dynmi / black_jack

21 points(21点) / 强化学习 / Black jack(黑杰克) / 蒙特卡洛法

Solution for problem Blackjack in Sutton's Reinforcement Learning books

Language

• python 3

Package required

• numpy

• random

Setting details

• Action： a =0 or 1

  • a=0 stick
  • a=1 twist // automatically twist if sum of cards < 12

• State： s = [my_sum, ace_own, s_card] //[13,0,8]

  • my_sum ：the current sum of my cards.
  • ace_own ：agent own a ace card or not // 0 for yes 1 ;for no
  • s_card ：host showing card

• POLICY： pai[1] = [0.4, 0.6] // under State[1], prob for stick is 0.4, prob for twist is 0.6

• Value function： Q [2] [0] =23 // q of (State[1], stick) is 23