IPython startup script examples needed

[msukmanowsky]

The Jupyter notebook migration documentation does specify that IPython profiles are no longer supported, but that doesn't help users like myself looking to ensure notebook kernels run startup scripts normally located in the ~/.ipython/profile_<name>/ipython/startup/* directory.

Would be helpful to get a response on this thread as well as some more in depth documentation on migrating aspects like these to Jupyter.

[minrk]

There is no migration to do regarding IPython startup files. The key point that appears to be missing is that IPython continues to have profiles, which have not changed. The migration docs are to convey the fact that that IPython profiles no longer affect Jupyter projects. That is, the Jupyter notebook config file has moved from the IPython profile to ~/.jupyter/jupyter_notebook_config.py. The IPython profile and its startup files continue to affect the IPython kernel, and have not changed.

[msukmanowsky]

Thanks @minrk! I think where my confusion is now is in how to create a new IPython-based Kernel for Jupyter that runs with a specified IPython profile.

[minrk]

If the kernel should always run with a specific profile, you can hardcode that in your Application code, rather than allowing it to be configurable. Otherwise, you can add the --profile foo to the argv in the kernelspec. This is how you would specify a profile for the existing IPython kernel.

[msukmanowsky]

To provide a concrete example to @minrk's instructions for anyone struggling with this like I was.

IPython searches for kernel specifications in one of three spots:

Unix:

• /usr/share/jupyter/kernels
• /usr/local/share/jupyter/kernels
• ~/.ipython/kernels

Windows:

• %PROGRAMDATA%\jupyter\kernels

A kernel spec is nothing more than a folder with a kernel.json file that specifies how to launch that kernel. If I wanted to use a specific IPython profile (named my-profile), I could do something like:

mkdir -p ~/.ipython/kernels/my-custom-kernel
cat > ~/.ipython/kernels/my-custom-kernel/kernel.json <<EOF
{
    "display_name": "My Custom Python Kernel",
    "language": "python",
    "argv": [
        "python",
        "-m", "ipykernel",
        "--profile=my-profile",
        "-f", "{connection_file}"
    ],
    "env": {
    }
}
EOF

When launching a Jupyter notebook server, you'll now have the choice of launching either a Python 2 kernel or a My Custom Python Kernel kernel.

[orenshk]

@minrk wrote:

...that IPython profiles no longer affect Jupyter projects.

But they do. For example, adding exec lines to c.InteractiveShellApp runs them in the notebook.

[takluyver]

For Python notebooks, Jupyter is running an IPython kernel, which still uses the IPython config and startup files. Jupyter applications, like the notebook user interface, do not use the IPython config files.

[takluyver]

Closing this as I don't see anything concrete to do.

[Jeitan]

Can I request to reopen this and actually make an addition to the documentation as the original post asked?

See, I totally understand that Jupyter does not do profiles anymore. That is fine, no problem. However, IPython does still use profiles (http://ipython.readthedocs.io/en/stable/config/intro.html), and I use them when working both in a standard console (not a Qtconsole) and in a Jupyter notebook (running an IPython kernel), and I would like to use them in the IPython kernel that Jupyter runs.

I get that Jupyter doesn't use profiles for any of its own configuration, but what msukmanowsky posted above is the only place I've been able to find after lots of searching that tells me how to make Jupyter tell IPython to use a particular profile instead of just the default profile.

It would be nice if this could be documented in the Jupyter documentation, somehow. Even if not the exact example above, make it clear that in order to tell IPython to use a profile, you have to construct another kernelspec. I don't think this is obvious.

[avishapiro]

I'll second @Jeitan 's request to reopen this because there are still dots that are hard for a novice to connect and also since kernels are now a part of Jupyter and not Ipython (https://jupyter-client.readthedocs.io/en/latest/kernels.html#kernels). Try ipython kernelspec list to see.

I tried to follow @msukmanowsky 's example to add something like "--profile=my-profile" to my kernel.json created when I created a conda env. It is located at ~/anaconda/envs/<env-name>/share/jupyter/kernels/python3. However, when launching a Jupyter notebook server and a notebook, and selecting the desired kernel, it did not appear that the profile was activated or that any startup files in ~/.ipython/<profile_name>/ipython/startup/* were run. I believe the right way to do this currently is using ~/.jupyter/jupyter_notebook_config.py, but I couldn't get code there to run either.

What did work was to use a cell magic %run ~/.ipython/profile_default/startup/startup.ipy.

[takluyver]

jupyter_notebook_config.py is run when the notebook server starts, not when a kernel starts. The kernel startup is up to IPython (for Python kernels).