Package	com.lookbackon.AI.searching
Class	public class MiniMax
Inheritance	MiniMax SearchingBase AbstractProcess Object

MiniMax.as class. The idea here is that both players will try all possible moves in their position and then choose, respectively, the one which makes the value of the position as high as possible (the white side) or as low as possible (black). I have called one color 'WHITE', this is the side which tries to maximize the value, and the other side tries to minimize the value. You can see that player 'WHITE' starts with a value of -INFINITY, and then goes on to try every move, and always maximizes the best value so far with the value of the current move. The other player, BLACK, will start out with +INFINITY and try to reduce this value. Note how I use a function checkwin(p) to detect a winning position during the tree search. If you only check winning conditions at the end of your variation, you can generate variations where both sides have won, for instance in connect 4 you could generate a variation where first one side connects four, and later the other side does. Also, note the use of handlenomove(p) - that's what you need to do when you have no legal move left. In checkers you will lose, in chess it's a draw. If the (average) number of possible moves at each node is N, you see that you have to search N^D positions to search to depth D. N is called the branching factor. Typical branching factors are 40 for chess, 7 for connect 4, 10 for checkers and 300 for go. The larger the branching factor is, the less far you will be able to search with this technique. This is the main reason that a game like connect 4 has been solved, that checkers programs are better than humans, chess programs are very strong already, but go programs are still playing very poorly - always when compared to humans.

See also

---