chainer-docker is a template for a typical machine learning project using chainer, docker, and docker-compose.

• docker >= 17.12.0
• docker-compose >= 1.19.0
• nvidia-docker2 >= 2.0.2

Register the required environment variables to your .bashrc:

export UID
export GID=`id -g`

Register the required environment variables to .env:

• IMAGENAME: Name of docker image for this project (default: chainer-docker)

And set the name of this package in setup.py:

...

setup(
    name='name of package',

...

$ docker-compose build cpu

You can run a command in a new container:

$ docker-compose run cpu [COMMAND]

$ docker-compose build gpu

Check if a GPU is available in the container by running nvidia-smi:

$ docker-compose run gpu nvidia-smi

And then you can run a command in a new container:

$ docker-compose run gpu [COMMAND]

Start up Jupyter Lab using:

$ docker-compose up jupyter

You will get the response as follows:

...
Copy/paste this URL into your browser when you connect for the first time,
    to login with a token:
        http://0.0.0.0:8888/?token=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
...

Then you can access jupyter lab from the follwing URL:

http://0.0.0.0:8888/?token=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Note that you can change the port using environment variable as below:

$ PORT=9999 docker-compose up jupyter