With increasing focus on Machine Learning systems in almost every business area, it is important to build a great pipeline to train, test and deploy your models. In this session we will show a way to do that with Jupyter and Azure. The session will cover the following topics:

• creating a simple image classification service without coding
• creating a PyTorch model from scratch
• Training and Testing the PyTorch model
• Saving the model locally and on a cloud storage (with self-made versioning)
• evaluating multiple models
• deploying an API (via Docker) to get predictions from the model
• using DevOps to update the API

Notes:

Please bring your own device, as we will be running the workshop on individual machines.

To prepare:

• make sure you have a Python environment >= 3.5 installed (preferred using Anaconda)
• we will be using Azure, so make sure you have a Microsoft account - no Azure registration needed beforehand
• optional: Install Docker on your machine, install VS Code, install Azure Storage Explorer

Daniel Heinze

Affiliation: Microsoft

Daniel is a Data Engineer at Microsoft, working with customers to create services that get insights from data and take that to improve the system through Machine Learning.

visit the speaker at: Github