CS164 - Milestone 7

On 4/26, I committed a partial solution to M6 to the repository, in the minesweeper/ directory. It consists of sweeper.py, which implements the basic rules of the game, and contains a __main__ segment to run it with a random player. And the other module is constraints.py, which partly implements our constraint solving over Boolean variables representing the coordinates.

For this milestone, I’d like you to take it a few steps further. Here are some options: do one or more:

Implement ‘prettier’ output of the board in the render function of the Minefield class. Right now it shows an uninspiring matrix of integers:

[[ -1.  -1.  -1.  -1.  -1.  -1.  -1.  -1.]
 [ -1.   2.  -1.  -1.  -1.   2.  -1.  -1.]
 [ -1.  -1.  -1. -99.  -1.  -1.  -1.  -1.]
 [ -1.  -1.  -1.  -1.   1.  -1.  -1.   2.]
 [ -1.  -1.  -1.  -1.  -1.  -1.  -1.  -1.]
 [ -1.  -1.  -1.   2.   0.  -1.   0.  -1.]
 [ -1.  -1.  -1.  -1.  -1.  -1.  -1.  -1.]
 [  0.   2.  -1.  -1.  -1.   0.  -1.  -1.]]

But that could be turned into ‘fancy’ ASCII graphics like this:

a . . . . . . . .
b . 2 . . . 2 . .
c . . . X . . . .
d . . . . 1 . . 2
e . . . . . . . .
f . . . 2 0 . 0 .
g . . . . . . . .
h 0 2 . . . 0 . .
  1 2 3 4 5 6 7 8

Where we include the alphabetic/numeric labels to help us humans with naming coordinates. (Internally, they’ll still be represented as a pair of zero-based integers.)

Implement an input function that can ask for a move from a human player. They will input something like f2 or h8. To prevent accidents, you should error-check that it’s a two-character string, and that the row and column are in the correct range. Then you translate it to the zero-based internal coordinates, so f2 becomes (5,1) and h8 becomes (7,7). Although the size of the board is configurable in this code, your human input only needs to support up to \(9\times9\).

Extend the connection between the game and the constraint solver. Instead of always choosing randomly, or just relying on human input, you can take the set of known-zero coordinates from the solver, and click on any that haven’t already been cleared. You can go a long way with just that, clearing empty regions.

Here’s some code that roughly shows how I selected between random actions, human input, and partial human input (only when guessing).

    while not done:
        # Uncomment below to choose random action:
        #act = env.action_space.sample()

        # Uncomment below to have human player:
        #act = env.get_human_input()

        # Uncomment below to use constraint solver:
        if len(constraints.clearQueue) > 0:
            act = constraints.clearQueue.pop()
        else:
            act = env.get_human_input()
            # Nothing is known for sure, so we have to guess.
            #act = constraints.bestProbs.pop()

The next step after that would be to try calculating probabilities and choosing coordinates in cases where all the known-zeros have been cleared.

Extend the constraint solver so it can deduce more values. I recommend continuing the test-driven development approach we began in class. If you come up with a test case that you can’t figure out how to implement, you can consult me for help.

PDF