Marlo environment on headless server
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
docker
install
README.md

README.md

Using these scripts, you can easily install and run multiple marlo environment on a headless server.

This script automatically set up jupyter notebook on server(port 5000). If you open port 5000, you can access jupyter notebook easily.

It worked on Google Cloud Ubuntu18.04 with and without GPU, but I guess any clowds may work fine.

What will be installed and setuped

  1. Anaconda
  2. marlo as a conda environment(You can switch it by source activate marlo)
  3. docker
  4. jupyter notebook on server
  5. pytorch, gym
  6. cuda(optional)
  7. jupyter notebook vim extension(optional)

How to use

1. Create Ubuntu18.04 instance with opened port 5000

2. Install marlo

git clone https://github.com/syuntoku14/marlo_headless.git
. ./marlo_headless/install/gcp/install.bash

Make sure to use . command to run in the current shell.

Now you can use:

source activate marlo

3. Launch Clients

./marlo_headless/docker/launch_multi_client.sh [number of environment]

The environments will be connected to ports 10000, 10001, ..., 9999 + [number of ports].

For example, "sh ./run_docker.sh 3" will run three Minecraft environment connected to port 10000, 10001, 10002.

Optional

Install CUDA

. ./marlo_headless/install/gcp/others/cuda_install.bash

jupyter-notebook vim extension

. ./install/gcp/others/vim_extension.bash