Sunday, March 06, 2011

Analysis
























.
.
.
At a pace of 2 hours a day it will take me about a year to do all exercises that form board vision and tactical vision. Besides that, I wil have to work on my analysis. As NM Dan Heisman put it:

Analysis is the process of creating a move tree; this includes identifying
candidate moves for both sides at all depths of the tree.

I will stick to this definition.
For two months I have played 20 minutes blindfold chess per day. Somehow it doesn't feel that it quite satifies the needs to learn to analyze. It sounds reasonable enough, "with blindfoldchess you will learn to keep track of the pieces without actually seeing them". I'm sure there are elements in blindfoldchess that are quite usefull for analysis. Yet there are other elements that aren't covered. Even in blindfold chess you have the problem that you have to analyze a position.

It is important to know more about this process of analysis. I haven't reached any definite conclusions yet, so let me first sum up a few things that come to mind.

Tactical vision and board vision.
Automating the scans that identify CCT and tactical opportunities will diminish the workload of STM during analysis and will thus be helpful.

Pattern recognition.
Pattern recognition is an ability we can't help we have. It simply works. In order to work you need a database with recorded patterns. When you have a pattern in your database, you will be able to recognize it everywhere. No matter how distorted the picture is. But no pattern in your database means no recognition. If you don't know what a giraffe is you can't recognize it in a cloud. No matter how clear the picture is. If you do have a record of a giraffe it is amazing how little you need to recognize it. Databases of patterns typically contain a few hundred records at maximum. In chess these patterns are common knowledge above a certain rating.

Chunks.
Chunks are LTM structures. You find them typically at the end of a branch of the tree of analysis. Heisman gives a quite good example of a chunk here (page 4). Exactly at the same time we both recognized the same chunk (a mating sequence). That gives you a precise idea how chunks are going to help you. There are about 50.000 to 100.000 chunks to be learned, if I'm not mistaken. Chunks can only help you sofar, though. Most of the tree of analysis consist not of chunks. Only the end of a branch might.

CCT.
Most of the calculation effort goes into the branches which don't consist of chunks. Your list of candidate moves should be guided by CCT. And well in the order of Checks, Captures and Threats. I have long underestimated this, but it is kind of logical, so I will give it a try. The board vision exercises hopefully adress this too.

Sequential and parallel thinking.
Automated tasks of different nature seem to be possible to be processed in a parallel way. Just as you can sing a song and ride a bike at the same time. It seems logical that automated tasks of the same nature can be processed only one by one though. Since equal type tasks are probably processed by the same brain area. Despite being logical I don't know if this is true. What I do know is that automated tasks enlighten the brain resources necessary for conscious thinking. So you have to automate every task you can.
Conscience tasks can only be processed in a sequential manner.

The problem with the tree of analysis is that it consists of branches and knots. Branches are suitable for sequential processing while knots are not. One way or another we have to find a trick to handle knots. Since our conscious brain can't handle parallel branches. I will have a look at pruning techniques in computer chess.

Calculation training.
At this moment I'm experimenting with the calculation training of Fritz. Somehow that feels more promising than blindfold chess. I download a tactical position with a rating of 2000+ from Chess Tempo into Fritz and start the calculation training. A tree of analysis typically is about 50 - 150 ply wide.

9 comments:

  1. Khmelnitsky suggest to do complex tactics and endgame puzzles to improve in calculation. With standard/tactic puzzles: "preverence difficulty hard" you can do that at CT.
    You can see then in your tactics statistics if you improve at Fork/double attack.
    The difference in your Performance there seems to be high:
    Simplification 2041
    - Overloading 1849
    thats ~200, mine is ~60

    ReplyDelete
  2. I've done about 10,000-15,000 complex taxtical puzzles and about 5,000 endgame exercises and found that that is not very helpful.

    About simplification and overloading: since I'm gonna adress all 23 tactical tags, I'm not worrying about the order too much. But maybe it's motivating to do the bad area's first. I will think about that.

    ReplyDelete
  3. It surprises me there are so few patterns and so many chunks.

    Of the tasks, It sounds like you are doing one task to death than moving on to a new different task. Do you envision revisiting each task at some point during the year to maintain sharpness?

    I'm curious have you ever found analysing GM games to be of any help?

    Good luck down the new road!

    ReplyDelete
  4. I am missing one important point in your analysis of "analysis": The Plan/Strategy. If you have a plan, and if you know the plan of your opponent your load of calculation is reduced drastically. Possible plans comes from the Opening ( Analysing Mastegames ), Pawnstructure and "Inbalances" = Positional Structure.

    There are a lot of different opinions about "Thinkingtechnics" in chess. Your Analysis reminds me of Thinking like a Grandmaster http://www.amazon.com/Think-Like-Grandmaster-Algebraic-Classics/dp/0713478853 . I try to "understand whats going on", and "see what chances might be there" , first. Alpha-beta Pruning and so on is for computers.

    ReplyDelete
  5. JIMj,
    there are 23 tactical patterns. There are a lot of combinations of patterns though. That's why these are called combinations I guess.

    Analyzing games is useful, yes. But I always used the word analysis in the common sense, not in the chess sense.

    ReplyDelete
  6. Uwe,
    the mind has a problem with calculation. Calculation is typically done until quiescence.

    Strategic plans tell you what to do in quiet situations. Then there is no need to calculate. And thus the mind will not be overloaded.

    ReplyDelete
  7. "Strategic plans tell you what to do in quiet situations. Then there is no need to calculate. And thus the mind will not be overloaded."

    You need to calculate with a plan too, but less.

    So you have calculationproblems in "not quiet" = Tactic situations?
    How do you know that you have these problems?
    If your standard rating at CT would be lower then your bltz-rating at CT that might be a hint. I think it would be more precise to compare the fide-elo estimates based on Standards with the ones based on blitz.

    ReplyDelete
  8. The Chess Exam told me so. Besides that, it's my hypothesis that everyone who is below master has a problem with calculating tactics.

    And indeed, according the definitions of Heisman, the problems are about tactics.

    ReplyDelete
  9. I think Khmelnitsky made an error in his first book with the evaluation of the calculation:
    Uri Blass did both available Chess exams in a short time. His calculation was weak in the first book, in the second (vs. Fischer ) not:
    http://chesstempo.com/chess-forum/training_diaries/uri_blass_trainings_diary-t1848.30.html

    My rating at #1 was: 1565 and in the book #2 vs Fischer: 2078. There was about 2 jears in between and several thousands Standard Tactics and Engamepuzzzels but +500? But if it would be true that my calculation did impoved that much then i recommend Standard and Endgame at CT! That was about the only training i did. Simply sit and calculate till you calculate everything till the end! No guessing allowed! No: "Looks good, should do it"!

    ReplyDelete