我已经实现了具有alpha-beta修剪的minimax算法。为了获得最佳移动,我使用rootAlphaBeta
函数调用alpha-beta算法。但是,在rootAlphaBeta
函数中,我发现了一些非常奇怪的行为。当我用rootAlphaBeta
4调用ply
函数时,它会进行大约20 000次调用,但是当我直接调用alphaBeta
函数时,它只会进行大约2000次调用。我似乎无法找到问题所在,因为呼叫的数量应该是相同的。
两种算法最终找到的举动应该是一样的,对吗?我是这么认为的,至少移动的得分是一样的,我无法知道alphaBeta
在没有rootAlphaBeta
的情况下直接调用它时所选择的移动。
def alphaBeta(self, board, rules, alpha, beta, ply, player):
"""Implements a minimax algorithm with alpha-beta pruning."""
if ply == 0:
return self.positionEvaluation(board, rules, player)
move_list = board.generateMoves(rules, player)
for move in move_list:
board.makeMove(move, player)
current_eval = -self.alphaBeta(board, rules, -beta, -alpha, ply - 1,
board.getOtherPlayer(player))
board.unmakeMove(move, player)
if current_eval >= beta:
return beta
if current_eval > alpha:
alpha = current_eval
return alpha
def rootAlphaBeta(self, board, rules, ply, player):
"""Makes a call to the alphaBeta function. Returns the optimal move for a
player at given ply."""
best_move = None
max_eval = float('-infinity')
move_list = board.generateMoves(rules, player)
for move in move_list:
board.makeMove(move, player)
current_eval = -self.alphaBeta(board, rules, float('-infinity'),
float('infinity'), ply - 1,
board.getOtherPlayer(player))
board.unmakeMove(move, player)
if current_eval > max_eval:
max_eval = current_eval
best_move = move
return best_move
答案 0 :(得分:4)
您的rootAlphaBeta
未更新alpha
值。当它可以缩小除第一个子节点之外的所有子节点的范围时,它会调用其所有子节点的全部范围(-inf,inf)。这将防止修剪一些对最终得分没有影响的分支,并增加节点数。