Azure Machine Learning allows IT pros to create machine learning applications using just a Web browser and an Internet connection. It pulls together resources from Microsoft's existing big data services such as the Hadoop-based HDInsight, iPython Notebook and Python Tools for Visual Studio as well as products in the Azure store. Microsoft has also published a step-by-step guide for data science apps that takes the user from raw data to a consumable Web service.
"The Azure Machine Learning Studio has an amazing [user interface] that follows the Microsoft IDE user experience, which I would argue is among the best in the industry," said Kevin Felichko, CTO of PropertyRoom.com, an online auction company based in Frederick, Md.
Kevin Felichko,CTO of PropertyRoom.com
With the Machine Learning studio, IT can easily do things such as drag and drop the workflow steps and code transformation steps without having to fire up VMs, Felichko said.
The all-in one interface and familiar Microsoft development tools have Felichko, who also manages deployments on Amazon Web Services (AWS), strongly considering Azure Machine Learning. Felichko said his company could use it to predict the success of auction listing based on rapidly changing data points such as length of auction, scheduled close time and geographical area. Currently, PropertyRoom doesn't have real-time evaluations of this data and must rely on post-auction analysis.
"This is where Microsoft beats AWS," Felichko said.
AWS provides a variety of services to build big data predictive analytics and machine learning applications. These include Amazon S3 for big data cloud storage, Amazon Kinesis for real-time big data streaming and analysis, high performance NoSQL databases with DynamoDB, data warehousing with Amazon Redshift and big data analytics with Amazon Elastic MapReduce, as well as machine learning apps in its Marketplace. Amazon's Elastic Beanstalk platform as a service (PaaS) can also be integrated with Kinesis.
But while AWS has delivered the major pieces that enable business intelligence apps, "to utilize these resources for machine learning, you still have to be a programmer," said Daniel Heacock, an Amazon RedShift user and senior business systems analyst for online ticketing service Etix, headquartered in Raleigh, N.C.
There is an abundance of third-party products which can be used for Machine Learning, such as Neuron from ColdLight Solutions and TappingStone Inc.'s PredictionIO, which is sold in the AWS Marketplace. IBM also has machine learning capabilities through its Watson APIs; and Google has a Prediction API that performs machine learning functions.
Azure a 'one-stop shop'
One IT pro who investigated both AWS and Azure PaaS offerings said he chose Azure because his application is primarily written using the Microsoft .NET framework. Amazon's Elastic Beanstalk also supports .NET, but Azure PaaS is easier to manage and offers one-stop shopping from testing to deployment. Azure includes Visual Studio online and the Team Foundation Server, according to Karl Schulmeisters, CTO for Carver Global Health Group, headquartered in Ashburn, Va., which makes a software product called ClearRoadmap that provides regulatory guides for medical device designers.
However, Schulmeisters believes AWS will soon catch up to Microsoft because "the two companies are pushing each other forward," he said.
Microsoft has done a better job curating machine learning capabilities so they are easy to find, plug-in and leverage, analysts said.
"AWS has a smattering of big data and [machine learning]-like capabilities but in typical AWS fashion they are somewhere in the catalog and thus hard to locate, put together and take greatest advantage of," said James Staten, analyst with Forrester Research based in Cambridge, Mass.
Other Azure apps including the Excel-based data processing service PowerBI and Data Factory, a cloud service for processing structured and unstructured data, are also among Microsoft's strengths in this area, Staten said. With Microsoft's acquisition of Revolution Analytics last month, it owns the best R language implementation in the market, he said.
Azure Machine Learning has a free tier which supports up to 10 GB of data. Beyond that, its pricing is broken into three hourly components: a Machine Learning Seat Subscription, which costs $9.99 per user per month; ML Studio Usage, which is priced at $1.00 per hour; and a $2.00 charge per hour for production Machine Learning API usage.
AWS declined to comment for this story.