What is best way to memorize patterns chess

[vectorchessc]

I heard that it is important to know the patterns but what is the best and fastest way to learn these patterns?

[gohoos02]

Just search for problems that meet the criteria you want, e.g.,

a) in the tactical motif or mating pattern that you want to strengthen and
b) with the rating that you want to work on.  

Then create a custom set of problems and go over them.  

There are patterns at all levels.  Which level and which motif do you want to work on?  

[gohoos02]

One other thought:  You might want to check the "primary tag" option in your search.  That will increase the probability that the problems in the set will contain the patterns you're looking for.  Adding problems rated 2.5 stars should also improve the dataset.

[vectorchessc]

Ok, thank you

[piggypiggy]

If you use Chess Tempo, the first step is to understand the tags. Outside of the checkmate pattern tags, the patterns you want to learn are much more fine grained than the broad, high-level tactical motifs that are tagged on Ctempo.

There are lots more mating patterns than those listed as tags on CTempo. I argue with Richard about classifying checkmates by the pieces used in the checkmates. He thinks it is a ridiculous idea though I see it used elsewhere like in Chess Steps and Yusupov's series of books. Such a classification would give a tool for studying patterns similar to the tags for the main non-checkmate tactical motifs.

In general, you are forced to use high-level tags because there are far more patterns to learn than can be named. Names for specific patterns can take you only so far. The question arises: If the tags are high-level, not specific, then what good are they? Answer: If you limit your problems by motif then you will get more repeats of the same patterns. These repeats will be of the same pattern in different positions, which is helpful in learning the pattern. You will notice the similarities between different problems you have worked. Repeating a single example of a pattern over and over is not as beneficial as repeating the pattern through multiple examples.

The problem, of course, is to make the sets of examples repetitive enough. That is an art form in itself. Here are some ideas.

• If you select two motifs together you will get a higher dose of patterns that combine those two motifs than by working problems of the two motifs separately. I guess you could try three motifs together, but I have not done that. The point is that the patterns are (most) often combinations of motifs.
• Use the is and is not operators on the beta site to select sets of problems that are tagged with exactly one motif.
• Separate endgame, middlegame and opening tactics. You can do this by limiting the number of pieces still on the board and/or by limiting the number of moves played. Different patterns tend to occur most often in different parts of the game. You will get sets of problems with higher concentrations of examples by filtering on these under appreciated criteria.
• The most important non-checkmate tags are pin, skewer, fork, distraction and not overloading, overloading, capture defender, discovered attacks, and advanced pawn. They can keep you busy for a while. (Unfortunately, overloading is not tagged often enough. Note that distraction is deflection.)
• These tags have the most issues/difficulties: counting, sacrifice, unsound sacrifice, zwischenzug (unfortunately), exposed king (a given in the endgame, might be useful otherwise).
• These preparatory move tags are usually useful: attraction, blocking, interference. There are some kinds of preparatory moves not covered by any CTempo tag. I find people shoe horning them in with attraction or interference.
• Patterns will also vary in frequency with rating. Limit the rating range to change the patterns you see.

More ideas:

• Knowing the motif is a strong clue as to how to work a problem. You don't get those clues over the board handed to you directly. So, limiting yourself by motif has its drawbacks. You also want to work sets of problems where you have no clues ahead of time.
• Checking primary-tag-only will exclude a lot of patterns. I use a vote threshold of 5 or 6. None of these methods will give perfect matches, but having a mismatch every so often will keep you more honest. (See previous comment.)
• It is commonly advised to work on lower rated problems for pattern recognition. Dan Heisman recommended 1250 standard. That's a good start. You can go higher, say 1500. The vast majority of problems on CTempo are lower rated problems anyway. The tactics in games that really require high levels of skill just aren't as common. Therefore they are not as important.
• I have been using 2.25 stars as my lower limit on quality. You might even be able to go lower. Not sure about that.
• I do not use the spaced repetition feature. I don't know how to use the settings, but fortunately you don't have to use spaced repetition to benefit from the spacing effects and lag effects. Spaced repetition is not the name of the psychological phenomena (plural) that make that method work. To accomplish the same thing, I just use smaller, sorted problem sets. I can repeat problem sets as I wish which accomplishes the same goal as spaced repetition. Spaced repetition algorithms try to take advantage of the spacing effect in the most efficient way possible, but what is most efficient is not the same for everyone or for every type of learning situation. My approach to taking advantage of the spacing effect may not be the most efficient, but it is simple and a more complicated approach may not really be better.
• As IM David Pruess points out in a youtube video, the goal with learning patterns is to learn patterns, not solve problems. Ergo, to learn patterns, one should not waste time trying to solve every problem. You only want to learn the pattern, so don't be afraid to give up. However, see the next point.
• As one of the Chess Steps authors points out, learning patterns is not the same as learning to play chess. In particular, I think that focusing on pattern recognition can foster a bad habit of not thinking carefully. Also, thinking carefully is not the same as skill at calculation (working out sequences of moves in your head).
• Working out sequences of moves in your head is a narrow definition of calculation. Pattern recognition is part of the broader process of calculation just as thinking carefully is,

[richard]

There are lots more mating patterns than those listed as tags on CTempo. I argue with Richard about classifying checkmates by the pieces used in the checkmates. He thinks it is a ridiculous idea

I don't think it is ridiculous, I just think it is an idea that doesn't scale well to the total number of possibly separable mating patterns. The alternatives have problems too, and it might be worth investigating as a high level grouping, but still doesn't resolve the problem of separating what are likely 1000+ mating patterns when you start to look at things at a fine level of distinction.

[piggypiggy]

I suppose that nothing will resolve the issue of separating 1000+ mating patterns except learning to calculate efficiently with different combinations of pieces. Actually, I'm really not sure why separating patterns is that important in the first place if we can construct them at will. The latter is what calculation gives us.

I suspect that those calculations are based on a smaller set of patterns that give rise combinatorially to a large number of mating patterns. I believe that there are fewer patterns to learn if we focus on a lower level goal: coordination among the pieces at hand. There I think the patterns are so intuitive that we do not get sidetracked with this business of separating.

Grouping checkmate problems together by the pieces used is one way to focus on the more fundamental task. FWIW Chess Tutor uses piece trapping exercises (both to trap and to avoid getting trapped) as well as grouping checkmate problems together by the pieces used. Of course, trapping a piece is just a generalization of checkmate. I did a set of queen and bishop mate-in-twos last night.

[richard]

I suppose that nothing will resolve the issue of separating 1000+ mating patterns except learning to calculate efficiently with different combinations of pieces. Actually, I'm really not sure why separating patterns is that important in the first place if we can construct them at will. The latter is what calculation gives us.

The point of separating them is that the system can then tell you which ones you are still doing badly at recognising so you can target them directly.

[vectorchessc]

Thanks a lot @piggypiggy

[piggypiggy]

The point of separating them is that the system can then tell you which ones you are still doing badly at recognising so you can target them directly.

I know that is a favorite idea of yours, but I've never been convinced of the utility.

• It would seem that the best approach  to diagnosing weaknesses is to do a post mortem on every problem missed at the time it was tried. That is when one can best understand how one's thought process went wrong.
  

• If you have a large number of errors to diagnose (which might justify a spreadsheet approach), you may be attempting too many difficult problems and, as a result, are practicing mistakes rather than practicing tactics. The correct response would seem to be to change to an easier problem set and get a fresh start.

[richard]

It would seem that the best approach  to diagnosing weaknesses is to do a post mortem on every problem missed at the time it was tried. That is when one can best understand how one's thought process went wrong.

Doing a post mortem on each mistake, and asking the system to tell you which pattern types you are most often making mistakes on are not mutually exclusive ideas.

If you have a large number of errors to diagnose (which might justify a spreadsheet approach), you may be attempting too many difficult problems and, as a result, are practicing mistakes rather than practicing tactics. The correct response would seem to be to change to an easier problem set and get a fresh start.

If you are making very few mistakes you are likely wasting some of your time, especially if the goal is pattern training rather than calculation, as it means you are training patterns you largely already recognize. The point of being able to separate patterns with more precision is to focus on the patterns you don't know. The point of pattern based training is to learn new patterns not endlessly repeat patterns you already know well. There are other ways to do that besides tactical pattern separation, such as spaced repetition, but targeted separation seems to be quite an effective learning method, and also seems to be the preferred approach of many tactics books that often present puzzles grouped by pattern type, albeit at a very high level.

[piggypiggy]

Doing a post mortem on each mistake, and asking the system to tell you which pattern types you are most often making mistakes on are not mutually exclusive ideas.

I'm not saying they are mutually exclusive. I don't think the stats are all that meaningful.

The tags are noisy data. Tags may be a false positives. For instance, a tactical theme (not pattern) may occur on one variation of a problem, but the difficulty one had was with another variation without that theme. Or again, the cause of missing a problem may be that the tactical theme occurs several ply deep and we stopped our calculation too soon. Or again, the tactical theme involves a backward move (not usual attacking direction) or long moves by "remote" pieces. Or again, the tactical theme is superimposed with another theme and the difficulty was that we did not deal with the superimposed theme (themes are not necessarily independent).  Or we may not have gotten enough sleep that day, or been hungry, or been distracted, yada, yada, yada.

You have to have a lot of mistakes to overcome the noise in the data. Generating a lot of mistakes could come from taking a superficial approach to solving the problems (not consistent with OTB) which is practicing a bad habit. Maybe it comes from doing problems that are too difficult--which increases the chance of certain false positives because it would be more likely that different tactical themes are involved on different important variations, a Catch 22.

Ergo, I think that only one of these approaches is useful.

If you are making very few mistakes you are likely wasting some of your time,

You could be. But we forget patterns if we do not repeat learning them. The effects of spacing and lag make repetition an efficient way to learn and retain the info over a long time period (years). We forget more slowly. There also is the benefit of generalizing a pattern by solving problems with the same pattern in different positions. The simplest way to do that is to flip the board. Bottom line: you do need to repeat patterns.

seems to be the preferred approach of many tactics books

You want to do both, thematic and non-thematic. Books use both approaches.

The Chess Steps people comment that (more) recently they have tried to provide more problems of different types mixed together. Having the same type presented together gives a big clue for solving the problem which is not there OTB.

Books are not the best technology for learning patterns. That's why some people cut them up to make Anki cards out of them.  That still is not best because you cannot do something so simple as flip the board which is A GREAT FEATURE of Chess Tempo.

[richard]

The tags are noisy data. Tags may be a false positives. For instance, a tactical theme (not pattern) may occur on one variation of a problem, but the difficulty one had was with another variation without that theme. Or again, the cause of missing a problem may be that the tactical theme occurs several ply deep and we stopped our calculation too soon. Or again, the tactical theme involves a backward move (not usual attacking direction) or long moves by "remote" pieces. Or again, the tactical theme is superimposed with another theme and the difficulty was that we did not deal with the superimposed theme (themes are not necessarily independent).  Or we may not have gotten enough sleep that day, or been hungry, or been distracted, yada, yada, yada.

I don't find those arguments overly convincing. Yes, most real world data is noisy. We don't decide FIDE ratings are a waste of time because they reflect some games where we hungry, sleepy or distracted. We accept that our rating includes things that are not purely about our strength, but that across a wide number of games it does provide a good predictor of our strength. Similarly failing a problem involving pattern 'X' doesn't necessarily mean we don't know pattern 'X', however  if our performance on problems with pattern 'X' is well below our performance in other problems over a large number of attempts then it way well indicate we should spend some more effort on pattern 'X' than say a pattern which we seem to getting correct 100% of the time.

You could be. But we forget patterns if we do not repeat learning them. The effects of spacing and lag make repetition an efficient way to learn and retain the info over a long time period (years). We forget more slowly. There also is the benefit of generalizing a pattern by solving problems with the same pattern in different positions. The simplest way to do that is to flip the board. Bottom line: you do need to repeat patterns.

Yes, spaced repetition is great for approaching both the re-inforcement and initial learning within the one system. This doesn't mean focused targeting on specific weaknesses isn't also useful.

You want to do both, thematic and non-thematic. Books use both approaches.

The Chess Steps people comment that (more) recently they have tried to provide more problems of different types mixed together. Having the same type presented together gives a big clue for solving the problem which is not there OTB.

Obviously, but sometimes learning optimisation isn't all about making the process as close to OTB as possible. If it was, we'd just play games to improve and ignore everything else.

Books are not the best technology for learning patterns. That's why some people cut them up to make Anki cards out of them.  That still is not best because you cannot do something so simple as flip the board which is A GREAT FEATURE of Chess Tempo.

Yes, books put the onus on you to decide what you want to study, they can guide thematic practice by sectioning their book like that, but do nothing in terms of trying to optimise your learning such that you are spending the minimal time required on retention to retain, and not wasting retaining time on doing problems you are in no danger of forgetting, and already know well.  I'm also suspicious on how well balanced the patterns presented are in terms of matching real world frequency. There is often a massive bias in books for 'cool' positions, and I think it is tough for an author to produce a collection that is highly representative of the sample you'd see when playing games.

If you believe in the idea that there are a finite number of patterns covering the vast majority of positions, and that learning all these is the difference between master and non-master level play (which is a popular opinion amongst researchers), then a perfect book would be a dictionary of all those positions that you could set about learning, and once done, you've got master level pattern ability (even if you don't have master level calculation). Part of the reason this doesn't happen is that book authors are happy with a very high level of thematic distinction of Fork/Pin/Discovered attack etc, and would struggle to identify more than a small fraction of the required dictionary if asked to name all the patterns they knew. Mostly it is probably not explicit knowledge, so even if they wanted to, they'd be hard pressed to respond to a "describe all the patterns you know".  If a language was found to start to describe these likely thousands of patterns then we could move closer to a point where you have a checklist of patterns to learn and can start to set about learning them. Right now we hope we see enough patterns during random training (or more targeted , but still very broard thematic training) to absorb a high percentage of the required patterns, and spaced repetition is indeed a handy tool to help with that, but it seems like a more surgical approach should be possible.

[piggypiggy]

part a:

You are comparing: a) assigning numerical ratings to chess players with b) assigning sets of tactical problems to chess players. That is, you are comparing: a) apples and b) oranges.

You are ignoring the variety of the sources of noise I mentioned which can negate the usefulness of data needed for your favored approach, e.g., when a tactical theme occurs in a variation that did not figure into why someone gave a wrong answer. The yada, yada, yada includes opening vs. middle game vs. end game, defender's vs. attacker's POV, knights vs. other pieces, depth of calculation required—things that Ctempo data does not cover.

You are also discounting the usefulness of a straightforward, robust method of learning what kinds of problems one needs to work on most: an immediate post mortem on any problem one gets wrong. If someone always did that, then even a successful statistical analysis would reveal nothing not already known.

We do not need a classification system for feeding data into some statistical analysis when we can already identify what needs work by always looking at what went wrong when we miss a problem.

[piggypiggy]

part b:

In short, the arguments given do nothing to diminish the value of grouping checkmate problems by the pieces used in delivering checkmate. (That's how we got here.) Some benefits I see:

• Grouping checkmate problems by pieces used would be devoid of the bias you see in books of tactical problems. (I think that bias is real and present in many books. Neishtadt even says that this bias is a conscious decision for him.)
• This method of classifying checkmates would be comprehensive, not incomplete like the various tag definitions currently extant in chess literature.
• Come to think of it, grouping by the pieces used to deliver checkmate would group similar patterns more closely together. One probably can learn a pattern better by comparing it to similar patterns. Similarities among patterns in a set of problems would help to build the mental schemata needed for organizing the information.

At present there is no comprehensive classification of checkmate patterns that identifies similarities as well as differences. Also, there is no prospect of having such a classification. The only idea I can see that would move us in that direction is to start by classifying checkmates by pieces used and studying the patterns we find grouped together by that method. I know of no other way forward.