MNIST with Kubeflow

first time workflow

Posted on November 14, 2020

With microk8s setup, I was ready to try out first time workflow based upon the venerable MNIST image set.

Creating a Notebook Server

First off, I needed to launch a Jupyter Notebook on the Kubernetes cluster. Following this tutorial,

Notebook Create

After a minute or so, the notebook shows as ready:

Notebook Created

Clicking Connect takes opens the new Notebook:

Notebook Opened

MNIST Sample Project

To verify the Notebook server is working, I tried a created Python3 workbook and pasted some sample code:

Sample Notebook

Sample Run

Ignoring the deprecation errors, this sample downloaded a MNIST data set and ran a gradient descent training and output the accuracy value. In my case: 0.9002.

More in this series…