Icohedron / Hexitricity

Deep Convolutional Neural Network, Reinforcement Learning, and Monte Carlo Tree Search to play Hex.

Asynchronous Advantage Actor-Critic (A3C) with Monte Carlo Tree Search and Hex

To train a new network, remove the following files:

• saved_networks/checkpoint
• saved_networks/Hex9x9-v0-Hexitricty.checkpoint
• saved_networks/Hex9x9-v0-Hexitricty.checkpoint.meta
• saved_networks/tf_summaries/events.out.tfevents.*

Then execute async.py using the command

python async.py

To evaluate the network, set async.py's TRAIN variable to False and run the file again.

Dependencies:

• [tqdm] (https://pypi.python.org/pypi/tqdm)
• [gym] (https://gym.openai.com/)
• [tensorflow] (https://www.tensorflow.org/)
• [numpy] (http://www.numpy.org/)