Sergey Nivens - Fotolia
Looking for something else?
SEATTLE -- While SQL Server 2017 continues to get attention for opening up to Linux, many of Microsoft's database advances revolve around various ways the company is opening up analytics on its flagship database. Case in point: SQL Server machine learning services.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
Open source data frameworks and development languages increasingly have become a path to next-level data analytics and machine learning, and SQL Server support is central to this strategy.
The clues are various. Even before the 2017 release of the database, Microsoft brought Apache Spark and the R language into the mix. Earlier this year, the Python language joined R as part of a newly minted Azure Machine Learning developer kit.
The story took a new turn at PASS Summit 2017 last week, as Microsoft featured the capability for Azure Machine Learning users to bring their analytics models into SQL Server 2017 for native T-SQL runtime scoring. An essential element in machine learning, scoring is a way to measure the likely success of machine-generated predictions.
Native T-SQL scoring can process large amounts of data at an average of under 20 milliseconds per row, according to Rohan Kumar, general manager of Microsoft's database systems group, who spoke at PASS Summit. Native T-SQL scoring takes the form of a stored procedure for prediction that can be used without calling Microsoft's R runtime, as was the case with SQL Server 2016.
This capability is important because models built and trained to, for example, suggest new products to likely buyers can produce results while the buyers are actually web browsing. As SQL Server machine learning services head in this direction, their use could grow.
Machine learning models
Supporting such scoring in the Microsoft database could make machine learning analytics more a part of operations and less an experimental effort, according to Ginger Grant, advanced analytics consultant for SolidQ and a presenter at the event.
"Traditionally, what has happened is that you've had a data science group that sort of sat in the corner creating machine learning models. They then threw that 'over the wall' to developers who had to code it in another language," Grant said in an interview.
"Native T-SQL scoring allows people to modularize their work and environment, so things can be operationally implemented relatively quickly," she said.
Microsoft's new SQL Server machine learning services will help with real-time prediction, said Victoria Holt, who also took part in PASS Summit. She is an independent data analytics and platforms architect, as well as a trainer at SQL Relay.
"It is great to be able to leverage machine learning computation in-database," she said.
This year's inclusion of Python in the Microsoft Machine Learning workbench is also a step forward, Holt said. But it will take time for such new technologies to spread.
Holt noted that the "addition of Python extends the use of deep learning frameworks in the product. The retrained cognitive models will speed up consumption. But there is significant user training and upgrading that will need to happen before these models are adopted."
Beyond T-SQL stored procedures
Jen Stirrupfounder of Data Relish
Microsoft analytics advances discussed at PASS Summit were not limited to T-SQL. The company previewed scale-out features for Azure Analysis Services to improve response time for large query workloads on the cloud.
The company also moved to simplify data preparation for analytics in the cloud by releasing a public preview of Azure Data Factory that includes the ability to run SQL Server Integration Services in ADF.
Growing Microsoft SQL Server 2017 support for Python and R is significant, according to Jen Stirrup, founder of the U.K.-based Data Relish consultancy and PASS Summit board member.
Python is something of a portal to a crop of machine learning services entering the open source sphere almost daily. In Stirrup's view, deeper support for advanced analytics is the next step for big data, and Microsoft is tuned to that notion.
"The company understands that customers really want to do something with the data," she said.
"The data is such a key thing. It underpins your applications. Today, that means you have to reach out to software and languages that are not necessarily part of Microsoft's .NET," Stirrup continued. "Microsoft's moves are all about being more welcoming to open source communities."
Listen to a podcast covering PASS Summit 2017 news
Learn more about U-SQL, another approach to big data analytics
Find out what SQL Server for Linux says about Microsoft's future plans
How Microsoft machine learning is expanding with Azure Databricks