I love this idea, and it would make a great book if you could compare to some other methods. For example, I really like Barsky's Modern Guide to checkmating patterns. Despite his extensive cataloging--he's identified so much--but I can't seem to connect the dots and see these patterns.
I don't know if the problem is my pattern recognition or if my calculation strategy when I look at these puzzles is utterly flawed. I probably need better habits.
thats exactly what interests our type of chess ... are the intricacies, geometric patterns , mathematics and shear magic of chess.. so yes absolutely... a book would be great (i now easier said than done) but if possible.. ill be the first to buy it ..:)
I think you're on to something here! But I do think it still needs some refinement. For example, 1058, 1059, and 1060 are indeed almost all the same thing -- a very well known mating pattern of the knight and bishop against a fianchettoed pawn structure in front of the king. OTOH, 1057 seems very different (starting with the fact that it's two knights, which necessarily makes the geometry pretty different)
I would add that "attacker sum" doesn't necessarily express what you want. After all, in a bank rank mate, who cares if it's a Q or a R? For the cluster just described, who cares if the long diagonal is controlled by a Q or a B? (In other words, in many instances, the Q&B or Q&R are interchangeable as they serve the exact same function). OTOH, a knight is fairly unique, which is exactly why 1057 is so different than 1058 thru 1060.
Thank you for the constructive feedback that pointed to some things I have been working to improve! I have found a better model. Will post more details later this week :)
For the attacker_sum why did you add the different values for the piece type? Since each piece attacks a square only once, the attacking power should just be the number of pieces attacking the square, right?
Or if you want to identify which pieces attack the square, it might be better to use a value that is unique to the piece combination attacking the square.
Thanks! My reasoning was that I wanted to measure when different pieces controlled the same squares. A slight error is that the addition might give the sum of another piece, so I think it can be improved
A simple way to make the number unique would be to assign a prime number to each piece type, raise it to the number of times this piece type attacks the square and multiply them together.
For example, if two knights and a rook attack a square, the value would be 3^2*7
Martin, I'm curious if you compared your fingerprint with the lichess problems database which includes the tag for a number of mate patters, so we can know if it matches the current fingerprints or not. I have been working in various kinds of fingerprints (mainly opening related) and this one seems to be very promising for tactics!
also publish the source data intermediates in the whole process before book digestion. ways to get to position sets.. so we can also work our own "magics" without having to reverse engineer or reinvent too much. This is just preventative, as I noticed chess and database transparency not to be high value on the market.. This is an exciting direction.
books are lacking information. they are not lossless. Maybe I don't understand what you mean by book. Also, not yet read, but are you making a difference between static pattern definitions and dynamic patterns. I found named chess mates come in such different flavors of chess culture naming decision.. Not all dynamics paths get named, and not all static mates either.. this is about those that humans find nameable.. (it implies their internal representation of their own thinking about chess, and what is worth naming).
So to make sense of books in your survey, I wonder about why a book not a database of position. Because likely, it might be about paths to "mates". What is a "mate"? The mate in 0... or the move sequence to that. And also, I should say that static patterns are also probably cognitively causal to dynamic patterns.. I will read now.. invested by daring to express myself here..
I would take the puzzles at their starting point, eg mate in 4 or mate in two. And It is correct that making a puzzle book with the checkmates would be rather boring 😄 and of course part of the book would be a technical explanation of how the puzzles were clustered together and maybe together with the code, so others can play with it too
so this is the rosetta stone. the high dimension from number of board plane points or squares (empty in foresight context or not really necessarily yet having the mating conditions), the king field including the king at the center.
Jomega also had such variables.. the 3 interprettable. But perhaps not verbally aligned on the pin vocabulary more the attacker potential covering. There is also a notion of support when attacker itself is occupant.. and the blocking of same color.
And one can choose different aligment of interpretables (or visible on board) but it seems at that level you do have the exact information for calling a static mate... I would need to check the definitions of the words... and compare with jomega study (because I understand those to be complete).
I would be happy to help, but I can't promise productivity.. minimally sound boarding. and very curious about the negative results as, this work you did is what I would have liked to do one day... exploratory data analysis. I willl contact you. The negative result would be in that common curiosity I am happy to see shared, would feed my understanding.
but quick question from second pass reading. The tool that does clustering, its input "vocabulary" set, beside the pure exact mate information as above, does it accept a certain amount of transformation control. I can't say I would be technology savvy. It would be more about the math. of this toward board visible quantities. Than which is the hottest algorithm. And I would not like for you to have high expectations of results. I just have a basis of understanding of such things, at the math level which I might be able to make understandable in context beyond the math symbols.
I think you might anyway have a single book topic there. But books dont talk back when read. I guess a monograph in my previous culture would come have streaks of debatable articles that might have been discussed pre-publicatino through peer review but also in between more than one author clutch (or group or school of thought yes, even in well measurable science) tit for tat askinjg questions that the previous paper did not make explicit in their own struggle to thread the not yet known.
If you can digest the other books that this comment section is bringing to our awareness I would be glad to read about it. I also find that patterns in general in chess have not been systematically investigated, and mostly taken as black box, that can only be communicated through examples.
So, I am familiar (or was) with theory of various clustering approaches, the question, when tinkering with those, becomes the relation between metric choices in that big space and the clustering portrait. notions of boundaries between "attractors" , cluster membership interpretation etc.
how many techniques and choices are behind the result.. I think in your book it might be interestijng to have discussion starters about the many possible clustering angle choices... so we can start lookihg at the human chess board story values of such clustering. And perhaps even consider theories of learning that humans can see in relation to board visible clues at their experience levels.. things like that. A book is a nice thing, but that is a one or 2 persons creation, and sometimes such beautifuul topics like this one, are meant to be discussed at the many persons level allowing for errors and creativity that one human alone in the end, might not be able to explore or see, in that we do all have limitations, the best of us. I guess I would call that scientific culture approach. But a book might be a good conversation starter.. of course.
I agree that there is a human element in it with making all the choices. Made several versions that didn’t manage to cluster similar patterns. If you want to collaborate on this please send me an email: saychess1@gmail.com
I’m not trained in this field, so you most likely have a better understanding than me, so would like to hear your thoughts
Ok so the kill box pattern seems to have all three possibilities.. static pattern trigger, dynamic (combo) pattern, and possibly also "idiosyncratic" board configuration mate static pattern. I am throwing synonyms as emphasis of what I mean.. can you translate?
can you share the code? Jomega on lichess has done some work on automatic pattern detection in math compatible with coding further work, not sure the studies are public or private.. But it is meant to.. I am curious about your "fingerprint" defintions..
So using the kind donut region is common to both your work.. The king field (dotted or not).
it would be nice that mathematically intelligible version of such things be shared.
Having not read the book but heard about it. I heard that Bobby Fischer teaches chess spends a lot of time drilling into a player the ideas about back rank mate. With some of your previous books. I feel that you've presented a lot of useful exercises but haven't quite hit that itch of connecting exercises to the chess world.
So my suggestion would be this: you've got 74 patterns, pick 25.
Of those 25 give a page of 6 example puzzles for each. Then find a game for each of the 25 patterns with a relatively high quality player, and show that full game for that player. This will connect the pattern to an opening, a story, and a method to get to the place that we have just studied.
Of course that's a lot of work, but it is discrete chunks that can be done in a checklist style.
You're really good at finding interesting chess statistics, and positions and this seems like the type of book that would stand out versus other puzzle style books.
I love this idea, and it would make a great book if you could compare to some other methods. For example, I really like Barsky's Modern Guide to checkmating patterns. Despite his extensive cataloging--he's identified so much--but I can't seem to connect the dots and see these patterns.
I don't know if the problem is my pattern recognition or if my calculation strategy when I look at these puzzles is utterly flawed. I probably need better habits.
Thanks! I need to look that book up now:)
There's a reason my username on chess.com is <readeroftomes>. Barsky's book is among them, though I do have it in Forward Chess format.
OH HECK YES !! Martin ... thats a wondefull idea !!!
Thanks! Also quite happy about how it turned out. Learned some new mathematical techniques in the process :)
thats exactly what interests our type of chess ... are the intricacies, geometric patterns , mathematics and shear magic of chess.. so yes absolutely... a book would be great (i now easier said than done) but if possible.. ill be the first to buy it ..:)
I think you're on to something here! But I do think it still needs some refinement. For example, 1058, 1059, and 1060 are indeed almost all the same thing -- a very well known mating pattern of the knight and bishop against a fianchettoed pawn structure in front of the king. OTOH, 1057 seems very different (starting with the fact that it's two knights, which necessarily makes the geometry pretty different)
I would add that "attacker sum" doesn't necessarily express what you want. After all, in a bank rank mate, who cares if it's a Q or a R? For the cluster just described, who cares if the long diagonal is controlled by a Q or a B? (In other words, in many instances, the Q&B or Q&R are interchangeable as they serve the exact same function). OTOH, a knight is fairly unique, which is exactly why 1057 is so different than 1058 thru 1060.
Thank you for the constructive feedback that pointed to some things I have been working to improve! I have found a better model. Will post more details later this week :)
Very interesting post Martin, great work!
For the attacker_sum why did you add the different values for the piece type? Since each piece attacks a square only once, the attacking power should just be the number of pieces attacking the square, right?
Or if you want to identify which pieces attack the square, it might be better to use a value that is unique to the piece combination attacking the square.
Thanks! My reasoning was that I wanted to measure when different pieces controlled the same squares. A slight error is that the addition might give the sum of another piece, so I think it can be improved
A simple way to make the number unique would be to assign a prime number to each piece type, raise it to the number of times this piece type attacks the square and multiply them together.
For example, if two knights and a rook attack a square, the value would be 3^2*7
That sounds like a promising idea, thanks!!
I love this one!
Martin, I'm curious if you compared your fingerprint with the lichess problems database which includes the tag for a number of mate patters, so we can know if it matches the current fingerprints or not. I have been working in various kinds of fingerprints (mainly opening related) and this one seems to be very promising for tactics!
Thanks :) I think I will try to look into the overlap between clusters and specific keywords next week. Its a good question
also publish the source data intermediates in the whole process before book digestion. ways to get to position sets.. so we can also work our own "magics" without having to reverse engineer or reinvent too much. This is just preventative, as I noticed chess and database transparency not to be high value on the market.. This is an exciting direction.
books are lacking information. they are not lossless. Maybe I don't understand what you mean by book. Also, not yet read, but are you making a difference between static pattern definitions and dynamic patterns. I found named chess mates come in such different flavors of chess culture naming decision.. Not all dynamics paths get named, and not all static mates either.. this is about those that humans find nameable.. (it implies their internal representation of their own thinking about chess, and what is worth naming).
So to make sense of books in your survey, I wonder about why a book not a database of position. Because likely, it might be about paths to "mates". What is a "mate"? The mate in 0... or the move sequence to that. And also, I should say that static patterns are also probably cognitively causal to dynamic patterns.. I will read now.. invested by daring to express myself here..
I would take the puzzles at their starting point, eg mate in 4 or mate in two. And It is correct that making a puzzle book with the checkmates would be rather boring 😄 and of course part of the book would be a technical explanation of how the puzzles were clustered together and maybe together with the code, so others can play with it too
beautiful
It’s a fabulous idea.
Thank you, John!
3×3×3 matrix: (occupant, attacker_sum, pin_flag)
so this is the rosetta stone. the high dimension from number of board plane points or squares (empty in foresight context or not really necessarily yet having the mating conditions), the king field including the king at the center.
Jomega also had such variables.. the 3 interprettable. But perhaps not verbally aligned on the pin vocabulary more the attacker potential covering. There is also a notion of support when attacker itself is occupant.. and the blocking of same color.
And one can choose different aligment of interpretables (or visible on board) but it seems at that level you do have the exact information for calling a static mate... I would need to check the definitions of the words... and compare with jomega study (because I understand those to be complete).
I would be happy to help, but I can't promise productivity.. minimally sound boarding. and very curious about the negative results as, this work you did is what I would have liked to do one day... exploratory data analysis. I willl contact you. The negative result would be in that common curiosity I am happy to see shared, would feed my understanding.
but quick question from second pass reading. The tool that does clustering, its input "vocabulary" set, beside the pure exact mate information as above, does it accept a certain amount of transformation control. I can't say I would be technology savvy. It would be more about the math. of this toward board visible quantities. Than which is the hottest algorithm. And I would not like for you to have high expectations of results. I just have a basis of understanding of such things, at the math level which I might be able to make understandable in context beyond the math symbols.
I think you might anyway have a single book topic there. But books dont talk back when read. I guess a monograph in my previous culture would come have streaks of debatable articles that might have been discussed pre-publicatino through peer review but also in between more than one author clutch (or group or school of thought yes, even in well measurable science) tit for tat askinjg questions that the previous paper did not make explicit in their own struggle to thread the not yet known.
If you can digest the other books that this comment section is bringing to our awareness I would be glad to read about it. I also find that patterns in general in chess have not been systematically investigated, and mostly taken as black box, that can only be communicated through examples.
So, I am familiar (or was) with theory of various clustering approaches, the question, when tinkering with those, becomes the relation between metric choices in that big space and the clustering portrait. notions of boundaries between "attractors" , cluster membership interpretation etc.
how many techniques and choices are behind the result.. I think in your book it might be interestijng to have discussion starters about the many possible clustering angle choices... so we can start lookihg at the human chess board story values of such clustering. And perhaps even consider theories of learning that humans can see in relation to board visible clues at their experience levels.. things like that. A book is a nice thing, but that is a one or 2 persons creation, and sometimes such beautifuul topics like this one, are meant to be discussed at the many persons level allowing for errors and creativity that one human alone in the end, might not be able to explore or see, in that we do all have limitations, the best of us. I guess I would call that scientific culture approach. But a book might be a good conversation starter.. of course.
I agree that there is a human element in it with making all the choices. Made several versions that didn’t manage to cluster similar patterns. If you want to collaborate on this please send me an email: saychess1@gmail.com
I’m not trained in this field, so you most likely have a better understanding than me, so would like to hear your thoughts
Ok so the kill box pattern seems to have all three possibilities.. static pattern trigger, dynamic (combo) pattern, and possibly also "idiosyncratic" board configuration mate static pattern. I am throwing synonyms as emphasis of what I mean.. can you translate?
can you share the code? Jomega on lichess has done some work on automatic pattern detection in math compatible with coding further work, not sure the studies are public or private.. But it is meant to.. I am curious about your "fingerprint" defintions..
So using the kind donut region is common to both your work.. The king field (dotted or not).
it would be nice that mathematically intelligible version of such things be shared.
Having not read the book but heard about it. I heard that Bobby Fischer teaches chess spends a lot of time drilling into a player the ideas about back rank mate. With some of your previous books. I feel that you've presented a lot of useful exercises but haven't quite hit that itch of connecting exercises to the chess world.
So my suggestion would be this: you've got 74 patterns, pick 25.
Of those 25 give a page of 6 example puzzles for each. Then find a game for each of the 25 patterns with a relatively high quality player, and show that full game for that player. This will connect the pattern to an opening, a story, and a method to get to the place that we have just studied.
Of course that's a lot of work, but it is discrete chunks that can be done in a checklist style.
You're really good at finding interesting chess statistics, and positions and this seems like the type of book that would stand out versus other puzzle style books.