Monday, March 26, 2012

Loose ends















.
.
.
.
We have a method for plan A of which it is more or less proven that it works. Yet I feel that there are some loose ends to tidy up.

What is better, low rated problems in high quantities or high rated problems in low quantities.

Low rated problems in high quantities (aka "the saltmines").
What are arguments for this approach?

Pro:
  • A grandmaster has stored 50,000 - 100,000 chunks according to prof. de Groot et al. You can compare it to learning words of a foreign language.
  • The frequency of occurrence is high. Hence the relevance of the patterns.
Con:
  • The added intelligence per problem is extremely low.
  • It is a daunting and time consuming task.
  • No grandmaster ever did this.
  • I have been in the saltmines for five years and it never quite worked. I have assimilated tenthoussands of patterns, maybe not in the most efficient way, but nevertheless.
High rated problems in low quantities.

Pro:
  • DLM used low quantities of high quality.
  • Much intelligence added per problem.
Con:
  • You can do much fewer problems compared to lower rated problems.
Discussion.
Both approaches have one element in common: speed. You repeat the problems until you can solve them fast (<10 seconds).
The difference in time needed per problem is caused by the adding of intelligence, which is time consuming.
To me the main question is: how does the transfer work from one problem to another?
It is impossible to learn all possible combinations due to their sheer amount. If there is no transfer of skill from one problem to another, the saltmines are the only possibility. The very expression pattern recognition suggests otherwise. You assimilate just one pattern, but our ability to recognize that pattern everywhere makes that you can see the pattern in a new position, one that you have never seen before. The approach of DLM had two elements which were necessary for skill transfer: speed and adding intelligence. One of them he forgot to tell us about. To me, the patterns in 2300 rated problems and 1300 rated problems are the same. So there is no reason at all to work on a large number of patterns. It is the adding of intelligence that counts. In the process of adding intelligence, the transfer to new problems is assured. At the moment I'm experimenting with 1800 rated problems at high speed. That might be the best of both worlds. With medium rated problems you can do a set of 200 problems within a week very fast.

I don't think it is possible to settle the score by arguments. Only proof ackowledged by prof. Elo will be accepted.

20 comments:

James Stripes said...

Yesterday's Training Log discusses deliberate practice and offers links to relevant scholarly articles. Reflection must take place for learning, and speed does not always facilitate. In my own training, I employ both large quantities of relatively simple problems that must be solved fast, and small quantities of difficult problems that require time. Calculation must become accurate in long variations where many simple motifs might exist together.

Munich said...

James Stripes is right, but:
In my case I can play simultaneous blind folded games. Surely I drop in strength, but I know where my pieces stand.
So the calculation skill must become accurate, which I agree with. But there is a level reached where it is accurate enough. You hardly calculate deeper than 20 moves ahead. If you can do this, then in my humble opinion, the calculation training is not needed anymore. But for those who have still trouble with it, it might be a must.

At a certain level (which I reached), it is then most important to add intelligence. Puzzles I failed which I should not have failed, cause they are too easy. I am sure, one day I wont fail easy puzzles, but for the moment I still do. And those I train. Deliberate practice. I am glad to hear this confirmation from James Stripes.

Here my soft critic at temposchlucker: The added intelligence of a failed easy puzzle I consider more valuable than the added intelligence of a failed difficult puzzle.
The only trouble with failed easy puzzles is: you usually dont fail them. So you need to do a lot of them, just to find this trouble puzzles that are easy for others but not easy for you.

How long does finding easy trouble puzzles take?
If I do 1000 Puzzles at a range approx. 350 CT Blitz rating points below my all-time-peak, then I will fail 100 of them.

Lets calculate:
900 puzzles I solved in 9000 seconds (I assume 10 seconds on average. some a bit longer, some in 5 seconds).
These 9000 seconds are wasted.

I then repeat the found 100 trouble puzzles until I can do them very fast (within 10 seconds).
This will require about 100 x 10 repetitions x 20 seconds each = 20.000 seconds. (I assume that the average solving time of trouble puzzle is taking you longer, and it is pretty dry and pain staking and very frustrating to fail them over and over again or to need endless time for simple stuff. It makes me feel stupid from time to time. Anyway, that is why this training is referred as the "dry saltmines". It makes me sometimes jump that I miss a easy 1300 again.)

So the majority of the training time is still put on adding valuable intelligence.
This is the main error Tempo does here. He believes, the 9000 seconds is the major part of the training time. Well, it is not.

My calculation is of course simplified. Because there is not only "failed" and "very well known", but there is a transition between these two. I refer to them as "slow solved" puzzles.

I have always said, that there is a range where the balance of invested time is too bad. I believe you should not do puzzles that you can solve in 95% of the time. In my calculation that would mean 9.500 seconds are wasted, and 10.000 seconds are spend well.
Theoretically. But some trouble puzzles at a range you solved 95% correct might be due to careless lazy thinking (you get careless if it isnt challenging at all anymore), and 0.5% might be even due to mouse-slips.

A range where you fail 20% or 30% of the puzzles is also valuable. We need to keep an eye on the statistical relevance. Most tactics in chess seem to be between 1300 to 1500. You will find them most often in your own games. So it is not only a question what percentage you fail, but if the range 1300-1500 is not too easy for you. As long as you fail more than 10% of this range, then I recommend the 1300-1500 range. But as you become stronger, you might need to adjust to more difficult puzzles. Probably there is no "too difficult" (because failing is always "good"), but it might be detrimental for your motivation if you are a 1150 (like my 7 year old son) and try to solve the 1300-1500 range.

AoxomoxoA wondering said...

Only proof ackowledged by prof. Elo will be accepted.

In the saltmines you did gain 350? Elopoints and the Saltmine-Method was not that good / develped.

After the saltmines you did add inteligence in different ways and you lost 150? points

Pro Elo has spoken!

OTB-rating has many factors, Opening, Endgame, Concentration, Visualisation, Calculation and and and.
To measure the sucess of different forms of tactictraining by OTB-results is not easy, you have to keep all other factors stable. But if you are not doing any opening, endgame, Visualisation...-training anymore, then you will get worse at these factors. How can you tell the reason for any ratingchange?

I try to separate the effects and try to develop "instruments" to judge my progress. These instruments are programs to analyse my "History" at CT. But that might be not possible by any-one.

Still i think: A good method to judge progress in tactics is the CT-Blitz Rating and the best method is the Fide Estimate because many side effects are already compensated by richards statistical work.
If you think different then read again what prof. Elo said on top of this comment ;)

Munich said...

P.S. I dare to say, DLM "success" is a big cheat. and I dont believe DLM wasnt a 2000 before his training. I have a friend who is around 2100 Fide elo in strength but does not have a rating anywhere. If he starts deliberatly in easy tournaments below 1400 he would not only win prices - he also would start pretty low in rating, but would reach 2100 very fast. Like DLM did in reaching his 2000. Faster than any human before him. Faster than Magnus, faster than Kasparov. I doubt DLM, and so the "Pro" argument I consider as not true.

Temposchlucker said...

@James, I'm going to read your blog.

@Aox and Munich,
Clearly a repetition of moves. You both seem to have choosen (rabidly, if I may say so), I have not decided between the two methods yet. I consider all arguments rather weak. On both sides they can be easily flawed.

I recognize only one strong argument: statistical relevance. To decide if that is really an issue, we must decide on how relevant blunders of grandmasters in timetrouble (=base of CT)are for long OTB games of mere mortals. I don't know.

Munich said...

Well, for the difficulty of a range I am most willing to give in. I believe it might not matter so much, but is a question of optimizing.
What is really important though is: deliberate practice, repeating your fails (and probably also your slow solved puzzles).
And a good thing to safe on time is the principle "no DIY". It is not important if you came to the solution with a little help. All that is important is, that you understood the tactic and memorize it. Understanding helps in memorizing and understanding might even be important (but I believe even literally memorizing is better than nothing).
How you achieved this understanding is not important. You may look up the solution after 1 minute, and as long as you understand the tactic, you will be fine. So "no DIY" is important in terms of using your training time efficiently.

Using tag sorted sets will speed up things even more, the hint that it is a "fork" does not do any damage, but is just the principle of "no DIY" - you simple learn them faster because you solve them faster.

But the rating range? Failing is most important, but statistical relevance will put the pointer where you make the most of your training time. The puzzles you don't fail are a minority of the training part since you don't repeat them. And since you did them fast and only once, the sacrificed time in accidentally doing them is little.

No, I am not a fanatically insisting the easy range is a must. I rather believe it is the range that adds most intelligence (a failed easy puzzle is more relevant for you than a failed hard puzzle). But no doubt, difficult puzzles will add some intelligence, too. Probably not as much, but will work. The principle of "no DIY" and "deliberate practice" is more important than the right range, I guess. If now 1800 or 1400s puzzles - as long as you use tag sorted (no DIY) and spaced repetition sets (deliberate practice), I believe you will be fine as long as it adds intelligence.

Bright Knight said...

What you did previously worked exceptionally well. What you did afterwards took you backwards. People’s theories (including mine) are worth nothing in comparison with what actually worked for you. I suggest that you go back and repeat what you did before, as accurately as you can, using the same training material. You found the magic formula and threw it away!

James Stripes said...

Munich said...
P.S. I dare to say, DLM "success" is a big cheat. and I dont believe DLM wasnt a 2000 before his training.


I've spent some time examining DLM's rating history, noting

1) he started with a provisional rating.

2) His provisional rating in quick chess is ~300 Elo higher.

3) He played relatively few weak players.

Leading me to conclude that he was not as bad when he started as he makes out. If he was a rank beginner when he entered his first tournaments, the competition that he faced would have taught him some decent positional concepts at the outset.

Nonetheless, he did gain 400 points in 400 days after attaining an established rating, but I doubt he was married with children during that marathon sequence of chess tournaments and intense tactics training.

Munich said...

But DLM was a matured adult. And I dont know of anybody (children included) who made 400 points in 400 days.
I would expect that some high world class players (such as Magnus Carlson or Fabiano Caruana or Aronian Levon or Garry Kasparov, or Judith Polgar) have done a good training. Maybe they didnt really know what they did to become strong, but somehow instinctively they did the right thing. Of a million chess players, some are surely doing thing right (and better) than others.

But still: Nobody (except DLM) achieved 400 point in 400 days. If I am wrong, please show me an example that came at least close to 400 points. Of course I mean after they reached 1400, which is good enough to reach club level.
It might be possible to get from 1000 to 1400, because with 1000 you hardly know how to castle big or know the en passant rule.
But with 1400 you know most stuff about chess. After which even a gifted child with a good training and the will to work hard is not improving that strong.

So my conclusion: There is something doggy about DLMs success. Probably it is a cheat.

Furthermore, there were knights you did his recommended training, and they failed to show similar results. If I look at his training method I wonder how that was a good training anyway.
To hide your true strength is much more easy than the other way round: having a higher rating than you actually deserve. (though even that is possible, if you bribe your opponents).

Munich said...

P.S. and "That" is what DLM forgot to tell us: he was allready a 2000 before he started.

Temposchlucker said...

@Bright,
The first three years it worked. But after stalling for two years afterwards I abandoned it.

@Munich,
there is no real proof for what you say about DLM. It are assumptions.

Munich said...

just because it is written in a book it doesnt mean it is more likely to be true. This is a major mistake people do. I can show up with good arguments, but DLM says otherwise. I accuse him for cheating. If DLM would sew me, I might even lose, since I cant prove my claims.
I appeal at your common sense: nobody, not even a child, nor the best players in the world, came close to 400 points gain in just one year. Nobody. Show me just 1 person, that made 400 points like DLM did. You wont be able.
Wake up, DLM is a cheater. Its like believing into Santa Clause.

But lets put it the other way round: with my method I improved 800 points in one year.
So trust my method, I am telling you so.
I was a absolut beginner (proof: I have no rating and am not listed anywhere), my wife tought me how to move the chess pieces and now within 400 days I made 800 points. Well, I am going to make them, you will see.

And you wont have any proof that I was a capable player before I started my training at Chesstempo.

Temposchlucker said...

I don't believe anybody on his word. Not DLM, not you, not even myself. To me this changes nothing.

James Stripes said...

DLM played nearly every weekend, and he studied intensely. The playing is evident from his rating history. The studying is his claim. Thus, while an adult, he must have had money and time. No responsibilities. No children. No wife. No girlfriend. His quick rating shows that he was nearly 1700 when he began. You can believe him, but in order to reproduce his success, you must be able to devote yourself to chess full time.

Munich said...

Thanks temposchlucker, thats the right attitude here.

Just to make sure for other readers (since irony is much harder to grasp in the internet): I dont have a method that improves your skill by 800 points. In fact, "my" method (which is basically derived from Tempo, bright knight and aoxomoxoa) still isnt approved for sure. The increase in my CT Blitz rating is not high enough to be sure about it.

So I dont trust even myself, like temposchlucker doesnt trust himself, too.
I let myself guide by what makes sense to me. DLMs method doesnt make enough sense to me to believe him.

Temposchlucker said...

DLM was unemployed at that time.

Bright Knight said...

If you do not play very many games, odds against underperforming by 200 points or more in one season and over performing by 200 points or more in the next will not be that long, even assuming only random factors. 400 points in one year looks very convincing if you only look at the results of one player. However, there are millions of chess players with improvement programmes. Very many of them will chalk up rating improvements of 400 points or more in one year. If they then give up chess, they can write a book.

Munich said...

Show me just on of those "many". It never happened. 200 - yes. 300 we might find exceptions. 400? after 1400 I havent found any. Not even a youth. Nada.
Ah, besides: "de la Maza" sounds like latin american (spanish speaking people), and even though I like the caribbean (and am half of the year living there) and even though I like the the mentality of these people, I feel like I need to remind you that their attitude about telling the truth is usually a "bit" less strong then english native speakers (carefully formulated). I am not a racist, but people who deny cultural differences are blind. I know it is prejustice and bias, nevertheless people in these countrys like to ly just for the fun of it. And they dont take it hard on themselves if I call them "cheater". The attitude is more: "Even if it isnt the truth, it is a nice story, isnt it?"

That is o.k. with me, and knowing that cultural difference spares me from getting disappointed and I also am generously overlooking little fabrications. DLM can be "proud" that he got away with it, but it doesnt mean we have to "buy" it.

James Stripes said...

It took me nine years (1996-2005) to break out of C Class. I was in my mid-30s when I started playing USCF tournaments. My initial non-provisional rating was 1425. It took another 2 1/2 years to go from B Class to A, and nearly three to cross 1900 (last month). That's rapid chess improvement for the adult player that is realistic and not grounded in deceptive claims regarding initial skill level. If I make Master before I turn 60, I will expect you to buy the book!

Bright Knight said...

I will work out some numbers if I have time. No matter how long the odds, someone will eventually chalk up a 400 point improvement (without actually improving). MdlM’s initial results could have been depressed for a variety of personal reasons short of dishonesty. He probably did improve, but by much less than 400 points. He may have benefited from systematic as well as random errors in the rating system (e.g. geographical differences, rapidly improving juniors, unrated players). 400 points in a year is not enough though. You need to have no history and then give up.