Enabling user-created conda kernels in Jupyter

[rsignell-usgs]

It would be great to allow HydroShare JupyterHub users to create their own conda environments and to enable these to appear in the list of available kernels in Jupyter.

I opened a terminal edited my $HOME/.condarc to specify an additional directory for environments:

channels:
  - odm2
  - landlab
  - conda-forge
  - defaults
envs_dirs:
  - /home/jovyan/my-conda-envs/

then created a new environment widgets.

It shows up when I do conda info -e:

but it doesn't appear in the kernel list in Jupyter:

I think you just need to add nb_conda_kernels to the root conda environment so that custom environments created by users will appear.

[rsignell-usgs]

If you are using repo2docker as described here you should be able to just add nb_conda_kernels to the environment.yml like:

$more environment.yml

name: root
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.6
  - jupyterhub
...
  - ulmo
  - nb_conda_kernels

See also: http://repo2docker.readthedocs.io/en/latest/usage.html#using-repo2docker-with-a-jupyterhub

Is that about right @yuvipanda?

[Castronova]

@rsignell-usgs I'll see if can implement this in the next JH release.