You need to install docker, nvidia-docker, docker-compose, and nvidia-docker-compose.

Before running nvidia-docker-compose, you should run nvidia-docker at least for one time to let it create an nvidia docker volume. For example:

  nvidia-docker run --rm nvidia/cuda nvidia-smi

If you must use sudo to run docker, and that PATH does not contain binaries that you need to run docker, define this alias

  alias sudo='sudo env PATH=$PATH'

This repository contains several bash scripts that I use to run and control dockers

• rl_start.sh and tf_start.sh: start a docker image with port mapping, password, and directory mount
• R_start.sh and R_studio_start.sh: start an R jupyter docker image, and R_studio docker image
• execBashDocker.sh: open bash shell on a running docker process
• cleanDocker.sh: clean temporary docker files
• tensorboard_start.sh: list all folders that seems to be a tensorflow output folder, and let user choose which to run tensorboard on