Jupyter Notebooks and Virtual Environments

Jupyter notebooks are the defacto standard for exploratory data analysis. With increasingly fucked-up dependency management it makes sense to use virtual environments and/or docker to manage dependencies per project. However, it makes more and more cumbersome to use it with jupyter notebooks. With conda, the nb_conda package helps, but virtualenvs need to be registered separately.

0.) Install jupyter, if missing, and ipykernel for the default user

pip3 install --user --upgrade jupyter ipykernel

1.) Create a venv

python3 -m venv `~/venvs/this_env`

2.) Install jupyter and ipykernel inside the venv

source ~/venvs/this_env/bin/activate
pip install  --upgrade jupyter ipykernel

# install other packages

3.) Register venv

# while inside activate `this_env`
python3 -m ipykernel install --user --name this_env

4.) Launch a jupyter notebook from outside this_env and the kernel should appear as selectable