Here are some issues I came across during setting up the Python environment. Ray only works with Python 3.7 so you should set up the environment like:
conda create -n forl python=3.7
conda activate forl
Some prerequisites to avoid version incompatibility, (don't install anything prior to this, otherwise you can start over):
- To be able to use that specific Ray version that they provided, you also need the following
pip install wheel==0.38.0 setuptools==65.5.0 pip==21
pip install protobuf==3.20.*
Now you can install the rest:
pip install -e .
pip install rl-warp-drive --no-dependencies
pip install gym==0.21
pip install tensorflow==1.14
pip install "ray[rllib]==0.8.4"
For Ray you need to uninstall this...:
pip uninstall pyarrow
Now you can import everything and it should work.
If you have issues installing pycuda
when installing rl-warp-drive
you need to issue sudo apt install nvidia-cuda-toolkit
(on linux for GPU)