Accelerate open source forecasting with SAS (part 2 of 3)


Here is the second installment of the 3-part series by guest bloggers Jessica Curtis and Andrea Moore. (If you missed it, here is Part 1.)


First and foremost, SAS distributes the input data for forecasting.  SAS knows how to split up data intelligently for time series forecasting, such that time series groups are not split across different worker nodes.  From there, SAS distributes the open source scripts themselves across multiple worker nodes.  Then, SAS distributes the execution of the open source code.  That’s right, when the EXTLANG package calls the Python or R code, it targets every single Python or R interpreter across multiple worker nodes so that time series are processed in parallel.  Think about what that means from a scalability and efficiency standpoint.  This enables you to not only solve one forecasting challenge, but also expand to solve many different forecasting challenges across the enterprise.  And solve them quickly, at scale.  

Imagine, if you will, you’re a global retailer.  Your vision is to solve many different forecasting challenges across the enterprise with one, consistent forecasting platform.  From predicting a compelling assortment of SKUs, to determining the right amount of inventory to carry across the supply chain, to optimizing labor at your stores, your goal is to drive accurate and analytics-driven decisions.  Today, you’ve taken an initial pass at developing aggregate level forecasts for financial planning decisions using R.  While you are seeing a lot of success in the R forecasting approach, you are looking to expand upon these forecasting capabilities and develop a forecast at a more granular level to support store labor decisions.  With a small team of forecast analysts, you require an automated process that will enable you to efficiently scale and expand across many different forecasting use cases.   

For financial planning at an aggregate level, you are running 1,000 time series.  For store labor planning by Store and Department, this quickly expands to 100,000 time series.  For supply chain planning at the SKU/Store level, the time series are in the millions.  That certainly sounds like a large-scale forecasting challenge.  A challenge that could be overcome only by the power of distributed and scalable forecasting.  Enter: SAS Visual Forecasting.   

The key to any successful large-scale forecasting challenge is automation.  And that’s exactly what SAS does.  SAS automates the statistical forecasting process and the open source models, which drives efficiencies in your business forecasting processes.  Through the power of TSMODEL and the EXTLANG package, SAS accelerates open source run time, further driving efficiencies in the forecasting process.  This alleviates your team from building forecast models one-at-a-time; you can move to a true exception-based process.  Time is freed up to focus on planning the business and expanding to new areas of forecasting.  Simply stated, do more with less.  

Once the models are created, SAS then automatically generates multiple output data sets.  This goes beyond the forecast itself; this also includes many different data sets containing model specifications, statistics of fit, and parameter estimates.  These output data sets are then – you guessed it – distributed.  This rich output data provides both the data science team and the business with a lot of insight into the key drivers of demand and model details.  Remember those discussions you’ve been having with your business partners who don’t trust the statistical forecast?  Well, the output data sets that SAS automatically creates help bring visibility to why the models are doing what they are doing, to enhance the discussions with the business, and improve adoption 

SAS Visual Forecasting also augments open source models with built-in best practices.  From a patented data diagnostics and model building process to automatic hierarchical forecasting with reconciliation to integrated time series segmentation, SAS Visual Forecasting goes beyond the algorithms and offers an end-to-end forecasting process based in best practices.   

With all the automation, acceleration, and augmentation, your organization can scale.  Scale across many different forecasting use cases across the enterprise.  Scale down to the lowest levels of granularity in your product/location hierarchy to produce forecasts at the level at which you execute those forecasts.  No longer do you have to rely on a high-level forecast that is manually disaggregated to lower levels.  SAS automatically generates high quality forecasts at the level at which business decisions are made. 

(Go to Part 3)


About Author

Mike Gilliland

Product Marketing Manager

Michael Gilliland is a longtime business forecasting practitioner and formerly a Product Marketing Manager for SAS Forecasting. He is on the Board of Directors of the International Institute of Forecasters, and is Associate Editor of their practitioner journal Foresight: The International Journal of Applied Forecasting. Mike is author of The Business Forecasting Deal (Wiley, 2010) and former editor of the free e-book Forecasting with SAS: Special Collection (SAS Press, 2020). He is principal editor of Business Forecasting: Practical Problems and Solutions (Wiley, 2015) and Business Forecasting: The Emerging Role of Artificial Intelligence and Machine Learning (Wiley, 2021). In 2017 Mike received the Institute of Business Forecasting's Lifetime Achievement Award. In 2021 his paper "FVA: A Reality Check on Forecasting Practices" was inducted into the Foresight Hall of Fame. Mike initiated The Business Forecasting Deal blog in 2009 to help expose the seamy underbelly of forecasting practice, and to provide practical solutions to its most vexing problems.

Back to Top