A Tour of SAS Model Manager
The purpose of this blog is to provide a tour of SAS Model Manager and discuss at a high-level the role it plays in ModelOps. Are you a data scientist who not only builds machine learning models, but is also responsible for putting those models into a production scoring environment? Are you the model steward at your organization who oversees hundreds or even thousands of models and must monitor the performance of these models over time? Or are you just curious to learn more about this so called ModelOps stuff (by the way, ModelOps is short for Model Operations)? No matter what your answer is to any of the above questions, there’s likely to be something new for you to learn from this tour of SAS Model Manager.
SAS Model Manager is the flagship SAS product to perform ModelOps activities. ModelOps means making the deployment and monitoring of machine learning models scalable, to be able to handle any number of models used by an organization. With SAS Model Manager you can:
• track data, models, and score code
• work with models developed in SAS 9 and SAS Viya as well as Open Source languages
• run champion-challenger tournaments
• assess the performance of models over time
• use workflow to assign tasks to users when ModelOps activities are required
• and much more.
Here’s a quick tour and some highlights of the SAS Model Manager interface and its capabilities. There are 5 main category views for Model Manager, and each is represented by an icon in a menu found to the left of the interface. You navigate around Model Manager for different ModelOps activities using these category views. The five categories along with their representative icons are: Home 
Select any image to see a larger version.
Mobile users: To view the images, select the "Full" version at the bottom of the page.
When opened, Model Manager starts at the Home category view. The Home view shows existing projects and models and provides other sources of information that Model Manager users will find helpful. The Home view provides links to SAS communities as well as to documentation and videos for information on topics like “What’s New” and “How To”.
Let’s continue the tour by skipping to the third category on the list, Projects, since this is the category where most of the activities in Model Manager are performed and where users spend much of their time while doing ModelOps. The initial Projects view shows all projects that currently exist within Model Manager. You can think of this initial page as the “view all” page; all projects are shown. New projects are created in this “view all” Projects view, but can also be created from the Home view. In the screen shot below, the Projects view shows there are currently 3 projects that exist in this instance of Model Manager. Some basic metadata about the projects is shown.
Individual projects can be opened to reveal what they contain. Projects may contain one or several models, information and results from scoring tests and performance reports, and a history record of all activities performed within the project. When a project is opened, there are several tabs across the top that are used to navigate through and work with the project. Some of these tabs include Models, Variables, Properties, Scoring, Performance, and History. In the screen shot below, a project named QS_HMEQ is opened. The project is open on the Models tab, where it can be seen that 3 models are listed, thus the project contains these three models. A model called QS_Tree1 has been declared the project champion (note the special symbol in the Role column next to the name of that model).
In the categories view list, the second category from the top is for Models. As with the Projects view, a summary of all models contained within Model Manager can be displayed, but individual models can also be opened to reveal their properties. The screen shot below is showing the Models page when no specific model is opened; this is the “view all” page for models. It shows that there are 9 models contained within Model Manager, the projects the models are contained in, and even the type of score code for each model.
When an individual model is selected, the Models view changes to show information and properties on that specific model. Machine learning models are a lot more than just score code. Additional files for models contain information on input and output variables, model properties such as code language and score code type, and summary information on model performance. Just as within projects, there are several tabs that are used to navigate the information about models. For example, in the screen shot below, a model named QS_Reg1 is open on the Files tab. Files for the score code, input and output variables and other properties about the model can be seen.
Now continuing to move back down the list of categories (the icons on the far left), second to last is for Deployments. How many models are deployed? Where are they deployed? Where are models published and how many have been validated? This type of information about ModelOps activities is summarized on the Deployment page. The screen shot below shows that 3 models are deployed and that 3 projects have published models.
The final category to navigate through Model Manager is for Tasks. Tasks use an application called Workflow, that comes with Model Manager, but is not required to be used with Model Manager. A workflow is simply a series of tasks. The tasks can be connected into a node-link diagram to create a model of the workflow called a workflow definition. The tasks in a workflow definition may be assigned to specific users or roles and may need to be performed in a certain order. These tasks consist of ModelOps activities such as importing models, setting and approving a champion model, and publishing of score code to a scoring environment.
Well, sorry to say folks, but our tour ends here. We’ve discussed the reasons for SAS Model Manager and what its capabilities are. We’ve also explored and seen highlights of all 5 category views. Now that you have the basic idea of Model Manager and the role it plays in streamlining ModelOps, the next step is up to you!
Are you interested in learning more about SAS Model Manager or ModelOps in general? There are lots of ways to do so. -Consider taking the instructor-led or self-paced e-learning course, Managing Models in SAS Viya.
-Check out some of the free videos and tutorials that are available, such as these from the SAS Users YouTube channel: Deploying Model in SAS or SAS Viya- Model Management/Execution for Python, R and SAS.
-I’m also covering the topic at the SAS Explore conference in Las Vegas, in a hands-on-workshop. My workshop will be held on Wednesday, September 13 at 4:00 PM. Here’s where to register for the conference, if you haven’t done so yet: explore.sas.com
