DoubleQAgentClass.py 文件源码

def _compute_max_qval_action_pair(self, state, q_func_id=None):
        '''
        Args:
            state (State)
            q_func_id (str): either "A", "B", or None. If None, computes avg of A and B.

        Returns:
            (tuple) --> (float, str): where the float is the Qval, str is the action.
        '''
        # Grab random initial action in case all equal
        best_action = random.choice(self.actions)
        max_q_val = float("-inf")
        shuffled_action_list = self.actions[:]
        random.shuffle(shuffled_action_list)

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

        return max_q_val, best_action