Machine Learning in SQL Server 2017

SQL Server R Services has been renamed SQL Server Machine Learning Services, to reflect support for Python in addition to the R language. You can use Machine Learning Services (In-Database) to run R or Python scripts in SQL Server, or install Microsoft Machine Learning Server (Standalone) to deploy and consume R and Python models that don't require SQL Server.
SQL Server developers now have access to the extensive Python ML and AI libraries available in the open-source ecosystem, along with the latest innovations from Microsoft:
  • revoscalepy - This Python equivalent of RevoScaleR includes parallel algorithms for linear and logistic regressions, decision tree, boosted trees and random forests, as well as a rich set of APIs for data transformation and data movement, remote compute contexts, and data sources.
  • microsoftml - This state-of-the-art package of machine learning algorithms and transforms with Python bindings includes deep neural networks, fast decision trees and decision forests, and optimized algorithms for linear and logistic regressions. You also get pre-trained models based on ResNet models that you can use for image extraction or sentiment analysis.
  • Python operationalization with T-SQL - Deploy Python code easily by using the stored procedure sp_execute_external_script. Get great performance by streaming data from SQL to Python processes and using MPI ring parallelization.
  • Python in SQL Server compute contexts - Data scientists and developers can execute Python code remotely from their development environments to explore data and develop models without moving data around.
  • Native scoring - The PREDICT function in Transact-SQL can be used to perform scoring in any instance of SQL Server 2017, even if R isn't installed. All that's required is that you train the model using one of the supported RevoScaleR and revoscalepy algorithms and save the model in a new, compact binary format.
  • Package management - T-SQL now supports the CREATE EXTERNAL LIBRARY statement, to give DBAs greater management over R packages. Use roles to control prviate or shared package access, store R packages in the database and share them among users.
  • Performance improvements - The stored procedure sp_execute_external_script has been optimized to support batch mode execution for columnstore data.
Source : Microsoft

Comments

Post a Comment