Welcome to our Designing and Implementing a Data Science Solution on Azure course. This course focuses on preparing you for the DP-100 exam. Passing the DP-100 exam will earn you the Microsoft Azure Data Scientist Associate certification.

My name is Hallam Webber, and I have over 25 years of experience in the IT industry, with the majority of that time working with Microsoft and Azure technologies. Along with my colleague Guy Hummel and our lab designers, we have created lessons and labs that cover the topics tested in the DP-100 exam.

The exam covers four main topic areas: Design and prepare a machine learning solution. Explore data and train models. Prepare a model for deployment and Deploy and retrain a model. Let’s look at what each of these topics’ entails.

Design and prepare a machine-learning solution focuses on how Azure supports machine learning in terms of infrastructure and how you interact with the machine-learning service. Within this topic, we look at the types and scale of computing resources available for model training and inference. Inference is when a trained model is deployed for end-users. We see how the type of model being trained impacts computing resource choice along with the method and tools used to develop the model. We are introduced to the Azure machine-learning workspace, the context in which all elements of a machine-learning project are brought together. Data is an essential ingredient of model training, and we learn how to bring data into a workspace from different sources.

Explore data and train models introduces us to the Azure Machine Learning Studio, where models can be easily trained using the drag-and-drop graphical UI. This involves connecting components that represent data, algorithms, and result evaluations. We look at using AutoML, where Azure machine learning uses advanced trial and error techniques to determine the best machine learning approach for the training data. Both of these methods are convenient wrappers around software scripts. Scripting is more complex but offers the greatest flexibility and control. We see how to use scripts and the Azure ML software development kit with Jupyter notebooks to explore data and run experiments.

The Prepare a Model for Deployment section looks at how to package scripts as a job. So far, we’ve been interacting with data and model training through notebooks and the machine learning studio. This is okay when experimenting with small amounts of data but impractical for many training scenarios. This section looks at creating jobs and pipelines to streamline model training. We also see how to create custom components to be used in the drag-and-drop ML studio interface.

Deploy and retrain a model is about how to make a trained model available to end-users for inference and retrain it based on changed or additional data. Models can be deployed to online endpoints, as in a web service, offering real-time responses. Alternatively, batch deployment is best suited to models that meet at least one of the following attributes. The model takes in a large amount of input data, the model processing can’t or doesn’t have to be done in real-time, and you want to allow users to submit multiple prediction requests at a time.

This course assumes you have some experience using Microsoft Azure. If you’re new to using Azure, please take our Overview of Azure Services lesson.

DP-100 Exam Prep Introduction

About