Thursday, July 27, 2006

No conclusions yet

Just back from CTS. . .

Generalkaia asked me about my experiments at CTS. Since I haven't reached conclusions yet it actually is way too early to report about it. On the other hand it might inspire somebody else to start his own experiments. And I can use every help I can get.

Before I start I think it is appropriate to apologize for my crude way of thinking an calculating. It is my beloved method to measure two or three points of a graph, to draw a line thru these points and look to where it hits the sealing (or the floor).
At this way I stamp on subtleties and nuances of course, but it is a very fast way to find midway by its extremes.

Why these experiments in the first place?
For every ratingpoint improvement at CTS I have to solve 570 problems at average. 400 points in 400 days would mean 570 problems a day. This rises the question: can this be done more efficient?

These are the issues I'm experimenting with:
  • Repeating every problem 3-4 times.
  • Concentrate on the essential underlying patterns when repeating.
  • Visualizing the whole line with the eyes closed.
  • Speaking out loud what I think during the problemsolving.
  • Redo the session history afterwards.
OK time for some crude calculations:)
A grandmaster has stored 50,000 to 100,000 patterns in his system.
Let's take the most bad grandmaster (rating 2500): 50,000 patterns.
And let's take a person at the top of the bell curve: a rating of 1700.
I do the following assumption: a person with a mediocre rating (1700) has stored half the amount of patterns as a bad grandmaster.

rating #patterns
1700 25,000
2500 50,000

The amateur has to store another 25,000 in his LTM to increase his rating with 800 points to become a bad grandmaster. This means at average 31 patterns per ratingpoint.
I consider a problem, together with its solution and variants to be 1 pattern.

To increase from 1470 to 1600 (130 points) at CTS you need to store 4400 problems in your LTM. That means at average 34 patterns for 1 CTS-ratingpoint.

The database of CTS contains 24,000 problems.
17,000 of them are reachable to me over time. For the other 7,000 problems my rating is too high. Mastering those 17,000 patterns would give me an additional rating of 550. That's a point which we can measure overtime:)


  1. "I do the following assumption: a person with a mediocre rating (1700) has stored half the amount of patterns as a bad grandmaster."

    I bet that some of the patterns that the mediocre rated person stores are incorrect as well. I wonder given that many are taken from unsound games.
    I wonder what effects this has

  2. pictures now? himmmm? very good!

    i say no chessblogs without human photos of bleary eyed, or odd persons after training!

    we thus show the human comedy or divine comedy, either Balzakian or Danteian as prefered.

  3. Tak,
    that's an interesting point. Given the fact that you have to repeat patterns often at least 7 times before you have committed them into LTM "for the rest of your life" it looks logical that there is some sort of survival of the fittest. It's hard to believe that a non-working knight fork is committed to memory by doing it wrong 7 times. Some alarmbells will go off along the road. But for a positional pattern it's not unthinkable that you make the same errors over and over again, just by lack of (or wrong!)knowledge. So the quality of your problembase is all-important.

  4. this is a post i am going to reread a couple of times. i totally agree. this is the sort of chess thinking which i think is profound, and from which lasting insights can be derived.

    i think similarly, just in different aspects of chess--i.e. thinking to myself: "if i review 941 GM games THEN review the top, say, 341 of them, starting especially with the most elemental 62, and analyze them myself, in my head, no board, no annotations, THEN study the annotations, i will absorb a lot of chess." same thing, only you say it much better. thank you dirk.

    i read somewhere with 100% certain recollectin that as GM Psakhis [spelled this way in informator, but other renditions to english common, i think] was cutting his teeth as an aspiring master, that he ONLY studied some 150 or 190 carefully selected games from an informator, and "burned them into the circuitry of" his "brain", to again use Yasser Seirawans nice phrase.

    ive told him my plan to do this on those 341, and he thinks its a great idea. wormwood said the same, i think it was he, in recent weeks.

    to make not only the elements of tactics and tactics on the clock part of our "habitual ways of thinking", but elements of strategy, pawn structure, middlegame to endgame transitions, and of course endings.

    Yasser said in Russia, as is well known, they had some 2000 games or 300 endings that were considered to be part of any candidate master's itinerary or working kit of BASIC knowledge. so much of what we say here is about tactics, but so much else, but, again, as you tempo have amply writen of in the past. thank you.

    my tea is almost drunk, and so now back to work from lunch... the sun still shines here!

  5. Pattern recognition seems to be an equation with many variables. One of them is the number of patterns stored in LTM. Another one is the search patterns that make jump them from LTM into awareness. And yet another one is the ability to check if patterns are valid or not. In CTS it often happens that I see a pattern A (say a back rank mate) and try to exploit it with my move. Bad luck. Because within 3 seconds I missed a counter-pattern that avoids back rank mate. Instead, there is a pattern B (say a skewer of the queen, winning a knight). Had the problem before been a skewer, too, my mind (search pattern) would have been tuned to skewers, and I would have exploited it instantly, neglecting the back rank pattern.

  6. Tempo, I agree. Incorrect tactical moves against good play are punished quickly. They most likely won't stay in long term memory. I was thinking more in the lines of moves which aren't punished immediately especially in the opening or where one makes a move where a better positional move is available. for example, I am uncertain as to the best lines in the scotch as black. I have burned certain patterns of play in my memory that aren't very good.
    Also Transf.. Some of the games & position that are recommeded to memorize are available in the book GM-ram

  7. takchess: thanks for mentioning that. GM-Ram is a book that i like a lot. i have it right next to me. ive got in qued for the next year.

    first i wish to read seirawans winning chess endings. ive done pandolhpini, chernev's practical chess endings also.

    i do a LOT of CTS and crArt3.0, so im pretty well covered, for now. alburts chess training pocket book is simply something to carry around or sit in the bathtub with.

    thx, dk

  8. (This is an old post, I know.)

    I dont think the linear assumption of patterns is correct. If (if!) a "bad" GM has stored 24.000 patterns, that does not automatically mean a 1700 player has stored approx. half of them.
    It could be 500 only, but probably therefore the most common 500 patterns.
    The amount (volume) of patterns to be learned does not increase linear, but I believe the effort in getting higher and higher in rating is exponential. You get much quicker from 1200 to 1700 then you go from 1700 to 2200, and from 2200 to 2700 it is again a much higher "step" to tacle.
    The reason is, that the patterns you need to add into your LTM will become more complex, more weird, more seldom. The most common patterns you learn early. A 1200 player probably know maybe just 50 patterns, but therefore the most common patterns.

    greetings from User "Munich" (you find me on "chess tempo")