This project has been to create an algorithm for completing Minesweeper most optimally. The current bot uses constraint satisfaction solving to first check all for all spaces that can be determined as either definitely a bomb or definitely not a bomb and it will then use probabilistic methods to make optimal guesses when faced with uncertainty.
My favorite part about Minesweeper is the uncertanty. You can develop logical rules for different scenarios to tip the odds in your favor, but at the end of the day certain boards are just going to require guessing. I originally created this project because I was curious how much of the uncertainty you could remove by playing as optimally as possible. My current algorithm solves about 40% of expert Minesweeper boards (18x30 with 99 bombs), which is about 4 points better than what I was averaging myself.
For the algorithm itself, while it will always pick the tile that is least likely to be a bomb, there is currently no good huristic for breaking ties. Ideally, if two tiles are both equally unlikely to be a bomb, the algorithm should open the tile that is most likely to lead to information that will help avoid future guessing. What this huristic should be is not straightforward at all because it has to balance learning information about already open tiles with the chance of opening a new plot entirely. It seems likely that an optimal approach here would need to use some machine learning techniques.
The UI/UX is very rudementary intentionally on this project. Still, it would be nice to do a more clean-up and beautification of the design if I were to actually host this application.
I would like to add a user game mode that can alert users if they play non-optimal moves, or if they lose despite playing optimally. This is more just sometihing I'd like to see about myself and know how closely my play is to what my algorithm considers optimal.
In the project directory, you can run:
Runs the app in the development mode.
Open http://localhost:3000 to view it in the browser.