Q-learning尝试

参考了Q-learning理解、实现以及动态分配应用（一）_洋葱专栏-有灵魂的程序员-CSDN博客_q-learning简单来说Q-learning需要Q表和R表，利用R表来保存reward，在应用Q-learning时，利用np.matrix得到矩阵类型的数据方便处理，然后利用公式更新Q表。Q-learning需要知道每一步的奖惩。

import random
import numpy as np
def get_qlearning(generation):
    R = np.matrix([[-1, 0, 100],
                    [0, -1, 100],
                    [100, -1, 0]])
    Q = np.zeros((3, 3))
    Q = np.matrix(Q)
    l_rate = 0.8
    action = []
    for i in range(generation):
        state = random.randint(0, 2)
        for j in range(2):
            nextstate = R[state, j]
            if nextstate >= 0:
                action.append(nextstate)
                Q[state, j] = R[state, j] + l_rate * (Q[j]).max()
    print(Q)

if __name__ == "__main__":
    get_qlearning(5)