6-Dyna-Q-算法

1
class DynaQ:
2
    ...
3
    def q_learning(self, s0, a0, r, s1):
4
        td_error = r + self.gamma * self.Q_table[s1].max(
5
        ) - self.Q_table[s0, a0]
6
        self.Q_table[s0, a0] += self.alpha * td_error
7

8
    def update(self, s0, a0, r, s1):
9
        self.q_learning(s0, a0, r, s1)
10
        self.model[(s0, a0)] = r, s1  # 将数据添加到模型中
11
        for _ in range(self.n_planning):  # Q-planning循环
12
            # 随机选择曾经遇到过的状态动作对
13
            (s, a), (r, s_) = random.choice(list(self.model.items()))
14
            self.q_learning(s, a, r, s_)