Hubzero 2.4

Tools

Using Python packages from Jupyter notebooks

Which Python packages a notebook can import depends on the kernel it runs under. A kernel is a prepared environment installed on the hub, and a notebook remembers the one you choose, so the choice travels with the notebook when it becomes a tool.

Kernels

The hub installs Python packages into conda environments and registers each environment as a Jupyter kernel. Selecting a kernel points the notebook at that environment, and its packages become importable.

You cannot install a kernel yourself: creating an environment and registering it is an administrator's job, done on the execution hosts. If the packages you need are not in any kernel, open a support ticket on your hub and ask for them. Say which packages, which versions if it matters, and what you are building. The procedure the administrator follows is Jupyter notebooks in the administrators section; reading it tells you what you are asking for, and why adding a package to a shared environment is not always the answer.

Selecting a kernel

For a new notebook, pick the kernel when you create the notebook: the launcher lists one entry per kernel the hub has installed. Save the notebook afterwards, so the choice is recorded in it.

For a notebook that already exists:

  1. Open it in the hub's Jupyter tool.
  2. Shut the running kernel down first — Kernel > Shutdown.
  3. Choose Kernel > Change kernel and pick the one you want.
  4. Check the kernel name shown in the notebook's corner. It should be the one you chose.
  5. Save the notebook.

The exact menu wording follows whichever Jupyter version your hub deploys.

Finding out what a kernel contains

Two ways: read the environment file the administrator built the kernel from, or ask conda directly. For the second, start the hub's workspace tool, open a terminal, and put the hub's Anaconda installation on your path:

use anaconda-X

X is the version your hub deploys; use |& grep anaconda lists what is installed. Then list the environments:

conda info --envs

List the packages in the environment currently active:

conda list

Or in any other environment:

conda list -n <envname>

To capture an environment as a file — useful for a ticket, or for recording what your tool was built against:

conda activate <envname>
conda env export > <envname>.yml

Further reading

Reviewed against 2.4-main @ e097e0236d on 2026-09-10.