SAS Model Manager supports a stress-free handoff between the data scientist training the model and the MLOps engineer managing and the deploying the model. To easily pass models from the development environment to our unified repository, SAS Model Manager provides various methods for model registration. These methods were designed to register models using a similar approach to how the models were trained. Models developed using point-and-click in Model Studio, can be registered in a button click. Models developed using Python and R can be registered using the Python-sasctl package and the R-sasctl package. Models developed in SAS Studio can be registered using a SAS macro. Now with the rising popularity of Flows in SAS Studio, we want to provide another mechanism for registration.
We’ve been working with the SAS Studio team to create a new step to register SAS models. This step fits seamlessly within a flow. Data scientists can train their models and register them in the same flow! This step supports models that generate a Data Step score code, such as regressions and decision trees, as well as models that generate an Analytical Store (Astore) table.
To get started with this new step, do the following:
- In your model training code, be sure to save the model scoring resource. This will be the Data Step file or the Astore table. As a best practice, I also recommend saving the scored data table. This data table is used to generate the metadata required to compare models.
- For Astore models, you will need to add a Table node that references the Astore table. A Table node is also necessary for the scored data. Be sure to run all Table nodes before filling out the Register SAS Model Step.
- Add the Register SAS Model Step and attach the Table node(s) as input(s). If you need to add a second input port, right-click the register step and add an additional port from the options.
- Fill out the required information. To add the scored data, check the first box on the Data tab and fill out the information about the scored data table.
- Run the step and see the models within SAS Model Manager, ready to manage, deploy, and monitor.
To see the step in action, check out this demo video:
Given the support of Python code within SAS Studio flows, we are currently working on adding another step to Register Python models. Stay tuned for more! In the meanwhile, what enhancements would you like to see for the Register SAS Model step and what steps would you like to see next for SAS Model Manager?
Want to learn more about SAS Studio flows? Check out the following resources:
