The New Sudoku Players' Forum

[dobrichev]

Being a poor manual solver I fully agree in this. The rating limit over 8 to 9 I am taking from you.
Yet another answer to the original question by Cenoman.

Are there doubts that the same is not always applicable for expert manual solvers? (asking directly for what you said indirectly in last sentence)
I.e. taking the difference between solving <=> searching for known exotic patterns from one hand, and attempting to invent new exotic techniques from other - should the solving be applicable to the experienced solvers in the same way, only more effectively and with less errors?

[David P Bird]

Hi dobrichev, well said from your side.

"At which point a manual solver should stop searching for more and more complex known patterns and gracefully switch to applying effective to a manual solver guesses?".

From my side I would suggest the forum consider the opposite question: At which point should it be considered bad form to use assumptive methods for solving the easier puzzles?

To re-use my old analogy, we shouldn't need to bring scaffolding with us if we are only climbing molehills.

David PB
.

[ghfick]

Solving methods have developed substantially since 2007. That is 11 years ago. Maybe, over the next 11 years, there will be further development of non-assumptive methods. I now have a thick file folder of partially solved puzzles. At the present time, I choose to consider only non-assumptive paths. The solution is of limited interest to me. The path(s) are interesting. The very elaborate forcing chains with SE scores over 9 are not interesting to me [so far]. So I elect to leave many puzzles unfinished. Maybe I will return to them as new methods emerge.

[dobrichev]

At which point should it be considered bad form to use assumptive methods for solving the easier puzzles?

I have no answer to this question, but somehow this should be a personal choice depending on own learning curve.
I think it is normal for some people to have such a brain that they are effective and find satisfaction in assumptive methods, and respectively develop and use them.

[SpAce]

Solving methods have developed substantially since 2007. That is 11 years ago. Maybe, over the next 11 years, there will be further development of non-assumptive methods.

It's possible, I guess. Then again, if I remember correctly, smarter people than me have opined that if there were any hugely impactful exotic patterns left, they would have been found by now. Maybe it's a bit pessimistic, but I personally wouldn't hold my breath.

I now have a thick file folder of partially solved puzzles. At the present time, I choose to consider only non-assumptive paths. The solution is of limited interest to me. The path(s) are interesting. The very elaborate forcing chains with SE scores over 9 are not interesting to me [so far]. So I elect to leave many puzzles unfinished. Maybe I will return to them as new methods emerge.

What exactly is a non-assumptive method? A pattern that can be recognized with just counting certain elements and then applying the related elimination rules? What's so great about that, actually? Isn't it just about memorizing patterns and then finding them? Isn't that turning a logic game into a pattern recognition game?

I'm not at all criticizing such an approach, or any approach for that matter (as long as it's fun for the user), but I don't see it necessarily better than the alternatives either. In fact, I find it a bit limiting. For me the real fun would be in finding such patterns on my own instead of blindly using others' inventions. Unfortunately I'm very late to the game and probably not smart enough anyway, plus I don't see much value in reinventing the wheel, so I'm quite happy to use others' inventions as much as possible. But, I don't mind using so called assumptive methods either, because it brings some logic back into the game.

Also, have you considered that your limited approach might prevent you from finding new solving methods yourself? I would think that it would require trying to solve those kinds of puzzles you now leave partially solved any way you can and then seeing if there were some recurring patterns in them. I don't have that kind of skills, but you, for example, have much more experience than me. Why do you want to wait for others to invent those new patterns instead of trying to find them yourself? Wouldn't that be much more fulfilling than just using others' inventions, not to mention how it would benefit the community if you did find them?

[David P Bird]

At which point should it be considered bad form to use assumptive methods for solving the easier puzzles?

I have no answer to this question, but somehow this should be a personal choice depending on own learning curve.

Exactly!

So, by personal choice, I don't like using assumptive methods to solve simple puzzles in one step in preference to using two non-assumptive ones. I believe this deters some newcomers and misdirects others. It also does not suit me to rely on a grading system that employs forcing chains long before they are needed.

With difficulty I can program, but I get far more satisfaction from (in your words) dancing the night away, looking for ways I can increase the difficulty of the puzzles I can solve. From what you say, it seems that for you it is the other way round. Regardless, I respect and admire what you are able to achieve.

But Mladen, this is an old discussion topic, and there is little point in resurrecting it.

David
.

[SpAce]

David, I can suspect (at least) one simple reason why you so strongly prefer unassumptive methods: you've developed or at least patternized many of them! It means you've worked out the logic involved when doing that, which probably means you can partly reexperience that satisfaction when using those patterns, too. For those who've only studied and memorized those same patterns it's probably less satisfactory, and thus they may prefer to resort to other methods with more original logic even if they're less elegant or efficient.

The same happens in the programming world a lot, and it's called the Not Invented Here Syndrome (NIHS). Programmers love to develop their own algorithms, frameworks, and libraries even if perfectly suitable (and probably better) ones already exist. It's often inefficient and error-prone, but also more fun (and sometimes quicker than finding and testing an existing component). Customers rarely appreciate that tendency, and would prefer programmers to use standard components when possible and only develop new ones when it truly adds value. It could be argued that the best programmers have probably given in to that tendency at some point in their careers (earning deep-learning), but then developed the self-discipline to suppress it unless really needed (earning customer-satisfaction).



Solving sudokus is a lot like programming, but there's one crucial difference: no customer, which means that self-discipline is totally optional. Its main purpose is to be fun, and for some that fun (and learning) is better achieved by building unnecessarily complicated nets etc. than using standard patterns that do the same trick. So let them (and I don't mean to imply that you don't!). For others it may be the other way around. Personally I like both approaches, just like in programming. I don't think it's possible to deep-learn if you only stick to patterns and methods developed by others, but it would also be stupid to reinvent the wheel in everything.

[SpAce]

Sorry for repeating myself, but could someone please clarify the exact relationship between multi-fish and MSLS? Like Exocets and JExocets they're sometimes talked about as if they were synonyms (including David's MSLS document), and other times as if they're not (like Phil's solver which has both as separate techniques). This is very confusing. Are they synonyms? Or is one a subset of the other (like JExocet is to an Exocet)? Or what?

[champagne]

Hi Space,

MSLS is a rank0 set/linksets logic, as what I call "multifish"

In a multifish, the logic looks for sets within a limited number of digits, using row, columns, boxes sets and cells where the candidates are limited to the digits studied.
In a MSLS, the sets are based on cells.

Usually, when you have one multifish, you have also a MSLS giving the same eliminations, so

both are "rank 0 logic"
one is a kind of complementary view to the other one.

And if you have a SKloop, you have usually 3 different ways to build a multifish rank 0 logic and one MSLS rank 0 logic (at minimum).

Comparing both on the same puzzle helps

[Cenoman]

Hi champagne,
MMLS is an acronym new to me.

A search in the "Advanced techniques" forum displays nothing.
What does it stand for ? Is it your personal naming for MSLS ?

Regards

[champagne]

Hi champagne,
MMLS is an acronym new to me.

A search in the "Advanced techniques" forum displays nothing.
What does it stand for ? Is it your personal naming for MSLS ?

Regards

just dislexy

[StrmCkr]

Muti fish are mutiple digits setup similar to fish patterns and quickly stoped being developed
(sector based)

muti sector locked sets (msls) (cell based)
using the overlapping properties of a naked Subset and a hidden subset to perform eliminations. (home away sets)

Which points to something very lacking development a combination of als-xz rule and Hls-xz rules.

Most of the sudoku stuff only uses one or the other data sets to perform logic eliminations and never both.

The only solver Ive seen use both is Allans xsudo.


Ps using the forum search function dose t find result very easily in many cases it returns nothing or can't search the terms.

it's often better to use google which finds the desired information on here alot easier.

[ghfick]

It is clear to me that MSLS, MultiFish and SK Loops can sometimes yield essentially the same exclusions. Maybe the initial wave of exclusions is not quite the same but then some very easy steps follow and one gets to the same place in the path. For me, the logic does appear very different though. I think it is also true that each of MSLS, MultiFish and Sk Loops do not always yield the same exclusions. Experience with Phil's solver suggests this. Now, it may be that one needs to extend the methods [in ways I do not understand] to show that all 3 methods can offer the same exclusions.

I have commented at length about assumptive and non-assumptive in U296 on Andrew's site. Here it is, repeated :

"Andrew has a number of forcing chains in his solver strategy list. Some of these strategies are also sometimes called Krakens. One can have Kraken Units, Kraken Cells, Kraken Rows, Kraken Columns... There are also Kraken Fish like Kraken Finned X-Wings, Kraken Finned Swordfish...

Many Sudoku writers would say that most, if not all, of these strategies are 'assumptive'. To try to prepare a definition of 'assumptive' requires a definition of a 'pattern'. Then one needs to determine the boundary to what will be called 'assumptive' or what will be called a 'pattern'.

One can create a list of well known 'established' patterns, but the edges to this list are elusive. An MSLS is a pattern. Its logic is clear and the MSLS pattern appears quite often in extreme puzzles. Once you see it and understand its logic, the exclusions follow from it. The Junior Exocet [JE] is a pattern. Its logic is well understood and it appears often in extreme puzzles. Most writers would say that Exocets [that are not Junior] are not patterns. But even the boundaries to JE continue to be expanded. Recently, David P. Bird provided an interesting addition to the JE world in his Compendium thread. Fascinating stuff. Andrew's solver contains some of the JE exclusions [but not all].

A strategy may be called 'assumptive' if its exclusions can only be seen starting with an IF statement [an assumption]. If a chain can be expressed in a bidirectional way, most authors would say that such a chain can be described as a pattern and so it is non-assumptive. If a chain can only be expressed as unidirectional, many writers would say that it is 'assumptive'. Such a unidirectional chain must start with an assumption. Many chains that have 'memory' along its direction would be called assumptive.

A bidirectional chain can have nodes that are patterns of various kinds. A node [or nodes] can be ALS's, X-Wings, JEs... So all of these chains are non-assumptive.

Many authors would say it is worthy and interesting to exhaust all non-assumptive steps in solution path construction. Some authors prefer to leave a puzzle unsolved rather than complete solution paths with assumptive steps. I now have a [growing] binder of puzzles that do not have a solution path. I anticipate that new Sudoku strategies will continue to be developed and that some [?] of these puzzles might eventually have paths.

Then there is the term 'human solvable'. Andrew [in his excellent book] offers a fine introduction to the POM. I think that most writers would say that the POM is non-assumptive. Nevertheless, any application of the POM [by a human] would take hours, days[?] to implement. So is the POM a human solvable strategy? Andrew calls the POM a 'headache'. Nice.

The boundary between assumptive and non-assumptive is hazy to me. A single finned Kraken X-Wing can be seen as a bi-directional chain with an X-Wing as a node. So it can be said to be non-assumptive. [maybe?] Sometimes a chain with memory appears to be unidirectional but then the reverse direction becomes apparent [perhaps with different logic] and so maybe such a chain is now non-assumptive.

Occasionally, a difficult puzzle has more than one interesting solution path using two apparently different but crucial strategies. For example, sometimes a puzzle appears to need a very long list of rather simple chains but then early on in a path an ALS-XZ just kills the puzzle. Some authors enjoy finding a small set of strategies [sometimes even just a single killing step] that solves a difficult puzzle. There is a thread on the forum entirely devoted to finding that one crucial step."

Best
Gordon

[SpAce]

Thanks for the answers for the multi-fish/MSLS question! I'm not sure how much they lessened my confusion, though Let me double-check if I understood anything...

So they are in fact different things? But somehow connected, so that if you have one you also have the other (kind of like naked and hidden sets/fishes)? Sometimes an SK Loop also joins the gang? And they all produce the same eliminations? So far so good -- except then comes Gordon who says that's not always true. My head is already hurting

Could someone produce or link an example that shows a multi-fish and its corresponding MSLS (and possibly SK Loop, too, if applicable)? So far the SK Loop seems the easiest of them to understand (and probably to spot, too).

The one thing that seems clear is the rank-0 logic shared by them all, but that much I'd already gathered. Looks like I need to get Xsudo before I can hope to really understand this stuff. Too bad it only runs on Windows, but I guess I can install it on a virtual box.

[StrmCkr]

Diffrent In regards to construction (hidden/naked set overlapping in cells (more like a complex als +hidden als combined)

) vrs naked set restrained to cover sectors. (exactly like a fish)

So yes they are identical using diffrent data sets because of the data they use and how they are built they have strengths and weaknesses and overlap each other. Which is why they may or may not match eliminations in full.

SK loops are simplified versions of msls (as they were found first) and later known as virus patterns.... Then better understood as msls.

using-multi-sector-locked-sets-t31222.html