Paquita wrote:I think that people try to define what is a hard puzzle.
Good luck with this. No two people agree on what "hard" means - not even on whether Hidden Singles are harder or easier than Naked Singles.
Paquita wrote:ISER or skfr rating them and storing them is a step, to collect hard puzzles so bright minds can study them and come up with definitions of what is a hard puzzle.
First, there's no "SER or skfr rating". There's the SER, which has become some kind of common reference, by default. And there's skfr, the buggy re-implementation of it. (Not that a difference of 0.1 or 0.2 means anything real - but, considering the original goal, the fact is there).
Again, good luck for finding a definition.
About "bright minds studying" the SER rated collections, can you name anyone, apart from me, who really studied them and came out with some result (with references to such studies, please)?
Paquita wrote:IYou do have some valuable criterium for T&E(2) puzzles. If all T&E(2) puzzles can be solved with it and it is an indication how hard the puzzle is. My question is, does this cover the hardness definition for all T&E(2) puzzles? Or are there other factors in some hard puzzles. For example, the highrated SER 11.9 puzzles that do not have a tridagon are usually BxB<7. Does this mean they are not hard? (I have a vague idea that "hard" means : hard to be solved by a human. Possibly a human who knows the tricks and constellations).
I don't know what "hard" means, even if you add "hard to be solved by a human"; see the first paragraph.
I'm not in the least interested in defining a unique rating for all the puzzles or in defining what "hard to be solved by a human" means.
My classification system is based on a radically different philosophy. It satisfies precise conditions:
- universal (meaningful in any finite CSP);
- pure logic (pattern-based);
- intrinsic (depends only on the puzzle at hand);
- invariant under isomorphisms;
- invariant under expansion by Singles;
- (decreasing) monotonic wrt expansion.
These properties are essential for any systematic work about classification. Two examples:
- when I expand a puzzle, I don't have to wonder whether that can make it "harder"; in SER, it can;
- when I look for the minimals of a puzzle, I don't have to wonder whether some of them can be "easier"; in SER, they can.
I could add one property:
- doesn't depend on oracles (such as uniqueness).
Moreover, for the specific hierarchical classification system I've introduced:
- T&E-depth defines a partition of all the puzzles;
- B is a complete sub-classif of T&E(1);
- BxB is a complete sub-classif of T&E(2);
- BxBB is a complete sub-classif of T&E(3).
Note that the universality condition implies the classifs cannot be based on any pattern that is not meaningful in all the finite CSPs.
Does this cover all the possible definitions of "hardness": NO. And I couldn't care less.
One question: did you try to recompute the SER of your "anomalous" B7B+ ¨puzzles, without uniqueness?
.