amaynez / TicTacToe

Tic Tac Toe game, designed to be used to train a Deep Neural Network via Reinforcement Learning (DQN). It can also be played by 2 humans and features a hard coded AI that never looses and will win if you do not do perfect play against it.

Deep Q Learning TicTacToe

This program implements a PyGame TicTacToe that can be played by two humans, by a human vs. an algorithmic AI, and a human vs. a Neural Network trained by playing against the algorithmic AI.

The training algorithm uses Deep Mind's DQN recommendations:

• A replay experience memory was implemented
• Batches of random experiences from the replay memory are used for every training round
• A secondary neural network was used to calculate the future Q values and then it was updated with the main network's weights every 10 games

OwnNNet_TicTacTow.py is the main program that uses the ML library I created.

TF_TicTacToe.py is the version that uses TensorFlow (I used to validate my code)

PlainGame_noNNet.py is a version with just the plain game.

For more Information: https://the-mvm.github.io/deep-q-learning-tic-tac-toe.html