I’m particularly partial to Simon Tatham’s version, which only gives you solvable puzzles. There is definitely something reassuring in that you always know a puzzle is solvable, so it’s just you against it. (There is a desktop version on the main page for those so inclined).
This was a lot of fun, well done on the interactive article. I wish the solver-assisted final version would pointed out some deduction you could have made when you err, as it is making a mistake is needlessly frustrating. Edit: turns out the known cells are marked a subtle (to my eyes) red.
It’s also quite easy to get it spinning in seemingly clear situations, such as when you’ve uncovered nothing but a few isolated numbers. (Try it: ask for help right from the start, and spread those over the board. Here it gets very slow starting at 5 or so uncovered squares.) It should not be too hard to modify the AI to only permute all squares that neighbor uncovered ones, and treat the other squares equally.
Hand-crafting doesn’t scale! When I wanted minesweeper puzzles (but I was OK with small ones), I implemented a brute-force solver, a very primitive pattern-based solver, and then ran them in a loop: generate a field, if nothing can be opened ot marked — open a random empty cell next to the already opened ones (if there are any), if pattern-matching allows doing something — do it, else let the human try. That actually produced quite interesting (small) puzzles.
I’m particularly partial to Simon Tatham’s version, which only gives you solvable puzzles. There is definitely something reassuring in that you always know a puzzle is solvable, so it’s just you against it. (There is a desktop version on the main page for those so inclined).
This was a lot of fun, well done on the interactive article.
I wish the solver-assisted final version would pointed out some deduction you could have made when you err, as it is making a mistake is needlessly frustrating.Edit: turns out the known cells are marked a subtle (to my eyes) red.It’s also quite easy to get it spinning in seemingly clear situations, such as when you’ve uncovered nothing but a few isolated numbers. (Try it: ask for help right from the start, and spread those over the board. Here it gets very slow starting at 5 or so uncovered squares.) It should not be too hard to modify the AI to only permute all squares that neighbor uncovered ones, and treat the other squares equally.
I’d dispute the claim that this is more fun than classic Minesweeper, though! Nothing wrong with a bit of twitch. And if it’s the puzzling you’re going for, it will be hard to beat a good hand-crafted minesweeper puzzle, such as https://www.gmpuzzles.com/blog/2017/12/minesweeper-john-bulten/ or (shameless plug) https://maybepuzzles.files.wordpress.com/2016/05/mines.png.
Hand-crafting doesn’t scale! When I wanted minesweeper puzzles (but I was OK with small ones), I implemented a brute-force solver, a very primitive pattern-based solver, and then ran them in a loop: generate a field, if nothing can be opened ot marked — open a random empty cell next to the already opened ones (if there are any), if pattern-matching allows doing something — do it, else let the human try. That actually produced quite interesting (small) puzzles.