Tic-Tac Toe

Software Architecture:

RL-Glue mechanism:

• "Environment" - Returns a 'state', set of valid actions and a reward.

• "Agent" - Returns an action

• Run on a common platform

• A sample invocation would be

  ./main.py 100 "OptimalAgent" "TicTacToe:random:RandomAgent"

  this starts the TicTacToe with the Agent being the OptimalAgent, and the opponent is a RandomAgent and is randomly chosen to start first;

• Another sample invocation would be

  ./main.py 100 "PolicyGradient" "TicTacToe:false:OptimalAgent"

  This does the same, with the PolicyGradient as the Agent, and the OptimalAgent as the opponent; though now the Agent always starts first.