I'm working on an Othello game that tries to use minimax?

I made a heuristic function for it:

    int getHeuristic() {
        // give each square a certain value
        // corners are most valuable

        int[][] values = {
                {10, 2, 7, 7, 7, 7, 2, 10},
                {2, -4, 1, 1, 1, 1, -4, 2},
                {7, 1, 1, 1, 1, 1, 1, 7},
                {7, 1, 1, 1, 1, 1, 1, 7},
                {7, 1, 1, 1, 1, 1, 1, 7},
                {7, 1, 1, 1, 1, 1, 1, 7},
                {2, -4, 1, 1, 1, 1, -4, 2},
                {10, 2, 7, 7, 7, 7, 2, 10}};

        int theHeuristic = 0;

        for (int y = 0; y < 8; y++)
            for (int x = 0; x < 8; x++) {
                if (boardColors[x][y] == 'b') theHeuristic += values[y][x];
                if (boardColors[x][y] == 'w') theHeuristic -= values[y][x];

            }

        return theHeuristic;
    }

Anyone know how to make this better?

Also, has anyone tried using a neural network to power the AI for Othello?

You could consider indexing the board etc by a counter 0..63 rather than all the x,y stuff, where you construct such a counter using board[y][x] on each access.

Functionally, nothing will change of course.

I don't know how to program othello, but I would be very surprised if the Internet has nothing about it. As a general rule, assume that if you think of something, someone else in the past in the world has had that thought already too.

As such, the answer to your final question is most likely “yes”, and the Internet can likely provide all the details and more.

@Alberth Why do you think a one-dimensional array is better?

Thanks.

ms75214 said:

@Alberth Why do you think a one-dimensional array is better?

Thanks.

You asked for “better” without really specifying what that means, ie nicer code, faster, improved heuristic, etc.

I went for “nicer code”, and one of the things I noticed is the same [x][y] things everywhere (although values does [y][x], not sure why). If you merge both counters into one, you can compute the offset once, and use it everywhere, or even drop the computation completely since you have a fixed order of looping over the fields, so you can write a loop that simply produces the correct offset directly.

I don't know how to make a better heuristic function for Reversi. You can try to find the opinion of experts online.

Using a neural network for a board game like this makes perfect sense, and you can see how projects like Leela Zero (go) and Leela Chess Zero have done it.

The current breed of NN projects use MCTS instead of minimax, but there is no reason you can't get this working with minimax. A few years ago I trained a small neural network to play Spanish checkers using minimax and it worked really well.