Tic-Tac-Toe-RL

Teach computer how to play Tic Tac Toe via Q-Learning (Pytorch)

• Q_Table : Tabular Q-learning implemented in Jupyter notebook. Positive Q values suggest good move for first player while negative Q values suggest good move for second player. Next state is the state immediately after a move is peformed.
• Q_Supervised : A single neural network to fit the Q-table values learned from Q_table.ipynb.
• Q_Network : Q-Learning with neural network from the ground with two agent compete against each other. Next state the state after the opponent has performed a move.
• game_tab : Environment class for tabular learning
• game_nn : Environment and agent classes for neural network learning and supervised learning
• table : Serialized byte file for storing the Q-table

Q-value update rules for this project

• Q-table / Q-network update immediately after each move
• Q(s_t , a_t) = reward + 0.9 * Q(s_t+1 , a_best)