RELAAX_demo.md

Quick start

We recommended you use an isolated Python environment to run RELAAX. Virtualenv or Anaconda are examples. If you're using the system's python environment, you may need to run pip install commands with sudo. On OSX / macOS, we recommend using Homebrew to install a current python version.

• Install Docker

• Clone RELAAX repo.

git clone git@github.com:deeplearninc/relaax.git

• Install RELAAX

cd relaax
pip install -e .

• Build DA3C bridge.

algorithms/da3c/bridge/bridge.sh

• Install TensorFlow

• Create training directory

cd ..
mkdir training
cd training

• Pull DeepMind Lab demo docker

docker pull 4skynet/lab

• If you are going to use trained checkpoint downloaded it from here: https://s3.amazonaws.com/dl-checkpoints/lab_demo_checkpoints.tar.gz. Unpack it to training directory.

• Open new terminal window, navigate to training directory and run honcho:

honcho -f ../relaax/config/da3c_lab_demo.Procfile start

• Use ifconfig command to find IP of your localhost. Remember it.

• Open new terminal window and run environment inside docker image. Use sudo if needed.

docker run -ti -p 6080:6080 4skynet/lab LOCALHOST_IP

Replacing LOCALHOST_IP with your localhost IP from above.

• Open http://127.0.0.1:6080/vnc.html URL in browser. You will see web form to enter your credentials. Leave all fields intact and press 'Connect'. You will see running game.

• Browse TensorBoard output using http://localhost:6006 URL.