cubetastic33 / dql-for-tic-tac-toe

I trained a deep Q-learning model to play tic tac toe

dql-for-tic-tac-toe

I trained a deep Q-learning model to play tic tac toe

Preview

Network Architecture

• Input layer of 27 neurons
• Fully connected layer with 64 neurons and ReLU activation
• Fully connected layer with 32 neurons and ReLU activation
• Output layer of 9 neurons

Train the model

Note: If you just want to play against the model, there's already a pretrained model that you can play against.

• It is recommended to use a venv
• Install the dependencies with pip install -r requirements.txt
• Run the training script: python train.py
• It saves the model to a checkpoint file every SAVE_FREQUENCY episodes, which is set by default to 500.
• If you directly run the train.py file, it will continue training from where it left off. You can either delete the HAL9000.pt file or change the SAVE_FILE path to train from scratch.

Play against the model

Just run python play.py after installing the dependencies