• Ability to rapidly prototype new reinforcement learning agents within the OpenAI gym.
• Use gym terminology.
• Eay integration with Keras.

• env.monitor (ran by test_gym.py) requires ffmpeg (installation instructions)

• Heavily based on https://github.com/tambetm/simple_dqn/ (particularly coding style) and DeepMind's Atari Deep-Q-Learner
• Many thanks to all!

Required:

• act(self, observation, reward)

• A3C.py reproduces Asynchronous Methods for Deep Reinforcement Learning