Tuesday, December 15, 2015

The miracle of pattern recognition

Since prof. De Groot wrote about pattern recognition in chess, there have been a lot of people, including me, who tried to construct a database with patterns in memory. Especially for tactics. De Groot made an educated guess that such database might consist of about 100.000 patterns for a grandmaster. But his guess wasn't restricted to tactics alone. Those 100.000 patterns comprise openingspositions, positional play and endgames too.

The approach of the pattern-database-in-memory constructors always has been rather straight forward, and with little knowledge or understanding of the subject. The adagium always has seemed to be: the bigger the database, the better. But let's have a fresh look at it. What do we know about pattern recognition in dayly life?

If you look in the sky at a cloud, you might recognize an animal in it.
There are a gazillion ways that clouds can form a familiar animal. The miracle of pattern recognition by our brain, consist of the ability to recognize an animal in all those gazillion different types of clouds, with hardly any conscious exercise at all. The only thing we have to do, is to build a database of, say, 200 animals, and we can recognize them everywhere. Be it in a cloud, an inkspot, a smeer, a tree bark or whatever.

If you take this as an analogy for tactics, what are we doing when we are trying to construct a database in our memory with patterns by solving exercises? Basicly we are trying to construct a database in our mind with gazillion clouds which look alike an animal. That doesn't look very logical.

If I take the labels of  CT as starting point, there are about 30 different tactical motifs, and 28 different mates. An average problem like the diagram below that can be thrown at us by CT, consist of 5 tactical motifs which form a combination. 5 tactical motifs in one combination isn't extra ordinary at all. 30 different motifs can combine in a 5-combination tactic in 30 exp 5 = 24.3 million ways. There is no way we can learn so much positions by heart.

In accordance to the analogy of the animal in the sky, it suffices to learn only the 30 tactical motifs and the 28 mates. So what do we have to learn then?

Today I got this problem at CT. I remember this position from a few years ago. I even remember that the first move was 1.Nxd5.
But after considering 1.... Rxc2, I am soon overwhelmed by the position, since I have difficulty to keep account of the gained points on both sides.

What you see here, is that the remembrance of the position, and even of the first move, is of little help.

The question is, what must I learn from this position? I must find a way to simplify the position. I recognize the five tactical motifs:
  • Nxd5 double attack on Rc7 and Nf6
  • Nxf6+ double attack on Kg8 and Qd7
  • Pin of Rc6
  • Overloading of Rc6 for Rc7 and Nf6
  • Qa8+ double attack on Rc6 and Kg8
  • Skewer on Rc7 (update)

Rxc2 counterattack by black

The problem is to stitch these five tactical motifs together in the right order, so that the usual counter measures of black fail. Usual counter measures are: defend a piece, exchange an attacker, move the HE away, put a piece between attacker and defender, launch a counter attack et cetera.

As long as I haven't downgraded this position from a complex one to a simple one, I haven't learned anything from this position that might be useful for the future. I haven't learned anything from my mistakes.


  1. There are many HE's Nf6 and Bh6 are forkable by a pawn, h3 is unprotected, there is some weakness in the backrank..
    After detecting the tactical weaknesses you need to sort out the no realistic ones.

    Nxd5 is a natural first candidate move after seeing that a check at a8 and c8 is no good.
    There is a second reason why Nxd5 is "necessary": for tactics you need contacts and the knight at c3 blocks everything.. Nxd5 opens the position to enable tactics.
    Move the knight and the activity of all pieces explode, keep it at c3 and nothing goes.

    Funny that the good move is a centralisation of a minor piece again, a positional player will think about such a move at once.

    The ability to ubdate the material balance at every move is very important. White has +1
    at the begining:
    Nxd5 +2 Rxc2 -3
    Nxf6+ +0 Kxx -0
    Nxd7 +10 ...

    The chess visualisation course Book1 By Ian Anderson is a good book to do some exercises about this.

  2. I am not sure if the list of chess motifs you have just mentioned is completed.

    - Nxd5 double attack on Rc7 and Nf6
    - Nxf6+ double attack on Kg8 and Qd7
    - Pin of Rc6
    - Overloading of Rc6 for Rc7 and Nf6
    - Qa8+ double attack on Rc6 and Kg8
    - Skewer on Rc7 (update)

    Here are a few of mine:
    - WINNING a pawn (gaining a material)
    - X-RAY at rook c7 [it is the same as "Skewer on Rc7 (update)"]
    - double/triple (?) attack at Rc6 (directly), Rc7 (with an X-ray) and Nf6 (directly)

    Anyway the puzzle is very interesting. In my opinion it should be analysed in detail and after that we should EVALUATE all the final variations (who is ahead and how much).

    In my opinion this puzzle is MUCH harder than we originally think. It takes some hidden (surprising) moves and variations into the consideration.