Dist-A3C

TODO: Have server use mp - one thread for server, one for testing. Keep counter to know once finished. Also be able to send push notifications to kill running clients once counter done.

Distributed asynchronous advantage actor-critic (A3C) [1] with generalised advantage estimation (GAE) [2]. Run python server.py <options> to start the server and python client.py <options> for as many clients as wanted.

Requirements

• OpenAI Gym
• MessagePack
• msgpack-numpy
• Plotly
• PyTorch
• PyZMQ

To install all dependencies with Anaconda run conda env create -f environment.yml and use source activate dista3c to activate the environment.

Acknowledgements

• @ikostrikov for pytorch-a3c

References

[1] Asynchronous Methods for Deep Reinforcement Learning
[2] High-Dimensional Continuous Control Using Generalized Advantage Estimation