I recently set up a headless RaspberryPi 3 to act as a Jupyter Notebook server on my local network. I needed an easy and quick way to keep to rapidly do spots of data science work and practice from the command line and browser without having to boot up a separate machine. I’m loath to bung up my regular laptop with loads of anaconda packages that I won’t necessarily use, or worse get into the potentially messy business of a custom virtual environment install involving esoteric python libraries. I prefer to keep things separate and clean.
Very good instructions here for setting up, basic setup is a headless RPi3 running Raspbian Lite with virtualenv for the virtual environment to keep python installation nice and clean, and supervisor for the process control. Pretty soon you have a tab on your browser with a Jupyter Notebook always ready to go. Nice.
Installation of most of the basic data science packages such as pandas and matplotlib is fairly straightforward. Some heavy googling, head scratching for missing dependencies and pinning to at least one old version of a key library was necessary to get the more specialist geopandas package fully installed (yet to test how fast this will be, probably not very).
First experiences are good. Despite the obvious limitations in terms of the computational power of the RPi3, the machine actually runs code acceptably fast for analysing small to medium size datasets such as UK election data. I’ll put up some example work in due course.