In the 9 months or so since I last wrote about Revolution Analytics they have released a new version of Revolution R Enterprise 7. This is focused on delivering “Write once deploy anywhere.” R, of course, continues to expand in popularity with the recent Rexer Data Mining survey (reviewed here) putting the percentage of data miners that use it at 70%. Revolution Analytics has been focused all along on helping address some of the challenges of open source R by delivering a version that is not in-memory bound and that supports parallel threading for instance. Version 7 has a number of new capabilities.
- First, Revolution Analytics has continued to add to the set of R packages available on their scalable multi-threaded platform, with Stepwise Logistic Regression, Stepwise GLM, Decision Forests (a random Forests implementation) and tree visualization. While there are still packages that are being worked on the core set used by the vast majority of projects are now available as multi-threaded and scalable implementations on v7.
- The data source integration in Revolution R Enterprise has been extended to include HP Vertica and Teradata Aster via ODBC (in addition to existing text, SAS, SPSS, HDFS and Teradata connections)
- V7 offers improved support for PMML with the ability to generate PMML models directly from the high performance Revolution R Enterprise packages rather than having to use the core R packages.
- Revolution Analytics continues to add to its BI integrations with new support for easy integration into Excel and into Tableau in addition to existing work with Qlikview.
- A partnership with Alteryx allows Revolution R Enterprise scripts to be dropped into Alteryx workflows. Alteryx previously supported open source R but the new partnership allows the more scalable and performance Revolution Analytics implementations to be dragged and dropped into the workflows. This gives a more business friendly interface that still has the ability to leverage the full Revolution Analytics platform.
- Revolution R Enterprise 7 allows the same R code used on servers to be run in a Cloudera or Hortonworks Hadoop cluster. This eliminates the need to rewrite code and avoids having to export data from Hadoop to build a predictive model.
This last item is really important. There has been so much change at the data infrastructure level that deploying models is becoming much more complex – Hadoop, streaming environments, databases and data warehouses all play a role. Picking one of these platforms and committing to it is risky so Revolution Analytics is allowing customers to build powerful predictive models on Big Data regardless of where the data comes from and then implement a production process on whichever platform makes sense so that models can be trained (and re-trained) in-situ. The code can run locally, on a RedHat or SUSE Linux server/cluster, on a Windows server/cluster, on Hadoop (Cloudera, Hortonworks) or in Teradata. All it takes is a single line of code to establish how to get the data that the model needs.
Revolution Analytics is one of the vendors in our Decision Management Systems Platform Technologies Report, and you can get more information on them here.