tic-tac-toe: extended

Applying RL to play tic-tac-toe

This work builds upon this source here

note: this is no longer an install-and-play product because it was extended. I should have branched out in retrospect :)

RL formula

TODO:

• Added player 2 policy
• Improved human vs computer
• Improved logging
• Performed action space analysis
• Add the ability to train on top of previous policies
• Make above ^ more user-friendly
• Pass command-line arguments
• Extend to 4x4? -- this endeavour has revealed many problems
• Implement other variants?

---

The Ultimate Draw 3x3:

note: this is based on policy 582023_19237

Most Played Endings:

note: information is extracted from ./analysis_3by3/games.json

This position was played 1549 times!
  p1: x  p2: o
   
   x|o|x
   -+-+-
   o|x|x
   -+-+-
   o|x|o

This position was played 1490 times!
  p1: x  p2: o
   
   x|x|o
   -+-+-
   o|x|x
   -+-+-
   x|o|o

This position was played 1483 times!
  p1: x  p2: o
   
   x|o|x
   -+-+-
   x|x|o
   -+-+-
   o|x|o