Project done during the year 2016/2017 at Ingesup (Enimia Lab) :

• Project : Discover DeepQLearning with a TicTacToe.

• Contributors :

  • Clément ROMAC
  • Nicolas Luvison

We used a FeedForward Neural Network (implemented with Keras) with an experience replay buffer.

The main file to start is game.py.

The current code in game.py allows you to play against the best trained AI.

Uncomment the training code in the main function of the game.py. We've implemented two types of training an AI :

• Against a random player
• Against itself

The training loss is monitored in Tensorboard. You can start Tensorboard with :

tensorboard --logdir=./Graph