In one of the latest SAS Viya releases (2023.07), SAS made available the Distributed Open Source Code (DOSC) node in SAS Visual Forecasting. The DOSC node enables forecasting users to use native Python or R code and preferred packages in a seamless way to perform time series forecasting, and at the same time, scale their process to run in a distributed way using the SAS Viya compute nodes. Why is this important?
Scalability
Users that want to use specific open source packages, are no longer limited by the computing power of their own machines. With the growing volume of data they are dealing with, they can take advantage of SAS Viya’s compute nodes and run their forecasting process in a distributed manner. This opens the door for more experimentation with more granular data that can significantly improve the end results of their forecasting process.
Visualization and Filters
Forecasters get out-of-the-box visualizations of their forecasts using the ‘forecast viewer’ capabilities of the DOSC node. Using this functionality, they can visually inspect the generated forecasts as well as apply various filters to identify and further investigate problematic series. As such, it enables a forecaster to identify and resolve any issues quickly and effectively.
Interactive Modeling and Overrides
Another great advantage that comes with this feature is that forecasters can employ an Interactive Modeling (IM) node right after they produce their open-source forecasts. This allows users to replace the forecasts that don’t meet their business expectations on the fly using either auto-generated models from the IM node or by creating their own custom models within the IM node. Users can also use the IM node to generate forecasts for series when the open source model has failed. If users want to infuse business knowledge that is not represented or captured in the data, they can still do this at a later stage in the Overrides tab of SAS Visual Forecasting. The combination of these capabilities offers users the greatest level of flexibility when it comes to using open source algorithms, proprietary and custom algorithms, and applying judgemental/business knowledge.
Reproducibility and Collaboration
After expert statistical programmers develop DOSC nodes (where they would apply different open source packages and algorithms), they can save those nodes and corresponding pipelines into a collaboration area (aka “The Exchange”) to share with their colleagues for re-use in other SAS Visual Forecasting projects. This enables users that are not experts in statistical programming apply the open source modeling strategies that their colleagues developed and achieve true democratization throughout the organization when it comes to advanced analytics.
Pipelines’ Competition
Finally, users can develop different forecasting pipelines, using both SAS and open source algorithms and compare them as part of the same process. This ensures that they are always selecting the best methods for their forecasting purposes based on a reliable and consistent process.
Below you can see this functionality in action which was presented by SAS’ Forecasting Product Manager, Joe Katz, during July’s ‘SAS Viya Release Highlight Show’.
For more information regarding the DOSC node properties and how to use it make sure to check the documentation using this link.
