MLOps for Pirates: R-sasctl
R-sasctl was released in January of this year and in our SAS Model Manager community. R-sasctl helps generate the score code and metadata for R models as well as allows R developers to import their model into SAS Model Manager directly from their R development environment. This makes the hand-off between our data scientist developing in R and our MLOps engineer managing and operationalizing the model much easier. R-sasctl has several other neat functions as well, but we are going to focus on the registration today.
R-sasctl helps generate several pieces of metadata. To explain why we collect each piece let’s dive into the anatomy of a model.
First, the score code and scoring resources are necessary for scoring new data, both within SAS Model Manager, like score testing or performance monitoring, or outside of it, like publishing into CAS or containers. Next, the input and output variables are critical for knowing what data the model expects as well as what outputs it generates. It also gives our engineers and project owners a quick view of what variables the model uses, to ensure only approved data sources were used for modeling. Following, properties tell us which file represents which role as well as who our modeler was, what algorithm they used, and more. These are all important factors when it comes to auditing your model. And finally, our diagnostic files include numerous fit statistics, lift charts, roc curves and more. This helps us compare our models on accuracy to know which model is the best fit for our use case.
The following demo video will take you through the training and registration process for a R model within R Studio. At the end of the video, we will get a view of the model within SAS Model Manager, where it is managed, tested, monitored and deployed.
Demonstrated above are just a few of R-sasctl’s capabilities, but there are several things we didn’t touch on, including R-sasctl’s support for PMML models, the project management capabilities, model publishing, and execution of published models, including parallel execution of batch data. Check out R-sasctl’s documentation to learn more about these other neat functions! What to try out an example yourself? See R-sasctl’s GitHub page.
Have you tried R-sasctl out? Tell us your thoughts below!
