In my last post, I’ve installed a Jupyter Notebooks with TensorFlow support on a Raspberry Pi’s Kubernetes cluster. Today I will train and run a Convolutional Neural Network to recognize handwritten digits on it.
I’m using the MNIST dataset which is like the “Hello World” of CNN. …
Last time I’ve deployed CockroachDB into my Raspberry Pi’s Kubernetes cluster now it’s time for Jupyter Notebooks with TensorFlow support.
TensorFlow has support for Raspberry Pi running a 64 bits OS but they don’t publish Docker images for the arm64 platform so I decided to create one.
There are lots of articles and posts on how to install TensorFlow X.Y on Raspberry Z but most of them are for older versions of the library or the 32 bits version of the Raspberry and they don’t work with my configuration (Raspberry 4B + Ubuntu 20.04 + Python 3.8).
Installing the latest Ubuntu Server was super simple thanks to Raspberry Pi Imager you can check some instructions here: https://opensource.com/article/20/4/raspberry-pi-imager-mac
After reviewing different approaches to get K8s running on my “cluster” I’ve decided to go with Rancher’s K3S which is super easy to install, is lightweight, and works perfectly.
Hacker: Go, C++, Python, Js, NoSQL MindTraveler, Serial Griller/Asador Serial. a.k.a Gato