Choosing clues to show

anon · Posted: Sun Jun 18, 2006 10:44 am Post subject: Choosing clues to show

Hi, I'm trying to make a logic based symmetrical sudoku puzzle generator.
So far I have methods for singles doubles and naked triples.

However, the problem is that practically all puzzles I get are solvable by singles only, and I see the other methods effectively used only in failed attempts to solve the puzzle.

I guess it must be something with the method I'm using to open givens for the solver. My method is to count the number of cells where a candidate will be eliminated for a given pair of clues (without repetitions), and I choose from among the cells that will result in most eliminations.

(In the end the program tries to solve the puzzle without a pair of givens in the same order as they were added, and removes this pair if the puzzle is still solvable.)

Unfortunately, this also seems to eliminate the need for any advanced solving methods. Is there a better way? E.g, should I try to find a pair of cells that would have made a naked triple (or what ever is the most advanced techique for the user selected difficulty) succesful that failed last time?

(I have written a sudoku program before, and I had the same problem there - all the puzzles were just as easy to solve, only the "hard" ones took more time to create. There I chose a pair of givens that had the greatest number of candidates between them.)

Ruud · Site Admin

Hi,

Even when you change your method, you will have this problem, because it is simply a statistical thing.

When I randomly create sudokus, this is the result:

75% easy
12% super hard
8% hard
5% moderate

So, to generate a moderate sudoku, on average you must create 20 to find one. Within those 20, there are 4 that make your solver choke.

My generator has no bias, except that all puzzles are generated symmetrically minimal. This makes the overall difficulty a little bit higher.

Ruud
_________________
Meet me at sudocue.net

daj95376 · Posted: Sun Jun 18, 2006 3:41 pm Post subject:

I have a very convoluted set of logic in my puzzle generator that I implemented with the hopes of creating very hard to fiendish puzzles. Although I've been modestly successful -- mostly just hard -- I would not advise trying this because the total execution time probably takes as long or longer for a hard puzzle as doing what Ruud described above.

Good luck!

anon · Posted: Mon Jun 19, 2006 9:32 pm Post subject:

Thanks for the answers.

I simplified the method for choosing the givens (as it doesn't matter much), and I'm now getting 2-7 puzzles with naked and/or hidden pairs out of 100 puzzles generated (none with a triple so far). So I guess I'll just make and rate a bunch of puzzles and choose from among those.

The good thing I discovered: most of the evil ones at websudoku.com don't have anything else but a few hidden pairs either most of the time .