Light(er) weight Jupyter Notebook docker images without conda.
Differences from Jupyter's docker stacks project:
- No Conda (why have Conda if I have Docker?).
- Lighter and smaller (Based on Python:3.8-slim, scipy is < 1GB, pyspark is 2.7GB).
- No annoying
jovyan
user that you always mispell and never notice (instead, the user isjupyter
).
Hosted on DockerHub, you can find images here.
- dockerbooks/base Base notebook, nothing but Jupyter.
- dockerbooks/scipy Extends base. Pandas/Numpy/Numba/Scikit-learn/Seaborn.
- dockerbooks/pyspark Extends scipy. Pyspark (Change the version in one line!).
User is jupyter
and work directory for mounting:
-v "$PWD":/home/jupyter/work
Install extra python packages via pip with --user
option:
pip install --user foo
By default, the notebook runs as the jupyter
user with UID/GID 1000
. To change this, you should build your own. For example, build the image in the /base-scipy folder as "base-scipy" and then:
docker build --build-arg="BASE=base-scipy" --build-arg="USERID=1000" --build-arg="GROUPID=1000" -t dockerbooks/scipy .