Overview

This article describes adding your mamba environments to Jupyter Notebook to use as a kernel. You can learn about creating mamba environments in Managing Python Modules Through the Mamba Environment Manager.

Use the Public Kernels

The public jupyter kernel can be directly used without installation. They are generated from the public mamba environments. Everyone can use them, but they cannot be modified (e.g., adding packages). Here are two ways to find them inside the jupyter session on the web portal:

The first option: initiate a new notebook directly

The second option: choose a kernel for an existing notebook file

Creating Your Own Private Kernel

The following command will create a jupyter kernel from an existing mamba environment:

Do not activate the environment when running this command.

$ mkjupy <environment> "Icon Label"

Example (do not activate the nobel environment):
$ mkjupy nobel "nobel"

Expected output example:
$ Installed kernelspec nobel in /home/spock/.local/share/jupyter/kernels/nobel

This takes the environment called nobel and creates a Jupyter kernel for it. When creating a new notebook, it will have an Icon labeled nobel on the launcher page.

If you need to add more packages, the steps are:

Open a shell/terminal, connect to Sol/Phx, request an interactive session
Load the mamba module, activate the nobel environment
Add packages to the nobel environment
Deactivate the nobel environment
Use the mkjupy command again to re-create the nobel kernel

Customizing logos is possible but not currently documented due to potential complexity. If interested, please come by office hours!

Removing a Kernel from the Web App

From the terminal, execute the following commands (once again assuming a fresh shell session). This will only remove the kernel entry in Jupyter and does not affect the environment. Besides the example below, an alternative way is to delete the kernel folder directly.

$ module load mamba/latest  
$ jupyter kernelspec remove nobel
Kernel specs to remove:
  nobel                 	/home/jyalim/.local/share/jupyter/kernels/nobel
Remove 1 kernel specs [y/N]: y
[RemoveKernelSpec] Removed /home/jyalim/.local/share/jupyter/kernels/nobel

Remove the kernel by the environment name, not the Jupyter kernel title.

Creating Kernels from a Data Directory

Since the command above only works for environments installed in the user’s home, to make a kernel from environments installed in /data directory:

If needed, re-create the env in this /data directory, then each user in your group can access this env and make a kernel from it.
Activate the env
Once the env is activated, install "ipykernel" with mamba: mamba install -c anaconda ipykernel
Then each user can make a kernel from this env for themselves:
1. They need to activate this env first by source activate ENV
2. python -m ipykernel install --user --name=<exact name of the env>

Known Bug

The Python package Plotly is known to have some incompatibility issues with Jupyter version 4. Some plotting functions don’t work properly inside Jupyter Notebook, for example, the FigureWidget function from Plotly. Currently, we don’t have a good solution, and we recommend exploring Matplotlib and Seaborn for your plotting needs. Please note that plotly, matplotlib and seaborn are installed on all of the public mamba environments and their corresponding public jupyter kernels.

Research Computing

Preparing Python Environments for Jupyter