This is part one in a series on low-code machine learning with Azure ML.
Machine Learning with Less Code
Getting started with machine learning is a real challenge, and there are two reasons for it: one mathematical and one programmatical. First, you need a good understanding of statistical and mathematical concepts to get very far with data science and machine learning. It’s not like you need a PhD to get started, but there’s so much depth and nuance to the field that it can take years of experience and a willingness to dive into research papers to get the most out of an analysis.
Then we have the development side. While you’re boning up on techniques and trying to understand what’s available, you may also be trying to learn a new programming language (typically Python) and a variety of new libraries.
That’s why we’ve seen a cottage industry in low-code and no-code machine learning solutions. The one that we’re going to focus on in this series is Azure Machine Learning. I should note that Azure ML is much more than just a low-code/no-code solution for machine learning, but for our purposes here, we’ll focus on it. In the next series, I’ll get beyond the basics with Azure ML.
Creating a Workspace
The first thing that we need to do is create a machine learning workspace. In the Azure portal, search for “machine learning” and choose the Machine learning option.
From there, select Create to add a new workspace. In the Basics tab, you can see that we need several resources for Azure Machine Learning.
You’ll need a workspace, which is hosted in a region. Workspace names only need to be unique within a resource group—they don’t need to be globally unique. Then, we need four separate resources: an Azure storage account, a Key Vault, an Application Insights service, and a container registry. Azure Machine Learning uses the storage account to retain input data and store the logs from different ML runs. We need a Key Vault to store any relevant credentials the AML service uses. Application Insights stores monitoring information about deployed models. And we use the container registry to store those generated models as containers for deployment.
Resources: New or Used?
There’s a good question here about whether we should allow Azure Machine Learning to create the relevant resources for us or if we should reuse existing services. If you’re just trying out Azure Machine Learning for yourself, it’s probably easier just to let AML create the resources for you. But if you’re working on a corporate solution, you’ll probably want to create these resources first, especially if the person creating the AML workspace doesn’t have rights to create all of the relevant resources.
The Advanced tab has a few more options of interest.
The last two options are particularly interesting to us. First, data encryption allows us to use either Microsoft-managed keys or our own keys. If your company necessitates using custom-generated keys for encrypting all data in the cloud, you can choose a key from Key Vault for this purpose. Finally, if you select the “High business impact workspace” option, Microsoft will collect less diagnostic data.
After you create your workspace, it will be available to you for further use. In the next post, we will take a look at what’s available in Azure Machine Learning Studio and create the resources we’ll need if we want to get going on an ML project.
2 thoughts on “Low-Code ML: Creating a Workspace”