Skip to content

Latest commit

 

History

History
90 lines (71 loc) · 2.88 KB

software_installation.md

File metadata and controls

90 lines (71 loc) · 2.88 KB
Raw
title nav_order
Software installation
4

User-defined kernels

The standard kernels in our JupyterHub setup include Python 3.6 and R 3.6.3, both with only few packages installed. These are sufficient for demonstration purposes, but may not contain specific packages that you may require. As a solution, you may set up your own kernel based on one of your local environments.

We will illustrate this process for both a MiniConda-installed Python and R environment. Please consult the VscDocumentation page on Python package management or the VscDocumentation page on R package management for MiniConda installation instructions.

Environment creation: Python

We will create a MiniConda environment with the following Python packages:

  • python 3.9
  • numpy
  • scipy
  • pandas
  • tensorflow-gpu
  • ipykernel

NOTE: Instead of creating a new environment you can also introduce an existing environment to your JupyterHub. In this case you can skip the creation of the environment. NOTE: Always include the ipykernel package in your environment, as this is necessary to install the kernel.

  1. Create a new Conda environment (here named p39env):

    conda create -n p39env python=3.9 numpy scipy pandas tensorflow-gpu ipykernel

  2. Activate the environment:

    conda activate p39env

  3. Install the associated IPython kernel in ${VSC_HOME}/.local so that JupyterHub can find it:

    python -m ipykernel install  --prefix=${VSC_HOME}/.local/ --name 'p39env'

  4. Connect to JupyterHub and verify that the new 'p39env' kernel appears. It should look like this:

  5. Test the environment and make sure that it works as expected.

Environment creation: R

For R, let's create an environment with following packages, and let's name it r41env:

  • R 4.1
  • data.table
  • ggplot2
  • factoextra
  • irkernel
  • jupyter_client

NOTE: Instead of creating a new environment you can also introduce an existing environment to your JupyterHub. In this case you can skip the creation of the environment. NOTE: Always include the jupyter_client and the irkernel package, to be able to install the kernel.

  1. Create the new R environment:
conda create -n r41env r-base=4.1 r-data.table r-ggplot2 r-factoextra r-irkernel jupyter_client -c conda-forge
  1. Activate the environment:
conda activate r41env
  1. Install the kernel in $VSC_HOME/.local to be able to use it in JupyterHub:
Rscript -e 'IRkernel::installspec(prefix="${VSC_HOME}/.local/", name="r41env", displayname="r41env")'
  1. Connect to JupyterHub and verify if you can find your newly created R environment in the kernel list.