Book Image

SQL Server 2017 Machine Learning Services with R.

By : Julie Koesmarno, Tomaž Kaštrun
Book Image

SQL Server 2017 Machine Learning Services with R.

By: Julie Koesmarno, Tomaž Kaštrun

Overview of this book

R Services was one of the most anticipated features in SQL Server 2016, improved significantly and rebranded as SQL Server 2017 Machine Learning Services. Prior to SQL Server 2016, many developers and data scientists were already using R to connect to SQL Server in siloed environments that left a lot to be desired, in order to do additional data analysis, superseding SSAS Data Mining or additional CLR programming functions. With R integrated within SQL Server 2017, these developers and data scientists can now benefit from its integrated, effective, efficient, and more streamlined analytics environment. This book gives you foundational knowledge and insights to help you understand SQL Server 2017 Machine Learning Services with R. First and foremost, the book provides practical examples on how to implement, use, and understand SQL Server and R integration in corporate environments, and also provides explanations and underlying motivations. It covers installing Machine Learning Services;maintaining, deploying, and managing code;and monitoring your services. Delving more deeply into predictive modeling and the RevoScaleR package, this book also provides insights into operationalizing code and exploring and visualizing data. To complete the journey, this book covers the new features in SQL Server 2017 and how they are compatible with R, amplifying their combined power.
Table of Contents (12 chapters)

Monitoring the accuracy of the productionized model

In Chapter 6, Predictive Modeling, we discussed a number of predictive modeling examples. The model(s) created is/are based on trained data. In a real-world scenario, new data keeps coming in, for example, online transactions, taxi cab transactions (remember the earlier NYC taxi example), and air flight delay predictions. Therefore, the data model should be checked regularly to ensure that it is still satisfactory and that there is no other better model that could be generated for it. With the latter, a good data scientist would continuously be asking at least four of these questions:

  1. Is there a different algorithm to consider due to changes of the data?

For example, if the current model is using logistic regression (rxLogit), would the decision tree algorithm more accurate (rxDTree) either due to the size or due to changes...