Unlike Azure advocates who are Microsoft employees and will not tell you whole truth, here I am describing Microsoft’s offering in Machine Learning area that’s part of their cloud platform named “Azure” and I am telling how it is, no marketing pep talk.
WARNING: Azure has no spending limits. So if you use Azure via MSDN or via some promotion then all is good, but if you are buying Azure services for commercial purposes you must be careful so that spike of usage or cyber attack will not result in many thousands of dollars charged to your credit card.
So, let’s get started, here is super short overview of Machine Learning (and a bit of artificial intelligence services) in Azure: Azure Machine Learning / AI comprises:
- Azure Machine Learning Studio
- Data Science Virtual Machines
- Azure Machine Learning services
- Cognitive Services
- Other Azure constructs that can be used for Machine Learning
And here is overview of Machine Learning offering as seen from Azure Portal:
As you can see above, 2 items are still in “preview” mode so clearly it’s a hot area on which Microsoft is working on big time: apart from Internet of Things, Machine Learning and other areas of Artificial Intelligence is where Microsoft sees the bright future.
What is Azure Machine Learning Studio? It’s a web app where you can use drag and drop with boxes and connect them. Microsoft designed Azure Machine Learning Studio to democratize Machine Learning: so that more people and not just professional data scientists can benefit from power of Machine Learning. Once models are trained in Machine Learning Studio they can be productized in form of web services. Here is how Azure Machine Learning Studio looks like (keep in mind: it’s a web app, so you use it in a browser):
Data Science Virtual Machines are pre-prepared VMs that have all necessary tools for machine learning. There are various types of these machines, there is even one with GPU to enable effective deep learning.
Azure Machine Learning services enables development of machine learning algorithms in Python and then offering them as web services.
Azure is cloud service so can it run absolutely everything also non-Microsoft technologies but let’s take a look at other Machine Learning offerings from Microsoft:
- Microsoft Cognitive Toolkit that allows to build and use neural networks but seems like much inferior offering to Google’s deep learning gigantic TensorFlow
- Cortana Intelligence Gallery that offers many templates and experiments that you can clone and quickly achieve results or get inspiration
- big scale data analysis tools like HDInsights with Spark SQL
- SQL Server R Services so you can execute R scripts from T-SQL statements, quasi R and recently also Python embedded inside of Microsoft SQL Server – a popular database server in enterprises
- Cognitive Services where pre-trained models can be used for example to identify persons in images and to detect mood of persons (could be useful in development or chatbots – another super hot, apart from Machine Learning, subset of Artificial Intelligence)
Well, as you can see above I have presented short overview of Machine Learning in Azure – Microsoft’s gigantic cloud platform. As you can see Azure Machine Learning makes it easy for all people, not just data scientists, to use and even to operationalize (by exposing web services with trained models) the machine learning. Of course hard core data scientists and machine learning engineers will also find useful tools like deep learning VMs with GPU hardware and big scale clusters. Disadvantage of Azure Machine Learning is that some companies are not allowed to upload their data to the cloud (due to security and privacy reasons) and that Microsoft doesn’t offer hard limit for spending so you might wake up with many thousands of $ if Chinese botnet will abuse your Machine Learning web service…
I don’t claim that this overview is complete and if you see that something is missing or wrong please post comment below!