Override the kernel

[henryiii]

How to I override the kernel? I'm using nb_conda_kernels to access my conda environments as kernels, and jupyterbook doesn't seem to be triggering that mechanism, causing it to simply report it can't find the kernel. I've tried to override it:

sphinx:
  config:
    nbsphinx_kernel_name: python3

But it doesn't seem to be picking it up, it's still trying the kernel in the notebooks.

Also, related, any plans for a conda-forge package?

[choldgraf]

@henryiii - I'm a bit confused, you want to manually over-ride the kernel that is already specified in the notebook? Is this because there's a different kernel on a build system vs. your computer or something? And just a note that nbsphinx is a totally different project so its config won't work here.

for the conda-forge question, which package are you talking about?

[henryiii]

The problem I'm facing is when using conda, there's a package nb_conda_kernels which exposes all the conda environments as IPython kernels without having to setup user installs for each one. However, it only works in the notebook (or lab), not for nbconvert, nbsphinx (which for some reason I thought I saw in the docs was being used in the backend, sorry), or here. So, if I have a regular kernel install, I need to override it. For nbconvert, you can do that, but the functionality seems to be missing here. It's also useful for CI builds, which may not have the same kernel name; I usually do add it as an override for things that build in CI, so that someone opening a notebook with a different kernel local to them and saving it doesn't break the build.

I've currently worked around the problem by manually installing the kernel with exactly the same name that nb_conda_kernels gives it. Obviously, this would have to be done on each project/environment. It also makes two copies of the environment show up in the notebook.

Are there plans to make a juptyerbook feedstock on conda-forge? Conda is a fantastic place to put together scientific Jupyter environments, so being able to natively add jupyterbook to an environment.yaml would be nice. It works with the pip: workaround, but it's more natural to have a package. I can help a bit with setting it up if you need help.

[choldgraf]

ah that does seem to be a reasonable solution!

re: conda forge, there actually is a PR to get this on conda forge right now but it is stalled: conda-forge/staged-recipes#12516 we'd love to see this up there too 👍

[joelostblom]

I added a feature request for this here jupyter-book/jupyter-book#1348