1
$\begingroup$

I am currently in college, and trying to learn reinforcement learning by myself. My primary goal is building an agent that play games such as dots and boxes.

I have sufficient highschool maths knowledge and I started studying Sutton for RL. Along with that I also studied probability from bertsekas.I also did a course on neural networks from Coursera, though very basic.

Currently while reading Sutton I have many questions, so please tell me if this a good approach to study RL. My main aim is to create a dots and boxes agent and also learn fundamental maths involved with RL on the way.

$\endgroup$
  • $\begingroup$ I taught myself basics of RL mainly by reading Sutton & Barto, so it is possible. Whether or not is a good approach for you, we cannot really answer, too much depends on what you already know, how you learn best etc. Once that is removed (as mainly opinion/too difficult), it is hard to find an answerable question here. Perhaps instead you could ask one - please just one, you can always ask more later - of your "many questions" which I assume are going to be more practical such as how you would set up state, actions and rewards for your game. $\endgroup$ – Neil Slater Jan 6 at 17:27
  • $\begingroup$ I would rather say, Approaches like Hopfield Networks are good for those kinds of game. $\endgroup$ – user27450 Jan 6 at 18:09
  • $\begingroup$ Welcome to SE:AI! This post hasn't received closevotes, so I'm leaving open for now (but be advised it may get closed at some point, not being the type of query ideally suited for this stack.) Noticed you posted on chat, so thanks for heeding that advice. $\endgroup$ – DukeZhou Jan 6 at 21:58
  • $\begingroup$ I think this question includes multiple questions, with OP asking whether he or she should apply RL to dots and boxes, but also asking is Sutton a good approach of studying AI, and both of these are opinion based.. @DukeZhou $\endgroup$ – Clement Hui Jan 7 at 0:13
  • $\begingroup$ From RL point of view it's the same as connect-4 and gomoku. Both have open sourced implementation based on AlphaZero on Github: github.com/junxiaosong/AlphaZero_Gomoku github.com/Zeta36/connect4-alpha-zero $\endgroup$ – mirror2image Jan 7 at 6:20