人工智能游戏开发.pptVIP

Connect Four using Alpha-Beta Pruning Billy Landowski CptS 540 7 December 2010 第一页，共十七页。 Overview Background Connect Four as a search problem Alpha-beta pruning Details about winning Heuristics Implementation / Demo Conclusion 第二页，共十七页。 Background Sold by Milton Bradley in February 1974 2 players Alternate turns Goal: Connect four tiles in a row horizontally, vertically, or diagonally 第三页，共十七页。 Connect Four as a Search Problem ≤7 possible moves per turn Enumerate each move Continue for each board configuration Player 1 Player 2 第四页，共十七页。 Connect Four as a Search Problem States: Any board configuration with at most one player’s tile in each location Initial State: An empty game board with no tiles. Actions: Place a tile of the current player’s color into any column that is not full. Transition Model: Returns a board configuration with a tile added to the specified column. Goal/Terminal Test: A player has four of her tiles in a line either horizontally, vertically, or diagonally, or the game board is full (indicating a tie). Utility: +∞ if player has connected four, 0 if board is full, –∞ if opponent has connected four. 第五页，共十七页。 Alpha-beta pruning O (bd/2) time complexity b = branching factor = 7 d = depth = 7 × 6 = 42 Computationally intensive Need cut-off depth Can also add heuristics 第六页，共十七页。 Winning Connect Four To win, player needs a “winning line” of 4 第七页，共十七页。 3-out-of-4 Heuristic To win, player needs a “near” winning line of 3 第八页，共十七页。 3-out-of-4 Heuristic (cont.) Count total 3-out-of-4 “unblocked” winning lines Compare to opponent Utility(p,G) = f(p,G) – f(opponent(p),G) f(a,G) = # of 3-out-of-4 winning lines for player a on board G 第九页，共十七页。 Scoreboard Heuristic Extend 3-out-of-4 heuristic to n-out-of-4 for n ≤ 3 Award weighted points based on the value of n Score(p,G) = 100(n3) + 10(n2) + 1(n1) ni is the number of i-out-of-4 winning lines for player p on game board G 第十页，共十七页。 Scoreboard Heuristic (cont.) Five 1-out-of-4 winning lines (n1 = 5) Five 2-ou