feat: alpha-beta prunning

multiAgents.py

@@ -196,7 +196,35 @@ class AlphaBetaAgent(MultiAgentSearchAgent):
           Returns the minimax action using self.depth and self.evaluationFunction
         """
         "*** YOUR CODE HERE ***"
-        util.raiseNotDefined()
+        return self.result(gameState, 0, self.depth, -1e9, 1e9)[1]
+
+    def result(self, state, agentIndex, depth, alpha, beta):
+        if depth == 0 or state.isLose() or state.isWin():
+            return self.evaluationFunction(state), None
+        elif agentIndex == 0:               # maximize
+            (nextAgentIndex, nextDepth) = (0, depth-1) if agentIndex == (state.getNumAgents()-1) else (agentIndex+1, depth)
+            max_reward, max_action = -1e9, None
+
+            for action in state.getLegalActions(agentIndex):
+                reward = self.result(state.generateSuccessor(agentIndex, action), nextAgentIndex, nextDepth, alpha, beta)[0]
+                if reward > max_reward:
+                    max_reward, max_action = reward, action
+                if reward > beta:
+                    return reward, action
+                alpha = max(alpha, max_reward)
+            return max_reward, max_action
+        else:                               # minimize
+            (nextAgentIndex, nextDepth) = (0, depth-1) if agentIndex == (state.getNumAgents()-1) else (agentIndex+1, depth)
+            min_reward, min_action = 1e9, None
+
+            for action in state.getLegalActions(agentIndex):
+                reward = self.result(state.generateSuccessor(agentIndex, action), nextAgentIndex, nextDepth, alpha, beta)[0]
+                if reward < min_reward:
+                    min_reward, min_action = reward, action
+                if reward < alpha:
+                    return reward, action
+                beta = min(beta, min_reward)
+            return min_reward, min_action
 
 class ExpectimaxAgent(MultiAgentSearchAgent):
     """