Friday, February 17, 2012

Aaaarghh!!! Not those 7 circles again!


















.
.
.
.
I'm doing high rated problems at CT in sets of 50 at the time.
I'm trying to work out why I keep missing them and try to take measures to prevent to miss them in the future. Of course I want to test that so I do the circles in a slow way a second time after a while.
Since I missed a lot the second time too, it is not unlogical that at a certain moment I want to test the same problemset a third time in slow fashion. Since even the third time I miss a few I know I will have to do them a fourth time slow to get them all right.

And if I have them all right, it is logical I want to ingrain the patterns and speed them up. What do you think, would doing them 3 times at full speed be enough? I hope so.

But look what I have reinvented here: those darned 7 circles again!
Only with one little difference!

At the moment I started a new series of 50. I want to know if my method has any effect when I encounter new problems. Since I scored the old problems at 0% during the first circle, any progress will be substantial. I will keep you posted.

9 comments:

  1. I try to do something ~~like this too, i use spaced repetition ( instead 7 circles ). If you double the space in between the repetitons each time, then the workload for these repetitions is not that high. ( Depending how you start these repetitions you just repeat (for ex.) twice as much problems as you find new ones ). If you cant solve one of these problems again in the future, then you found a "new" problem you should repeat. That saves time too. If you could solve it ( in the right way ) it did take that much time and you kept your "skill" ( to be able to solve this problem ).



    To make an error onve is ok, but if you do it twice...

    ReplyDelete
  2. Somehow knowledge in itself doesn't seem to be good enough. The knowledge has to be transformed into patterns, before it becomes usable.

    The question is, does the training have an effect on new questions too? I'm about to find out.

    ReplyDelete
  3. It had with me!
    The CT Fide estimate based on CT Blitzrating is calculated on first-timers only. So I can clearly measure how well I am doing in unknown puzzles I see the very first time.
    My CT Blitz rating with the user Munich is now 1900*. My CT Fide estimate is now at 1990.

    *And due to the other proof I have given you by email (shhh - this is a secret!), we can assume, that 1900 wont be the temporary limit, but it could well be 20 points higher, and I guess the Fide estimate would raise then by further 20 Points, too.

    Your last post - you dont mean that you repeat all 50 puzzles in a set, do you? Because the ones you were able to solve reasonably fast the last time you have seen them - well you dont need to repeat them. Maybe only sometimes, just to check if you can still do it fast.
    I used to have the system like this. First time I solved the whole set. Then in the 2nd round I repeated only the puzzles I missed or I did too slow. And in the 3rd round I repeated only the puzzles I still did wrong or too slow.
    Only the 4th round I started to do them all again. Maybe I did an extra round for all the blunders I did, before I started the 4th round.

    But since Richard has now given the option to do a spaced-repetition set (it is in beta testing, so you need to ask richard to give to you), I will not keep track anymore when to repeat the puzzles, but simply trust Richard that he wrote a program well enough to deliver me more often blunders, not so often slow solved puzzles, and hardly puzzles I solved reasonably fast.
    My impression so far is, that richard did find a very good cycle algorythm.
    It seems to take into account how often I blundered the puzzle, how slow/fast I solved it the last time I solved it (and for the time it probably looks how many moves the puzzle has), and how often I have already seen it so far altogether.

    ReplyDelete
  4. I always (well, lately:) said that there are not so much patterns to be learned. A few hundred, at most. But once you have acquired those patterns you will be able to recognize them in millions of positions.

    If only two problemsets of 50 already have a significant effect on how well I score the third set of 50, I consider the statement above to be proven.

    ReplyDelete
  5. I know what you mean that there are only a few hundred patterns to be learned.
    However, now the "but":
    ...but isnt it in reading words the same?
    "-ing", "-en", "-an", "-og", "-ig", "-ave", "-it", "-at",....

    We (mum and dad) trained our son to be aware of these letter combinations.
    Nevertheless, Joey needed a lot of repetitions of whole example words that contain these patterns.

    If you for example play a black opening with the white pieces, the opening should be familiar to you. Nevertheless, you can feel the difficulty to have the picture of the board position mirrored. While as a black player you are used to have your casteled king on the LEFT side, with a white king it is different. The short castled king is on the RIGHT side.
    It makes a difference in blitz games (resulting in a little delay), even though the principal patterns you should know very well.

    Well, yes, "only" a few hundred patterns, "but" nevertheless you need to train them in different places all over the board several times.
    So even if you had a super well accomplished set of tactic puzzles, where each pattern only represented once and this set might possibly be only consist of 500 puzzles to be learned - even then it would probably not sufficient to learn just these 500 puzzles by heart. I agree, that a well chosen set would be more beneficial than a random set. But I dont think it would be sufficient on its own just to learn this "super set".
    Joey needed lots of repetitions: "set", "get", "let", "pet", "jet", "wet"

    and
    "ring", "sing", "thing", "King", "wing"

    ...and still he would have had trouble with words like "b-et-t-ing", "get-t-ing".

    I am afraid you need to learn 10 times more puzzles than there are patterns, to automize a good recognition of these patterns.

    Fotunately you are allready a capable player, who knows a lot of patterns inside out. So you only need to concentrate on what you dont know well: puzzles you failed or did too slow.
    The main question we havent solved is: what range? If your range is too easy, you hardly will fail them. If your range is too high, it slows you down and hence you dont have thousands of different examples to train what you like to automatize.
    And too difficult puzzles are a bit lacking statistical relevance. Most puzzles are in the CT Blitz range of 1350-1550 AND where failed by players where half of them had a fide rating of at least 2200 elo!
    My own problem is, that this range is a bit too easy for me (I hardly fail them), so the next step would be to move a bit up: 1550-1750 or so. Still reasonably relevant like "high frequency words" in reading, but not the most common patterns anymore. (so not the most basic words like "cat", "the", "it", "have", "this"..., but rather words like "threat" versus "thread", "once" versus "ones", so words we actually do now but often mix them up writing them wrong in blogs and forums all the time. Actually we do know how they are correctly spelled. But I guess a bit training on most mistakes we write in blogs and forums, and we would not consitently do them anymore.)

    ReplyDelete
  6. @Munich,
    Joey needed lots of repetitions: "set", "get", "let", "pet", "jet", "wet"

    and
    "ring", "sing", "thing", "King", "wing"

    ...and still he would have had trouble with words like "b-et-t-ing", "get-t-ing".

    I claim that the lower (<1550) rated problems are the "ring","sing" and "thing"-s.
    It is time to form sentences with those syllables!

    How strong is a blundering grandmaster? Given the frequency of blundering he plays like a 1500 player. That's why those problems are rated 1500 low. What are you learning by doing them? You learn to avoid to make blunders that rate you 1500. But that doesn't make you >2200!

    Trust that you already know all relevant syllables. Start to make words. Or even better, sentences. Narratives.

    Something totally new is added to the equation when you start to make sentences. Something you cannot derive from learning syllables. The frequency of syllables is totally irrelevant.

    ReplyDelete
  7. They are rated 1500 because we have the guidance "be aware - here is a winning tactic for 100% sure!"
    This hint the high rated players dont have during their games.
    Since about approx. every second game contains a blunder, and most blunders are to found at 1500 --> you will beat a master every second game.
    Well, only if you spot these tactics in your games, of course!

    But somehow we dont. And I doubt I ever will be able to. Even Magnus does frequently blunder.
    A master on the other hand has the skill to solve high rated CT Puzzles with the hint "here is a winning tactic for sure".

    We have no trouble in reading more difficult words. Contesters of speed reading train on reading words that cause them trouble. And the "trouble words" they are looking for are high frequency words. Then again, most high frequency words they can already read with super speed.
    They have the same trouble we have in chess: on one hand, the most frequent words are too easy and are not worth to be trained, since you cant expect any progress in speed reading if you train words you already know well. So they search for "trouble words" are words, that are worth to be trained. Words which have good chances to come up occasionally AND cause the speed reader trouble.

    You are right with "ring", "sing" are in the category < 1500. They dont make much sence to be trained anymore.
    But on the other scale, words such as "desoxyrybonuclein acid" are in the category > 2100.
    So a speed reader is looking for words that are maybe 6-8 letters long, are a bit more seldom, they cause them trouble, but are worth to be trained, since they appear every now and again in a text.

    We never looked at the middle range to be trained. I guess we should though, because here we have the best balance of statistical relevant patterns and finding puzzles we fail.

    I am not sure where "my" middle range is (I am at 1900). It probably does not matter so much, as long as I am not totally off the road. So I should not look at puzzles too difficult for me (>2100 CT Blitzrating), and also not at puzzles too easy for me(<1300).

    Here I have two puzzles as an example, please take the time and have a look. I promise it is worth it!

    If you know this one very fast:
    http://chesstempo.com/chess-problems/41450

    Will you be able to solve this one?
    http://chesstempo.com/chess-problems/74502

    ReplyDelete
  8. Those examples of you stemm from the same game. You won't find that in real life.

    But on the other scale, words such as "desoxyrybonuclein acid" are in the category > 2100.

    No. Those artificial words you can compare with composed chess problems. Those are both artificial and complicated. I will make a post about why the >2200 problems are sentences and not artificial words.

    ReplyDelete
  9. Oh, dear tempo. Of course it is!
    You missed the point! You cant tell the wood from the trees.

    O.k. let me help you.

    The CT Standard rating difference is 700 points. The time consumption is a lot more just for the two extra moves.
    What was the problem for people when they saw the difficult brother?

    Was it they did not see they could take the queen on a1?
    Well, I bet most people saw this pretty soon. They also saw: but if I take the queen on a1 then he takes my queen on d6.
    Hm. Can I give a check with my queen, so I can get it out of danger?
    No - I cant.

    O.k. lets play that through.
    I take his queen, he takes my queen.
    1...NxQa1
    2.NxQd6

    Hm. What then? Can my knight eat s.th. from a1? No.
    What about the check with my rook -->
    2...Rb1+

    And now, white is mated? No, he is not. He can play
    3.Bf1

    Hm. And then? (Now he is thinking about 5-10 seconds) I cant see anything promising.

    Lets go back to the start....

    - - - - - -

    That is how most people will think. They had the winning line found pretty soon I bet. But they failed to see that they win after 3.Bf1.
    Which is exact the position we have with the easy puzzle. 700 points rated below! Almost there! It was so easy. How can you not see that after 3.Bf1 you are not winning?
    The reason is: 5-10 seconds was not enough for strong people to see that after 3.Bf1 black is winning.
    Only very strong people saw in their search tree, that after 3.Bf1 their is somehow a pattern bell ringing.

    So, this is aox and my main point. If you cant solve easy stuff quickly, you will stop in your search tree after some seconds. I bet most people came so far. But it took 700 rating points in difference to make it within the search tree in your head to the solution.
    If you dont see it quickly --> chances are you are not seeing it at all.
    Or only after you repeadetly came back to the position 3.Bf1, only then you will evtually find the solution.

    Most people can solve the easy puzzle. But they fail if they have a look at it for just some seconds. Those who can solve the easy one pretty fast - those will find the solution.
    A player who is 400 rating points stronger will require only half the time. In CT Blitz rating, the rating difference is 500 points. So we can safely assume, that a 1850 rated player can solve the easy puzzle in 12.5 seconds on average (25 sec/2).
    This is just about fast enough to tacle the puzzle at ply 3 in your search tree. 1 out of 2 1800++ player manage in CT-Blitz-mode. Only those who can do it below 12 seconds.

    It is not enough to have hear of a pattern. It needs to be automated within you. Otherwise the breadth/width of your search tree suffers. at ply 3 a lot of 1800 rated player have the strength of 1100 only. And dont tell me, they did not try in their mind the variation "I take the queen, he takes my queen - and what then. Any checks?"

    Almost everybody came to Bf1. No doubt about that. Or have you doubts?

    ReplyDelete