Azure Machine Learning has evolved rapidly since its introduction into public preview back in the summer of 2014. There’s been a constant stream of changes, updates, and added features, as well as the integration of third-party acquisitions like Revolution Analytics and Jupyter Notebooks. Azure Machine Learning has also been partly absorbed into a larger set of offerings, the Cortana Analytics Suite. So it can sometimes be confusing for new users to navigate the service, know what’s where, and assemble a decent map of where to look for next steps and answers to questions. This post is an attempt to provide some orientation.
Azure Machine Learning Hubs
Azure Machine Learning has a number of key hubs or locations. There’s the Azure Machine Learning Studio, where you run and tweak your experiments. Links along the top give access to the Cortana Analytics Gallery, where you can access pre-existing templates, algorithms and APIs, and view a number of tutorials, videos, and general documentation. Any Microsoft account (formerly Windows Live ID) is sufficient to sign in and start experimenting. That gives you access to a very generous free tier; there’s no SLA, and the experiments you run and call from your applications won’t be very fast, but it’s enough to do lots of experiments and get familiar with the Azure spin on how Machine Learning should be done. There’s even a guest tier, requiring no sign-up, which resets after 8 hours.
More intensive users will want to quickly move on to the Standard tier. They can do so by signing up for an Azure account and going to another key hub for Azure Machine Learning, in the classic Azure Portal. Creating a workspace here automatically puts you in the Standard tier, and enables additional configuration options like attaching a new or pre-existing Storage Account, adding more users and leveraging Azure Active Directory Authentication. But you still need to flip back to the studio to compose experiments and train models. It may seem a little confusing that there are these different hubs at different URLs, but it’s likely we’ll see everything brought together once Machine Learning gets added to the new “Ibiza” Portal. (It’s also worth noting that browsing for Machine Learning in the new portal will eventually bring you to the same place, but it takes a few more clicks to get there).
Differences between the tiers can be seen below:
There’s really a vast amount of Azure Machine Learning-related documentation, guides, templates, and videos at the various locations mentioned above and elsewhere. That’s great once you’re up and running – there’s invariably a document or tutorial or experiment or guided video for whatever it is you’re trying to do next – but can be a little daunting at first. I’d recommend taking a look at the following.
Once you’re signed up, a good thing to do is simply browse the Cortana Analytics Gallery to get a sense of the kinds of problems Azure Machine Learning can solve without even doing much customised work. A word on the different categories of item you can find in the gallery. Machine Learning APIs package up various kinds of sophisticated data analysis for use as black boxes. Experiments can also be used for data analysis, but are rather less black box in nature. Templates provide end-to-end, industry-specific partner solutions. And Collections package groups of related items together. It’s important to note that everyone can contribute to this gallery, so when you’re starting out, you may want to constrain your search to gallery contributions from Microsoft. For example, the following link shows all Microsoft-produced content sorted by popularity, a kind of what’s hot right now.
If you want to move on to having more of a learning map for the product as a whole, I’d recommend the following.
The Cortana Analytics Process touches on areas beyond Azure Machine Learning and shows how all the different data analytics services on Azure can be composed together, while the bottom sector of the map shows all the principal stages of the Azure Machine Learning process and the key documentation pages for them on the Azure site.
There’s also a free ebook, Azure Machine Learning Essentials, written by Jeff Barnes from the Microsoft Partner Enterprise Architecture team, downloadable in various handy formats for phones and ebook readers from the Microsoft Virtual Academy. It should however be noted that there have been almost a year’s worth of updates since the book was published.
If you’re tempted to look beyond the Azure site for more platform-agnostic training in Machine Learning, Andrew Ng’s Coursera Machine Learning Course is a perennial favourite.
That wraps up an introductory look at the Azure Machine Learning platform and documentation.