Productionisation of Machine Learning is the hardest problem in data science.
That is a bold statement and one which often surprises people. When you think about Data Science, you could assume the hardest problem would be selecting an appropriate algorithm, obtaining and cleaning data or possibly Deep Learning. All these are problems and hard problems, but they are well solved in machine learning. What is not solved is what you do with a model once it has been trained.
This course is designed to take a data scientist from concept to Production in just 2 days. You will leave this course with all the skills you need to take a model and put it in production. The course will be delivered by Terry McCann, Artificial Intelligence MVP (Microsoft MVP). Terry holds an MSc in Machine Learning with a focus on DataOps. Terry is recognised for his ability to convert deep technical material in to bite sized understandable chunks.
- A basic level of python will help, but is not required.
- A laptop with a subscription to Azure
Pick up a book on Machine learning and it will explain the process for machine learning, many citing CRISP-DM as the ideal process. CRISP-DM is an iterative approach to Data Mining. It starts with business understanding the flows to data understanding, data preparation, modelling, evaluation, then either loops back around or your model is deployed. How it is deployed, well no one ever tells you that! It is this problem this course solves. If you look at the image in the top you will see the vast array of options available. With each comes more complexity and sophistication.
In this two-day course we will build a series of basic models and promote them into production, using a variety of techniques in Azure. The biggest problem a lot of customers face is not how to build a machine learning model, but it which technology they should use. This question depends on so much and as a result there are multiple options in Azure. This session will demonstrate the various options for deploying a model.
We will end the two days with an example which will allow you to deploy and model, in any language using Docker, Python and Kubernetes.
Machine learning in Azure
- Exploration services
- Data Science Virtual Machine
- Deep Learning Virtual Machine
- Azure Notebooks
- MLFlow & AML tracking
- Batch Machine Learning
- An overview of batch Machine Learning
- Batch in SQL Server 2019 with Python
- Batch in Azure SQL DB with R
- Batch in Azure Databricks with Spark ML
- Interactive Machine Learning
- Azure Machine Learning Studio
- Azure Machine Learning Services
- Docker & Kubernetes
- Interactive Batch Machine Learning
Creating your first Machine Learning model
- Exploratory Data Analysis
- Working in Notebooks
- Creating a model in Python (regression model)
- An introduction to Python
- Machine Learning in Python
- Understand Python dependencies
- How to share models
- Basics of Serialisation in Python
- Introduction to Pickle
Creating restful APIs
- Learning from Software Engineering
- Introduction to RESTful APIs
- Options to build a REST API in Python
- Options in other languages
An introduction to Docker
- A short history of application deployment
- Docker basics
- Creating an application in Docker
- Docker for Machine Learning
- Containers & Images
- Creating the right image for your model
- Container registry
- Images and tagging
- Docker compose
An introduction to Kubernetes
- A short history of Kubernetes
- Kubernetes basics
- Creating a Kubernetes Manifest
- Options for Kubernetes in Azure
- Deploying your model to Azure Kubernetes Service
DevOps for Machine Learning
- Automating the deployment of models in Azure