Human composed Sudoku vs Machine made

Everything about Sudoku that doesn't fit in one of the other sections

Postby motris » Thu Apr 27, 2006 5:33 pm

To throw in my two cents, having completed several books of Nikoli Sudoku puzzles as well as some of Pappocom's over here (most recently, Grand Master Ultimate - not sure how the UK offerings differ, but this is the name of the title in the States), I think both puzzle creation methods offer different things.

I do not think that many of Nikoli's puzzles, even in their hardest collections, reach the level of difficulty of computer-generated puzzles. This is not to say that all hand-written puzzles are easier as I've seen some that are very very hard (in fact, some of Tetsuya Nishio's super hard books such as the one I'm currently working on are probably a great counterexample to the statement that hand-written puzzles are easier). Its just that, in the printed books I own from Nikoli and from Pappocom, I've spent the most time solving the Fiendish ones in Wayne's books than I have on Nikoli's, even in their Gekikara (Super Hard) Sudoku series. Computer generated puzzles can more readily, I believe, be forced to have a particular kind of step required in their solving. As a solver who tends to never list all candidates in a cell (I solve for speed and minimize pencil marks to what I consider "useful" information), there are some techniques computer-generated puzzles require that are harder for me to see. They are not unreasonable techniques to require in a sudoku, just not ones Nikoli puzzles tend to have. It may go back to Wayne's point about what is a reasonable and an unreasonable puzzle to have someone solve on paper but, as Nikoli's puzzles tend to be more "reasonable", they may seem in most of their forms to be easier.

On the other hand, there is something tremendously organic and esthetically pleasing about Nikoli's puzzles. It is not, as mentioned earlier in this thread, just the shape of the givens in the grid. For some of their puzzles (not necessarily all), one of two things is true: (1) the number placement within the shape placement is fundamentally beautiful - admittedly gimmicky, but there is a repeated clustering of numbers in a repeated shape, a way that the center box is framed with diagonals that repeat the same number, etc. - that is not something computer-generated puzzles ever tend to have. Just by looking at a puzzle in class (1), you can tell it was likely hand-written and I'll make it a Turing test for sudoku for when a computer-generated one can have the instant esthetic beauty of a good hand-written one. (2) The look of the puzzle may not seem that special, but then the solving of the puzzle shows that it is. As an example off the top of my head, there may be a particular symmetry in the solving route that you use to answer the puzzle. There may be a trick such as a pointing pair to get information in box 8 that then solves a number in box 7. On the other side of the puzzle, the same exactly placement and rule gives information in box 2 that solves a different number in box 3. This kind of logic then propagates around the puzzle in an interesting spiral. While computer generated puzzles may be symmetric, few have been set up so that the solution also follows a symmetric path. On the other hand, it often feels like many hand-written puzzles have something along these lines. There is then the rare puzzle that combines both feature (1) and feature (2) and these will only, at the moment, ever come from hand-written puzzles. These tend to be too gimmicky to be "hard" puzzles. At the same time, I'd say they are very fun sudoku to do and what I would suggest someone who hasn't done many before try out as they may catch on to the "voice" of the author telling them what they should be looking for as they continue to make progress. I never think the computer-generated ones are hinting to me at how to solve them - I'll end up just going down my list of solving techniques (check each digit for any information, check rows, check columns, etc) until they are finally solved.

I would say the above argument holds - perhaps even better - for many many other logic puzzles that can be either computer-generated or hand-written. From slitherlink to masyu to fillominos to nurikabe, there are "gimmicks" that are rare, but special, in the hand-written ones that will not come from random computer generation.

Thomas Snyder
motris
 
Posts: 71
Joined: 13 March 2006

Machine generated

Postby Ant » Tue May 16, 2006 8:47 pm

I haven't played much Sudoku, but got started because my wife plays lots and lots. The free puzzles we get in the daily free paper however were occasionally not solvable (well as far as a bunch of us could work out). So I set about writing an algorithm - ok its based on what a few of us know - but it proved that they aren't solvable. Once I had the solver, it wasn't too hard to go the extra mile and create a generator. This works by taking a complete grid which obeys the rules, and then randomly removing numbers until the puzzle is no longer solvable. I took it one step further and now it generates other size puzzles too, for example 2x2, 2x3, 2x4, etc. [Edited by Luna, website containing adverts] Anyway, my point is that at least these computer generated puzzles are guaranteed to be solvable, which hand generated puzzles cannot be, due to human error. And personally, I find it very irritating when I get near the end of a puzzle and cannot solve it...
Ant
 
Posts: 1
Joined: 16 May 2006

Computer Generation

Postby davidecullen » Sat May 20, 2006 9:45 am

Hi,

There are as many ways to program a sudoku generator as there are
of solving the sudoku puzzles and possibly more, since most people
would never attempt guessing everything, which is a possibility for
creating.

If someone that has a good method for creating sudoku puzzles by
hand, writes a program that follows exactly the same logic they use
when making them by hand, there will be no difference in the quality
of puzzles.

The reason many people talk about computer generated puzzles being
not as good, is because many generators are optimized for speed of
generation and not to follow any of the human style logic. This is
probably essential in hand-held games that generate puzzles on the
fly, but puzzles you find on the internet can have been generated in
advance and take as much computing power as necessary to
create just one puzzle.

Regards,
Dave@sudokuoftheday.co.uk
www.sudokuoftheday.co.uk
davidecullen
 
Posts: 2
Joined: 20 May 2006

Postby Spok » Tue Jan 30, 2007 2:29 pm

I have been reading this thread with great interest and felt it time to add my two-penneth.

I visited the Nikoli page at http://www.nikoli.co.jp/en/puzzles/sudoku/hand_made_sudoku.htm where they go on about how wonderful their hand-made Sudoku puzzles are. Well, I have to say, I have never read such complete nonsense in all my life. All that stuff about the relationship with the solver etc. is just plain mumbo-jumbo.

Firstly, although I have no proof, I would be very surprised if the people at Nikoli actually have rooms of Sudoku masters creating Sudoku grids literally by hand. Think about what you would have to do:

(1) Create a valid Sudoku grid with all the numbers filled in.
(2) Remove numbers to leave enough so that the puzzle still has just one solution.
(3) Make sure the remaining numbers form a symmetrical pattern.
(4) Then solve the puzzle to check its validity and its strength.

How long do you think that would take a human?

By the way, if Nikoli have a different method to the above for hand-crafting puzzles, please can they let us know on this forum.

I actually think that Nikoli generate their puzzles by computer, but try to pull the wool over the eyes of the world by claiming they are made by hand, adding some mystery and magic into the language they use.

There is no task that a computer can do that a human cannot. The computer just does the task very much quicker. When a programmer sits down to design a computer program for a particular task, they first ask themselves how they would do it manually. The program is then made to follow the same steps.

The blind trial that Wayne suggested earlier would be proof enough that there is actually no difference between Nikoli Sudoku and others. All genuine Sudoku puzzles only have to obey three simple rules.

(1) Be solveable by pure logic alone.
(2) Have only one solution.
(3) Form a symmetrical pattern.

If a puzzle does not obey all three rules it cannot be called Sudoku and the silicon machine knows this as well as the carbon one .




Spok.
Last edited by Spok on Thu Feb 01, 2007 6:59 am, edited 1 time in total.
Spok
 
Posts: 6
Joined: 03 January 2006

Postby re'born » Tue Jan 30, 2007 6:39 pm

Spok wrote:All genuine Sudoku puzzles only have to obey three simple rules.

(1) Be solveable by pure logic alone.
(2) Have only one solution.
(3) Form a symmetrical pattern.

If a puzzle does not obey all three rules it cannot be called Sudoku and the silicon machine knows this as well as carbon one .


Your first rule is implied by the second rule, so you don't need it. Why must a Sudoku have a symmetrical pattern to be genuine? Certainly the rules for playing asymetrical Sudoku are the same. The techniques for solving them are the same (Gurth's Symmetrical Techniques notwithstanding). Other than some subjective aethestics argument, for what reason should we require our Sudoku to be symmetric?
re'born
 
Posts: 551
Joined: 31 May 2007

Postby tarek » Wed Jan 31, 2007 12:39 pm

I actually believe that computer generated puzzles (given the correct settings) would generate BETTER puzzles........

TOne other simple simple reason is that all puzzles are actually ALREADY THERE, Given the correct settings the computer also can filter a big number of randomly generated valid puzzles & will eventually find a TAILOR-MADE puzzle for......

You do not need to look far ........ Sudokus with an original rare shape, The hardest sudokus & the superior puzzle collections to name a few......

tarek
User avatar
tarek
 
Posts: 2611
Joined: 05 January 2006

re: symmetry - language confusion

Postby Pat » Thu Feb 01, 2007 9:09 am

re: symmetry - language confusion

Spok wrote:
    All genuine Sudoku puzzles only have to obey three simple rules
    1. Be solveable by pure logic alone
    2. Have only one solution
    3. Form a symmetrical pattern
      If a puzzle does not obey all three rules it cannot be called Sudoku

    tso (2006.Feb.2) wrote:
      Nikoli defined a SuDoku as a Number Place puzzle with symmetricaly placed clues.

      However, due to language confusion, it is likely that Nikoli insisted that the clues have order, but not necessarily symmetry.

      They have published puzzles in which the clues were not symmetrically placed,
      but formed a recognizable shape, letter or kanji.
User avatar
Pat
 
Posts: 3423
Joined: 18 July 2005

Postby Spok » Thu Feb 01, 2007 10:57 am

Spok wrote:All genuine Sudoku puzzles only have to obey three simple rules.

(1) Be solveable by pure logic alone.
(2) Have only one solution.
(3) Form a symmetrical pattern.

If a puzzle does not obey all three rules it cannot be called Sudoku and the silicon machine knows this as well as carbon one .


rep'nA wrote:Your first rule is implied by the second rule, so you don't need it.


Not my rules, by the way. However Rule 2 is easier to test with a computer.

rep'nA wrote:Why must a Sudoku have a symmetrical pattern to be genuine?


I agree again. It was Nikoli who decided that genuine Sudoku should have symmetry and it seems to have stuck. Most computer programs (including Nikoli's) now produce puzzles with either rotational or reflective symmetry. Symmetry has always been very pleasing to those of us with a strong mathematical mind. We tend to like order.

Spok.
Spok
 
Posts: 6
Joined: 03 January 2006


Previous

Return to General