IntelLabs / coach

Reinforcement Learning Coach by Intel AI Lab enables easy experimentation with state of the art Reinforcement Learning algorithms
https://intellabs.github.io/coach/
Apache License 2.0
2.32k stars 459 forks source link

Enable multi-process training with multi-node Coach. #240

Open balajismaniam opened 5 years ago

balajismaniam commented 5 years ago

Tested with the following presets on single-node and multi-node Coach: CartPole_ClippedPPO Mujoco_ClippedPPO with inverted_pendulum level

gal-leibovich commented 5 years ago

Sorry for taking so long to getting to review this PR. This PR leaves out mxnet from using multi-process training with multi-node coach, since it uses distributed tensorflow (which we intend to remove support for) for implementing distributed multi-training. Since we plan to remove future support for distributed tensorflow, I am not sure that this is the way to go.