Tensorflow code based on https://github.com/MorvanZhou/Reinforcement-learning-with-tensorflow/tree/master/contents/10_A3C

To start training from scratch, create a new directory saves/<mydir> and a configuration file saves/<mydir>/config.py (see existing models in saves). Training is started by calling python3 run.py <mydir>.

Available flags are

• -t, --test: Run in test mode. In this mode no training is performed and visualization is enabled.
• -n, --n_episodes: Train for a fixed number of episodes (useful when submitting jobs to a server). Without this flag, the program will run indefinitely until user interrupt.

Run tensorboard --logdir=saves/<mydir>/log to start tensorboard and monitor training progress on http://localhost:6006/