I grew up tired of using VS Code to edit Jupyter notebooks given all the bullshit around AI and their push to include it anywhere it has no place to be. And I didn’t want to launch a terminal and everything everytime. So I bought a 16Gb Raspberry Pi and started setting up a server.

First try: via YunoHost

I had been using YunoHost install in the past few months for some simple projects and playing around some server-based applications. Good news is, Jupyter is listed in their application catalog. Bad news, the install packages has been broken for the past 6 months. So no easy install but I raised an issue and switch to manual install.

Second try: manual install

Thankfully, there are a lot of resources on the topic. I ended using mostly this one and this one.

An hour later, I had a running Jupyter server running with the python kernel. A quick nohup later and it was done.

Additional kernel: Julia

Installing Julia on the RPi5 was straight forward. Following up the install procedure for Linux or Debian and done. Then it’s just a matter of installing IJulia via the Julia package manager and voilà, Jupyter has its new kernel.

No issue installing packages and I played with Agents.jl models just to get back on track.

Additional kernel: R

Installing the R kernel started easily, using the stand procedure for Debian.

My first project failed lamentably though. And from the get-go. Impossible to install the tidyverse package. And it was late. I didn’t want to dive into pages and pages of compiler error log. So I call it quit, kinda deciding to let go and not use R on my brand new Jupyter server.

And a few days later, when I wanted to do a quick visualisation, I remembered all that, once again called it quit and launched RStudio. ‘Cause who wants to spend hours parsing through error logs when all you want to do is a visualisation taking you 5 minutes top?

But it bothered me. And the next day I decided to bite the bullet and deal with that.

My first step was to check whether my R install was correct. This walkthrough, even though it was written about the RPi 4 had all the information I need. My issue was not with R itself but on the many system libraries the R packages depends on and which do not come by default on the Pi.

Installing the ones listed in the blog post then cross-checking what was actually missing and listed in the error log. And problem solved.

Next step was installing tidyverse and it worked.

Then same process again for caret and sf. No issue with the first, but installing the required projection libraries for sf took some looking around.

Next steps

Beyond just running this Jupyter server, this small project has guided me to the whole homelabing thing and I am also thinking about how to train AI models on clusters and all. A lot of things to learn, not enough time.