consider endgames to be a class of its own. I'm just considering tactics here.