Thursday, July 13, 2006

Inside the mind of Temposchlucker part II

Today finally my new personal chess trainingsstool arrived.
























The Blitz-O-Maza trainingsstool is manufactured by RatRace (TM) shock&punisystems Inc.
It works via a simple capital reward and -punishment system with which you will learn chess at lightning speed. An electrifying experience!
They guarantee that you never make the same mistake again. Or any mistake for that matter.
Gamma-burst inflatorcap is optional.



When it comes to training, what variables can we play with?

Category of problems.
The direction of our study can be determined by the choice of the problemset you work with.
There are different categories of problems possible:
  • Tactics
  • Positional
  • Endgame
  • Strategic
  • Opening
  • Medical (if you want to be able to treat your opponent)
The sky is the limit.
I can't think of a reason that when we find a system that works for tactical problems, it wouldn't work for positional patterns. After all, a move is a move, no matter if it adresses a tactical matter or a positional issue.

Quality of problems.
If you have a problemset with bad moves, you will learn wrong patterns.

Frequency on the board.
If you learn to solve problems that you will not face at the board within a lifetime it is spilled energy. For instance compositions fall in this category.

Depth of problems.
All complex positions consist of a lot of simple tactical structures. It is really important two learn the one-, two- and three-movers first. Lots of them.
It is my own experience that that helps you to solve complicated problems too.

Thematic approach.
We should test if a thematic approach to problems works best. Chess is a way too complex game for the human mind. Hence we simplify matters and isolate simple parts in problems. What masochistic tendency is that to artificial complicate matters again my randomizing the themes? Pattern recognition works via repetition of patterns and is stored in the procedural (implicit) memory. Thema's are stored in declaritive (explicit) memory. Who cares about the latter since that is of no use in a game? (see article previous post). Let's just experiment!

Size of problemset.
To ensure enough repetition the size of the problemset must not be too big. Less than 500

Total amount of problems.
Since a grandmaster has incorporated 50,000 to 100,000 patterns in his brain, 2 problems a day is not going to work miracles for you.

Amount of problems per hour.
This is connected with the move-depth of a problem.
You can do 80 simple problems in an hour (typical for CTS) or 2 complex problems in an hour (CT-art level 10?).

Speed.
The much critisized time controls at CTS are actually very good. When I started with CTS I thought it was impossible to have a look at everything within just 3 seconds. But patterns work so efficient that you can see it often in an instance if your plan is going to work. Just amazing!

[The End]

20 comments:

  1. Hey!

    It's ole Sparky! I saw him down at Big Mac here in McAliester Oklahoma!

    They don't use him anymorr though it's all Postassium Chloride these days. . .

    ReplyDelete
  2. I didn't know that chess is adventure sports. Forget bungee jumping and deltagliding. Soon insurance companies will ask new clients: do you play chess?

    ReplyDelete
  3. One comment about the time control at CTS. I have found that after training at CTS for long periods that my OTB game is worse in two ways: I lose my vision of the whole board and my ability to calculate in depth.

    I think I develop tunnel vision when solving CTS problems because when faced with a new problem it's more efficient to focus on the part of the board where the solution is most likely to be. However, if there's a rook on the other side of the board I will often completely miss it. I also get used to calculating a quick variation and not looking as deeply for defensive resources for my opponent because there simply isn't time to do so.

    I think CTS is an excellent test of tactical strength because if you really know a pattern you will recognize it within the short time allotted. I'm just not sure if it is the best way to develop that tactical strength. Right now, I'm going back and forth between CTS and CT-ART for my main tactical training, but I'm leaning more and more toward CT-ART as the best way to learn new tactical patterns.

    This doesn't seem to match your experience as you seem to have had great success using CTS. Have you had any of the problems I've described? And if so, how did you deal with them?

    ReplyDelete
  4. Hi Flounder,
    I'm pretty disciplined when making moves at CTS. That means I want to have a reasonable certainty that my move is good. That is, I don't want to gamble. I have a lot of tournament experience ofcourse, and the habits I developed there don't disappear at CTS. I think it is important to keep the success rate at CTS above 80%.

    My experience OTB is 180 degrees contra to yours, but I was always a terrible slow mover OTB. So I'm not used to make moves without calculating in depth. Thanks to CTS I can do that much faster now. But my caution always keeps firm grip on my play. I never play a move a tempo.

    ReplyDelete
  5. SanchoPawnza has commented on this a bit in his blog and you will find it in any good study recommendation: Your training must have balance.

    The short time limits of CTS are designed to give you high repitition of tactical patterns. The result is that these patterns are more easily recognized when you play a game. You will see them without having to look for them.

    Once developed, this automatic recognition will help you with your long in-depth calculations just as much as it would help you make quick moves in a blitz game. You may see the pattern 3, 4, 5 (or more) moves into your calculation. Visualizing the board several moves in will be significantly easier if the important tactical 'chunks' are stored in your memory rather than being calculated from scratch.

    There is of course a danger of treating a game situation like a CTS problem, but this is not the intention of the CTS training! If your training is balanced and you are disciplined at the board, the CTS training will augment your ability to recognize candidate moves, calculate deep lines, and calculate accurately.

    I have really enjoyed the many chess improvement blogs I have found here. There are so many good (even great!) ideas on what works, what doesn't work, and why.

    ReplyDelete
  6. Hi anonymous,
    5 years tactics, 5 years openings, 5 years positional play, 5 year endgames, 5 year memorizing grandmaster games, 5 year tournament play. If that isn't a balanced approach I don't know what is:)

    Here you can find the things I tried. http://temposchlucker.blogspot.com/2005/12/more-about-flawed-methods.html

    ReplyDelete
  7. I never thought I'd see a creepy electric chair picture on a blog from somone from the Netherlands! :)

    ReplyDelete
  8. Blue,
    I thought I put some American folklore in my blog:)

    ReplyDelete
  9. a comment on thematic vs. mixed problem sets: in the world of artificial neural nets, which are simplified computational models of biological neural nets, it's easily shown that mixed learning sets will result in better generalisation. that's because a thematic set will converge (ie. teach itself) towards the current theme and away from others. it's like brainwashing yourself. better generalisation means the net will perform better with test data (ie. new problems).

    of course brain is a hugely complicated system, and it might behave somewhat differently. but my guess would be that mixed sets will give you better results compared to thematic ones.

    ReplyDelete
  10. worm,
    that's a good argument I didn't know. thx.

    ReplyDelete
  11. I like how the Tasc Chess Tutor program does it. First, brief textual instruction and explanation (something a neural network could never understand: we can learn things by reading about them once; a neural network needs hundreds (or more) learning trials to get nudged in the right direction). Second, a few problems on that theme you just learned (to burn the idea in). Then, after a few such instructional modules, you get a bunch of problems without the theme.

    My bet is that for certain things, the above will work much better. For instance, in the endgame there are certain concepts it is just key to have, and much more efficient to learn them from the start than to work through thousands of examples (e.g., the key squares in K+P endgames, the pawn's square to see if it can promote).

    ReplyDelete
  12. I had a couple more thoughts.

    First, for those who already have spent years studying the word-heavy explanations, who understand all the major concepts, it seems practice practice practice is key. And perhaps a good chess coach going over your slow games with you.

    Second, it is interesting that ANNs (artificial neural networks, which have little to do with biology, as wormstar said, and are basically speculative psychological models, which is what we are discussing after all) are so horrible at learning to play chess! Even I could probably beat an ANN chess machine. And this, despite the fact that they are pattern recognition machines extraordinare. Perhaps it is the ability we have, that Tempo has discussed quite a bit, for a kind of meta-pattern recognition, the ability to see that multiple patterns we have learned are at play in a given position, and to combine them to generate a useful move. This would be analagous to more traditional symbolic, syntax-drive, AI. I wonder how many people have tried generating hybrid chess playing programs, combining traditional AI with ANNs.

    ReplyDelete
  13. Warning: the following comment is cultural and time period specific so all/some may have no idea what I am talking about.

    Tempo. this reminds me of the movie The Clockwork Orange where the Malachek Malcolm Mcdowell was getting reprogramed. I wonder if everytime you see a chessboard after your training you will yell in horror just like when he'd hear classical music. Bloody Horrorshow. 8)

    ReplyDelete
  14. so, tempo. you have now passed me on CTS. we are both close to getting above the 300 ranking! so, i'm curious how you are crushing everyone you play. do you think that it is simply the number of problems you have seen? or does your tactics rating seem to make a big difference? if that is so, then i just really need to work on my strategy or learn something from you. :) keep on chugging along at CTS!

    ReplyDelete
  15. Blue,
    you are quite right. First one have to acquire knowledge. Knowledge is so highly esteemed nowadays, that most people stop there. I made the same mistake. But then the knowledge has to be transformed to skill, by converting the "knowledge-particles" into concrete positions. Then these positions have to be trained as if it were a motor skill (maybe it is?).

    Most amateurs who are a lot of years around are filled to the eyeballs with knowledge, but they fail at the application of knowledge. It is widely recognized that the average level of play of correspondence games (high knowledge-low skills)is lower than OTB play by super grandmasters (high knowledge-high skills). Which means that time alone isn't enough compensation for the lack of skill. Hence your suggestion of a meta pattern recognition sounds reasonable.

    Now it is formulated, it is all very simple and logical.

    ReplyDelete
  16. Tak,
    sorry, I haven't seen the movie. But when I started with tactical training, I wasn't really charmed by it. I knew that I had to learn to love it though, to have a chance to prevail. Now, I'm loving it.

    ReplyDelete
  17. GK,
    yes, I even jacked up my alltime high with 13 points to 1577. My new acquired crushing skills are 100% CTS-based. I just have much more overview and I decompose a position faster to its tactical essentials.

    ReplyDelete
  18. tempo, I just wrote a bit about visualisation, memorization and master games on my blog. I'm extending things about tactics & memory to learning master games, and I thought it might interest you as well.

    ReplyDelete