SAS Visual Forecasting (VF) is a powerful and flexible tool which could cover all forecasting needs of an organisation. Users can automatically produce large-scale time series analyses and hierarchical forecasts – without human involvement. Having both a friendly UI and a programmatic interface it allows all personas to take part in the forecasting process. The list below discusses my top 5 capabilities of SAS VF and provides references to some useful resources so you can explore the topics discussed in more detail. The list goes as follows:
Open-Source Scalability
SAS VF can distribute open-source code (Python and R) to run in parallel in the same nodes that SAS Viya is installed. In that way you can easily scale up forecasting processes which are developed in open-source to millions of series and move away from ungoverned, inconsistent and error-prone processes that run in users’ local desktops. For more info on the parallelization make sure to check the paper here and if you are looking for a practical example on how to parallelise Facebook-Prophet open-source algorithm using SAS Viya make sure to check the blogpost here.
Time-Series Segmentation
SAS VF comes with an out-of-the box segmentation capability based on the time-series patterns that are detected in user’s data. As a result users can apply different modelling strategies on the different segments. N,ot every series requires manual intervention from modellers; especially when we have to deal with hundreds of thousand of series. Arima and Exponential Smoothing models which are automatically generated and optimized by the system provide surprisingly accurate results when we have enough data history and we don’t deal with very complex patterns. By automatically segmenting time-series based on their patterns (volume, volatility, seasonality, intermittency etc.), users can develop modelling pipelines for each one separately and focus on developing sophisticated forecasting modelling strategies on the series that require the most attention. Have a look here for detailed information on the automatic segmentation or if you want to bring your own segments then have a look here to learn more about the ‘external segmentation’ pipeline.
Deep Learning
SAS VF provides users the capability to run RNNs, LSTMs and GRUs in a simple way. This is done using the recently developed ‘TNF’ package. The data is automatically structured in the right way saving significant time compared to users having to manually apply feature engineering techniques themselves. For a closer look then check this resource which includes both the necessary theory to understand what’s happening behind the scenes and also some good examples to get you started. If you want to incorporate the above deep learning techniques in your VF pipelines and compare and select the most accurate results with other forecasting algorithms at a series level, then have a look at the this blog which takes you through the end-to-end process.
Hybrid Modelling Techniques
SAS VF includes out-of-the box proprietary nodes which use Neural Networks in combination with traditional Time-Series techniques to provide the most accurate results. Feature engineering and data aggregation is also taking place automatically saving massive amounts of time to users. Something to be mindful of when using these techniques is that the model is trained on all data and you don’t get different models per series as for example would happen with RNNs, Arima, Exponential Smoothing etc. These techniques work great when we want to incorporate many variables in our forecasts or we have to deal with complex patterns in our data. Have a look at this paper that takes you through all the available techniques in more detail.
Build Your Own Nodes
SAS VF gives the flexibility to users to create their own custom nodes and share them as plug-and-play solutions with other data scientists and forecasters around the business. The most efficient way to do this is by using ‘The Exchange’ which is a space where SAS Viya users can exchange assets that have created in a simple and robust way. Alternatively you can also upload the node as a Zip file directly to SAS Viya. For a step-by-step guide have a look at this resource which demonstrates the process of developing a manual Gradient Boosting node which can be loaded and used directly in any forecasting pipeline.
That’s all for now. Stay tuned as new capabilities are released monthly and this list is just going to get bigger.
