NIPS2017 - Learning to Run Challenge --->
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


A keras solution for 2nd place NIPS RL 2017 challenge.

There is a slide, a lecture and a writeup(arxiv) about our work.

To Run


These instructions expect that opensim-rl conda environment is already setup as described in : .

$ source activate opensim-rl

Other dependencies is needed as follow

  • Keras(since old version does not support selu activation)
  • TensorFlow
  • matplotlib
  • numpy
  • Pyro4
  • parse
  • pymsgbox(optional)


This version requires farming, before starting, you should first start some farms by running python on each SLAVE machine you own. Then create a in the working directory (on the HOST machine) with the following content :

farmlist_base = [('', 4), ('', 8)]

# a farm of 4 cores is available on localhost, while a farm of 8 is available on another machine.

# expand the list if you have more machines.

# this file will be consumed by the host to find the slaves.

Try python --help to get more information about how to set the environment.

More information can be found in .

Thanks to @ctmakro for providing us with this frame.


Test the model in parallel and calculate the average score.

We provide you with some trained parameters .

python -a=10 -c=5 -t=200 -p logs

# test the model for 200 times with 10 actor networks and 5 critic networks ensemble

# the network parameters should be placed as logs/actormodel1.h5 ... logs/actormodel10.h5

Try python --help to get more information .