search

rll / rllab

rllab is a framework for developing and evaluating reinforcement learning algorithms, fully compatible with OpenAI Gym.
Other
2.89k stars 803 forks source link

Docker intended running environment #244

Open vsoch opened 6 years ago

vsoch commented 6 years ago

heyo! We are building the Dockerfile to run on a GPU node, and running into issue when we try to use with GPU:

https://github.com/researchapps/sherlock/tree/master/rllab#testing-with-exo

The run with CPU actually works (sort of) ok, but for GPU there is verbose output about the version of theano being too new. When we load the suggested version, another (larger) error spits out.

I'm wondering:

Or if you know, can you share the libraries that you have loaded that everything works on? Have you tested with GPU? The only thing we've changed so far is the location of the install from /root to /opt so that when used in a Singularity container, the user can see it. Thanks muchly!

vsoch commented 6 years ago

I also noticed in the environment.yml the system platform is specified to be darwin?

        - https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-1.0.1-py3-none-any.whl; sys_platform == 'darwin'

but we are installing in a linux container?

vsoch commented 6 years ago

It looks like the dependency is ignored given linux, so I'm guessing this is there to support (also) a local install on a Mac.