Slashdot Mirror
The Godfather of Sudoku
circletimessquare writes "The New York Times profiles 55 year old Maki Kaji who runs Nikoli, in its article Inside Japan's Puzzle Palace. Nikoli is a puzzle publisher that prides itself on 'a kind of democratization of puzzle invention. The company itself does not actually create many new puzzles — an American invented an earlier version of sudoku, for example. Instead, Nikoli provides a forum for testing and perfecting them.' Also notable is how Mr. Kaji describes how he did not get the trademark for Sudoku in the United States before it was too late. But reminiscent of a theme many Slashdotters will find familiar about intellectual property: 'In hindsight, though, he now thinks that oversight was a brilliant mistake. The fact that no one controlled sudoku's intellectual property rights let the game's popularity grow unfettered, Mr. Kaji says.' Will Nikoli be the source of the next big puzzle fad after Sudoku?"
Nikolai makes my favorite type of puzzles, Nurikabe. Hopefully those will pick up in the US too.
Actually a generator which will generate puzzles at a given level of difficulty is not too hard if you organize the solver as a rule based system (Rule 1: first look to see if there is a square that only allows one number; Rule 2: look for pair of numbers that must exist in a pair of cells{sometimes called hidden pairs} . . . )
After you create the rules that will solve the puzzles you categorize the rules as easy, medium or hard.
Now if you want to create an easy puzzle you start randomly adding in (legal) numbers and each time you try to solve it using only the easy rules. The first time the solving part of the program solves the puzzle you are done. Alternatively you can start with a completed puzzle and remove a number and see if the number you removed is recoverable with the level of rules (easy, medium, hard) that you choose for the final puzzle.
Yeah, I wrote one in Perl. Also one for Kakuro.
;)
My Perl sudoku solver was probably more complete - it even does set analysis: [A, B, C] are blank, values [2, 5, 7] are available, but A can only be [5, 7], B can only be [5, 7] so C must be 2.
It reverts to brute force only where its algorithms can't find any possible moves. I've only found that to be the case for puzzles with more than one legal solution (crappy cheap books).
Kakuro is more interesting - first the format is a cow to transcribe to a data file. I'm using "X Y DIRN LEN SUM" as the format. Output is a HTML file because plaintext doesn't quite read easily (my sudoku solver just outputs ASCII art).
The solver itself actually turned out to be very easy to write, and I get the feeling that it's not even complete yet (there are things I can do by hand that it doesn't attempt) and yet every puzzle I've thrown at it solves fine. It doesn't have a brute-force mode. I suspect this may be because either the kakuro book I'm using isn't very advanced, or it's just not so popular yet that people have crafted difficult puzzles.
What I'd really love to see is a standard data file format for describing these puzzles so I can share them and/or use other peoples' puzzles as input to my scripts. But I don't have THAT much free time, really...