Sudoku

[duke o' york]

Google?

Try typing hard (or difficult) Sudoku, and see what Sergei Brings you.

[snoopy369]

Sudoku.com iirc has a great puzzle generator - shareware, iirc again.

[Adagio]

Google sounds like a scary place


I'll try that puzzle generator, thx

[Atahualpa]

[Q=Urban Ranger]I am still interested in developing an algorithm that solves Sudoku. Maybe some time Real Soon.[/Q]

I once programmed an evolutionary strategy to do the work. With the excessive amount of computational power nowadays, even such wasteful strategies are going to do the trick quite fast.
I guess a GA would have been better, but I was too lazy to think about CrossOver operators.

The fitness was the amount of violated constraints and the goal was to minimize them. The mutation was to exchange 2 or 3 positions in one of the smaller 9x9 grids.

[Dauphin]

I like the Times 'killer' sudoku that they started a couple of weeks ago. More fun than the standard sudoku.

Same rules, except the grid is additionally split by dotted lines. Numbers in squares enclosed within a dotted line add up to the total written on the dotted line. Also, you don't start with any numbers already in place (unless there is a single square enclosed within a dotted line, in which case it de facto tells you what the number is).

[Urban Ranger]

Originally posted by Atahualpa
[Q=Urban Ranger]I am still interested in developing an algorithm that solves Sudoku. Maybe some time Real Soon.[/Q]

I once programmed an evolutionary strategy to do the work. With the excessive amount of computational power nowadays, even such wasteful strategies are going to do the trick quite fast.

Geek

[duke o' york]

Dolphin, I notice today that the Grauniad have got something called Kakuro. Trouble is, their explanation of the rules wasn't great. Maybe if I google it.

[Lul Thyme]

Originally posted by Atahualpa

The fitness was the amount of violated constraints and the goal was to minimize them. The mutation was to exchange 2 or 3 positions in one of the smaller 9x9 grids.

Im not an expert in algorithms so maybe I missed something.
Wouldn't it be possible to get stuck with that algorithm?
Its possible that 2 or 3 position changing is not going to improve the solution anymore, specially if you consider only one 3x3 box at a time...
Of course for easier puzzles it will work...

[Atahualpa]

Yes that's a problem and that's why I thought a crossover or an ES with recombination operator would do better, but it was a quick work and I didn't put that much effort. Plus I could solve most puzzles with several tries. The easier ones even with a 1+1 ES, the tougher ones required a bigger population.