Sunday, April 08, 2012

Skill or pattern recognition

In a previous post I said that pattern recognition is the only way to transfer a solution from one problem to another. Maybe there is a second method: skill.

A skill is problem independent. The skill to see wether a piece is attacked for instance. If I improve this skill, it will have an effect on all positions, not just on the positions I used to train that skill.

To make matters more complicated, a skill can depend to a certain degree on the ability to recognize patterns or work in combination with it.

Aox said:
Saariluoma conducted a series of simple experiments which suggest that grandmasters are much quicker than novices in certain lower-level perceptual proc or esses. In the first of these experiments, a king of one colour was placed on the chessboard, along with a piece of the other colour. The subject had to state whether the king was in check or not. The average latencies were as follows: novices: 1550 ms, class players: 1250 ms, experts: 900 ms, grandmasters: 650 ms.21 The results show that skill is inversely proportional to reaction time.

This piece of text refers to a skill, not necessarily to pattern recognition. The skill to see whether a king is in check is transferable from one problem to another.

Aox said:
They argue that “the differences in chess skill manifest themselves in the speed with which successive new chunks are retrieved from longterm memory: 3 or 4 seconds for the Master, 6 or 8 seconds for the Class A player, and about 12 seconds for the beginner” (p.256). Therefore, not only does the stronger player have more chunks in LTM, he or she can retrieve them faster (Roring dissertation).

This points to pattern recognition (which might be just another skill).

The reason I bring this up is that when I'm busy with the fine art of seeing the invisible, it looks more and more like a skill and not like pattern recognition. Which is not the same as saying that pattern recognition plays no role in this skill.

CT is specialized in geometrical-pattern-of-the-pieces-recognition. As Aox pointed out, if you are disciplined enough during storage time, you can store the geometrical pattern of the squares too. And you can test it indirectly, if you are able to solve the problem quick, you prove to master the pattern of the squares. Well, he actually said "nothing new", but I think this is what he meant:)

But for a skill, an entirely different method for improvement might be needed. Something like the skilltraining in Fritz, for instance or something similar. What is needed for the skill to see the invisible topology of the board?

(the picture above, btw, is a shoe which plants a seed with every step)


  1. I think there is just a difference in complexity/quantity and not a difference of quality between tactical pattern recognition and these "skills". Tactical problems consits of tactical sub problems consist of tactical micro problems = board vision. A high speed low tactics training is almost the same as the board-vision-exercises at Fritz. A low-rating-mate-training is very close related to the fritz-board-vision-check-training.
    Board vision does help with allmost every position, sub-pattern with many pattern, the knowledge of a pattern with some positions...
    As easier, as more general as more (often) usefull.

  2. Nobody has clearly defined what they mean by a pattern - certainly not precisely enough to enable a computer program to be written to identify them. Nonetheless, surely one piece attacking another is a pattern. So is one piece forking two others. A pattern is something that someone who knows it can quickly recognise in a position.

  3. @Aox,
    This means that the transfer ratio is reciprocal to the complexity of the pattern. In order to maximize transfer we must minimize complexity.

    I doubt that.

  4. Bright knight, it could be worth though to program pattern recognition. And I am not even sure if there werent attempts towards this method. Maybe Shredder did?
    If I remember correctly, Shredder did "2 brains" looking at the position. One was of tactical nature calculating a lot of possible variations. The other was evaluating positional stuff, thinking how a possible endgame could look like (looking at the pawnstructure and thinking if the weakness in the pawnstructure seems to be of long lasting nature), and it might as well had some typical patterns stored in this respect.
    It was pretty fast in finding the idea to start a minority attack at the queens side, and back then Shredder became world nr. 1 amongst the chess programs with a clear lead ahead.
    However, meanwhile programs seem to be able to look so far ahead, that the positional thinking is not worth it anymore.

    For the Fritz attack training (board vision) I started around 10 attacks per minute, but could push it pretty fast (maybe within half an hour) to 20.
    Some hours later I even reached 47 attacks per minute. I think within 100 hours of training you might be able to push it certainly over 60 attacks per minute.
    I dont think it helped me much though, but then again it might have helped a little bit. And furthermore I did not invest more than 20 hours into this kind of training. Probably the 20 hours (if it was 20 hours) was spend well?!
    I am more aware of hidden pieces such as a black bishop on a7 attacking the white pawn on f2. Some puzzles in CT have solving times of a minute just because one of the main actors in a tactic is far away and hidden. In these kind of puzzles, my solution time is considerably faster than the average solving time, and this must result in a little CT rating increase. It also might have helped to avoid blunders in real games a bit.
    It could have sharpened my eye for judging when a position is tactical and when there is no tactic. If there are no captures possible, then the probability for tactics is very low.
    But I have doubts about the benefit after you reached 40 attacks per minute. Probably 30 attacks per minute is already enough. I have no argument why I actually believe that becoming even faster than 40 attacks wont help me to become any better.
    I might be wrong.

  5. @BK,
    I was afraid you were going to say that. It is difficult to give an example of a skill where pattern recognition plays no role. Yet I think it are two different things.

    Somehow it "feels" different.

    Allthough pattern recognition might be a skill, not every skill equals pattern recognition.

  6. Thempo said This means that the transfer ratio is reciprocal to the complexity of the pattern. In order to maximize transfer we must minimize complexity.
    That is what i think. A complex problem consist of sub pattern. If you are able to solve all subpattern (quicker) then you solve the problems (quicker).

    Thempo said I was afraid you were going to say that. It is difficult to give an example of a skill where pattern recognition plays no role. Yet I think it are two different things.

    Skills are highly automated "tasks". Board-vision is just very easy tactics. A check is just a very very easy pattern, to see all checks is board vision. To solve a tactic: "mate in one" is almost exact the same. Doing many easy (mate-)problems is automating skills wich are common in chess.

  7. Skill is a fuzzy concept. On the other hand, pattern recognition develops chess skill not only when these patterns comprise the physical relationship among pieces on the board, but also their dynamic relationships through what is less visible: where they can move. Stronger players recognize more of these dynamic patterns; that is, the GM instantly sees not only the bishop, but the diagonals along which it can move. Hence, a GM instantly recognizes whether that bishop is checking the king.