alpha/beta pruning
<games, algorithm> An optimisation of the minimax algorithm for choosing
the next move in a two-player game. The position after each move is assigned a
value. The larger this value, the better the position is for me. Thus, I will
choose moves with maximum value and you will choose moves with minimum value
(for me).
If it is my move and I have already found one move M with value alpha then I am
only interested in other moves with value greater than alpha. I now consider
another of my possible moves, M', to which you could reply with a move with
value beta. I know that you would only make a different reply if it had a value
less than beta. If beta is already less than alpha then M' is definitely worth
less than M so I can reject it without considering any other replies you might
make.
The same reasoning applies when considering my replies to your reply. An alpha
cutoff is when your reply gives a lower value than the current maximum (alpha)
and a beta cutoff is when my reply to your reply gives a higher value than the
current minimum value of your reply (beta).
In short, if you've found one possible move, you need not consider another move
which your opponent can force to be worse than the first one.
(1997-05-05)
Nearby terms:
ALPHA « Alpha « Alpha AXP 21164 « alpha/beta
pruning
» alphabetic language » alpha conversion » Alpha EV6
|