RMaxAgentClass.py 文件源码

def _compute_max_qval_action_pair(self, state, horizon=None):
        '''
        Args:
            state (State)
            horizon (int): Indicates the level of recursion depth for computing Q.

        Returns:
            (tuple) --> (float, str): where the float is the Qval, str is the action.
        '''
        # If this is the first call, use the default horizon.
        if horizon is None:
            horizon = self.horizon

        # Grab random initial action in case all equal
        best_action = random.choice(self.actions)
        max_q_val = self.get_q_value(state, best_action, horizon)

        # Find best action (action w/ current max predicted Q value)
        for action in self.actions:
            q_s_a = self.get_q_value(state, action, horizon)
            if q_s_a > max_q_val:
                max_q_val = q_s_a
                best_action = action

        return max_q_val, best_action