2021-10-17 2.1

# nA为动作空间数量， T为进行的总时间步
def epsilon_greedy(nA, R, T, epsilon=0.6):
    r = 0  # 初始化累计奖励r
    N = [0] * nA  # 对动作空间进行全零初始化

    for _ in range(T):
        if np.random.rand() < epsilon:
            a = np.random.randint(q_value.shape[0])
        else:
            a = np.argmax(q_value)

        v = R(a)
        r = r + v
        q_value[a] = (q_value[a] * N[a] + v) / (N[a] + 1)
        N[a] += 1
    return r