Installing Unity ml agents on Linux

Test environment

OS: Ubuntu 16.04 / x86_64
python version: 3.6
pytorch : 0.4.0
cuda : 9.0
cudnn : 7.0.5 for cuda 9.0

Installation manual for Linux

1. Make and activate virtual environment

replace pgtravel with your favor name.

$ virtualenv pgtravel -p python3.6
$ source pgtravel/bin/activate

2. Clone Unity ml-agent repository and install it

$ git clone https://github.com/Unity-Technologies/ml-agents
$ cd ml-agents/python
$ pip install .

3. Install pytorch

3.1. Install pytorch for GPU users

Download and install cuda & cudnn

• Link to the official homepage for cuda 9.0
  • https://developer.nvidia.com/cuda-90-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=deblocal
• Link to the official homepage for cudnn 7.0.5
  • You need Nvidia account for downloading cudnn
  • https://developer.nvidia.com/rdp/cudnn-archive

Download and install pytorch for gpu

pip3 install http://download.pytorch.org/whl/cu90/torch-0.4.0-cp36-cp36m-linux_x86_64.whl
pip3 install torchvision

3.2. Install pytorch for CPU users

pip3 install http://download.pytorch.org/whl/cpu/torch-0.4.0-cp36-cp36m-linux_x86_64.whl 
pip3 install torchvision

3.3 Install pytorch for other platforms

Refer installation manual for other platforms from the pytorch official homepage

4. Clone pg_travel repository and install the prerequisite

Note that the prerequisites other than tensorboardX are installed when installing Unity ml-agents

$ git clone https://github.com/reinforcement-learning-kr/pg_travel
$ pip install tensorboardX