A bot for Tic-Tac-Toe that uses deep reinforcement learning trained using keras-rl DQN agent.

All code in main.ipynb

Trained weights are present in dqn_weights.h5f

1. A PPO agent built with stable-baselines.

   All code in MlpPolicy.ipynb

   A trained model with PPO1 of stable baselines is in PPO1_model.
   A trained model with PPO2 of stable baselines is in PPO2_model.

2. A heuristic using bot. A n-step-lookahead bot which uses minimax algorithm to find the optimal move.

   import heuristicBot
   ...
   move = heuristicBot.nslAgent(no_of_step_to_lookahead, game_grid, available_moves_on_grid, player_number)
   ...
   # returns a valid index to where the move should be made
   # PARAMETERS =>
   # no_of_step_to_lookahead = (max 9, min 1)
   # game_grid = a 9x9 2D numpy array with value 0:empty, 1:X, 2:O
   # available_moves_on_grid = list of all empty places on the grid
   # player_number = 1 or 2

numpy==1.19.4
gym==0.18.3
stable-baselines==2.10.2
tensorflow==1.15.0
tensorflow-gpu==1.15.0
python==3.7
keras-rl

by Debashish Gogoi