jiumao2/myMineSweeperAI

Playing Minesweeper itself

myMineSweeperAI

Playing Minesweeper itself

Overview

• Minesweeper Version: Minesweeper Arbiter 0.52 distribution
• Time for each game: about 6s in Expert mode
• Win Rate: 31.1%
  
• Algorithm: clicking the block that is least likely to be bomb.

How to use it?

1. Install Python with proper environment.
2. Set hyperparameters in hyperparams.py.
3. Run ms.py in terminal python ms.py. You will see MineSweeper running itself.
4. Stop running by Middle-Click with your mouse
5. The outcome would be saved in log.txt. Each 0 denotes failure of 1 game and 1 denotes success of 1 game.

Outline of the algorithm

1. Get current minefield by reading the memory.

2. We implement 2 different algorithms to solve the task.

(1) Naive

• For each block, if the number of bombs around it equals to the unexplored blocks around it, mark all unexplored blocks around it with flags.
• For each block, if the number of bombs around it equals to 0, click all unexplored blocks around it.
• If nothing is done above, click a random unexplored block.
• Go back to the first step until the game is over.

(2) Normal

• We regard each of the unexplored blocks as a variable. The value of these variables is 0 (safe) or 1 (bomb). And these variables satisfy many equations according to the blocks with number. Therefore, the task is to solve the multivariate linear equations.
• We solve it by output_possibility = np.matmul(np.linalg.pinv(x), y). (not very accurate)
• Mark flags if output_possibility is 1 and click if output_possibility is 0. If nothing is done, then click the block with minimum output_possibility.