Hello and welcome to Designing and Implementing a Data Science Solution on Azure. The focus of this learning path is to prepare you for Microsoft’s DP-100 exam. If you pass the DP-100 exam, then you’ll earn the Microsoft Certified: Azure Data Scientist Associate certification.

My name’s Guy Hummel and I’m a Microsoft Certified Azure Solutions Architect and Data Scientist.

The DP-100 exam tests your knowledge of four subject areas: setting up an Azure Machine Learning workspace, running experiments and training models, optimizing and managing models, and deploying and consuming models. I’m not going to talk about every item in the exam guide, but I’ll go over some of the highlights of what you’ll need to know. 

Okay, the first section of the exam guide is about setting up an Azure Machine Learning workspace. This is just a place where you can put everything related to a machine learning project. Don’t worry if you’re not familiar with machine learning yet because our Introduction to Azure Machine Learning course explains the fundamentals.

One very important topic in this section of the exam guide is how to create datastores and datasets in a workspace so your machine learning experiments can ingest data from external sources. You also need to know how to create compute resources that will run your experiments.

The next section of the exam guide is about running experiments and training models. Microsoft provides two different ways to do this. One is a very user-friendly tool called the designer, which lets you create machine learning pipelines using a drag-and-drop interface. You don’t even have to write any code to use it.

The other way is to use the Azure Machine Learning SDK. With this method, you have to write code, so it’s more difficult to use, but it can do more than the designer interface can. One especially useful feature is the ability to automate the model training process.

The next section of the exam guide takes automation to another level with Azure AutoML. This service can automatically try different machine learning algorithms and options to find the optimal model to solve a given problem. You do still have to know how to configure it before running it, though.

A related service called Hyperdrive can automatically try different hyperparameter values to find the optimal values for training a machine learning model. Hyperparameters are a bit hard to explain if you’re not familiar with machine learning concepts yet, but basically they’re parameters that control how a model is trained rather than being parameters that are part of the model itself. Don’t worry if that doesn’t make sense yet.

Another topic in this section is how to use model explainers to interpret models. You may have heard that machine learning models are often black boxes that no human can understand, but it is possible to understand how a model works to some extent. Explainers can help with that.

The last topic in this section is managing models once they’ve been trained.

The final section of the exam guide is about deploying and consuming models. Once you’re happy with the performance of a trained model, you can deploy it in a couple of different ways. You can deploy it as a web service that responds to requests in real-time, or you can deploy it as a batch pipeline that takes in a large number of data records and sends the results for all of them at the same time. It takes longer, but it’s more efficient for large amounts of data.

This learning path assumes that you already have some basic experience using Microsoft Azure. If you don’t have any experience yet, then please take our Overview of Azure Services course first.

Now, are you ready to learn about doing data science on Azure? Then let’s get started! To get to the next course in this learning path, click on the Learning Path pullout menu on the left side of the page. But please remember to rate this introduction before you go on to the next course. Thanks!