Conda with Jupyter Setup

2024-08-08 – 2025-01-29

When working with Jupyter Notebook/Lab I found Conda to be one of the best ways to manage my virtual environments.

Python and package version differences can be breaking and so it is essential to have some method of managing different environments. My only issue with Conda is that the default install and settings seem suboptimal and incomplete so every time I need to set Conda up I need to search through multiple different resources. I hope this article could be an answer to all those questions for anyone else having the same problems as I was.

While I highly recommend using Conda this article also provides instructions for virtualenv in the rare cases when it is more convenient.

Install Conda

At the time of writing, Anaconda has recently changed their licensing terms regarding the usage of their defaults channel to require organizations (including non-profit and government) with over 200 employees to purchase a paid license. Educational institutions are exempt, if their usage is for ‘‘curriculum-based courses’’.

These changes only apply to the defaults channel and the Anaconda Distribution (as it comes with a bunch of packages from the defaults channel). Conda itself is still free and unless you have a good reason to I would recommend moving over exclusively to the conda-forge channel (not only is it free but it has more packages and is usually more up to date). Just note that conda-forge is mainly hosted by Anaconda and you will probably only want the defaults if you work in an industry that is more strict on security and stability.

Miniforge is a minimal install for Conda with the following features:

wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh" -O miniforge.sh
bash miniforge.sh

Then follow the installation prompts (I usually install to /home/<user>/.conda) and after you are done feel free to delete the installer with rm miniforge.sh.

Close and re-open your terminal window, then if you are getting a Command Not Found when running conda you might need to run

/home/<user>/.conda/bin/conda init

Anaconda or Miniconda

I would recommend Anaconda if you are new to Python or Conda and Miniconda if you know a what you are doing as you can just install what you actually need later.

Whichever you download run

bash <conda-installer>.sh

Then follow the installation prompts (I usually install to /home/<user>/.conda) and after you are done feel free to delete the installer with rm <conda-installer>.sh.

Close and re-open your terminal window, then if you are getting a Command Not Found when running conda you might need to run

/home/<user>/.conda/bin/conda init

Install Jupyter

The most basic method is to simply install Jupyter to every Conda environment and to activate that specific environment before running Jupyter every time.

The better method is to have Jupyter dynamically find the kernels in each Conda environment so we only need one install of Jupyter and can switch kernels as we please.

Method 1: Jupyter outside Conda

If you want to be able to run Jupyter without activating any conda environments you can use your package manager to install Jupyter then use the following instructions to create an expose your an environment.

conda create -n myenv ipykernel
conda activate myenv
python -m ipykernel install --user --name myenv --display-name "Python (myenv)"

sudo apt install python3-jupyterlab-server
jupyterlab

Now in the kernel selection you should see myenv.

To see what kernels are available we can list them with

jupyter kernelspec list

To uninstall the kernel we can run

jupyter kernelspec uninstall myenv

Using nb_conda_kernels Jupyter can see the kernels in the other Conda environments (I personally like using base as my notebook env).

conda create -n notebook_env nb_conda_kernels jupyterlab
conda create -n myenv ipykernel

conda activate notebook_env
jupyterlab

Now in the kernel selection you should see myenv.

To see what kernels are available you can list them with

python -m nb_conda_kernels list

To uninstall a kernel we just need to remove the ipykernel from the environment.

Other

Conda Tips

Don’t start conda when logging into the shell

conda config --set auto_activate_base false

List all explictly installed packages to conda (does not include packages installed via pip inside the env)

conda env export --from-history -n myenv

Delete a conda env

conda remove -n myenv --all

Virtualenv

If you prefer to use Virtualenv/venv you can install it for a certain version of Python (in this case Python 3.9) and create an venv by running.

sudo apt install python3.9 python3.9-venv
python3.9 -m venv venv
source venv/bin/activate

Note that 3.9 can be omitted if the version of Python doesn’t matter. Now we can test that are are indeed in the venv by checking which executable we are running when we type python or pip.

which python
# /absolute/path/to/venv/bin/python
which pip
# /absolute/path/to/venv/bin/pip