SAS Viya Copilot in Action (Part 2): Smart Next-Node Recommendations in Model Studio Pipeline
Views
54
In my previous posts—SAS Viya Copilot Explained: Building Machine Learning Pipelines in Minutes, Not Hours, and SAS Viya Copilot in Action (Part 1): Intelligent Assistance for Building Model Studio Pipelines—I introduced SAS Viya Copilot, its core concepts, and how it provides intelligent assistance for building pipelines in Model Studio. In this post, we will take a deeper dive into one of its key features called the Node Suggestion capability in model pipeline development, with a particular focus on how it assists users within Model Studio.
SAS Viya Copilot can help you decide what node to add next in your Model Studio pipeline. When you select a valid parent node, Copilot suggests the most appropriate next node, making it easier to keep your pipeline moving forward, especially when you’re unsure about the next step.
Node suggestions are available after the Data node and most data preprocessing nodes. However, a few specialized risk modeling nodes, Pairwise Selection, Pricing Optimization, and Reject Inference, are excluded.
When does node suggestion work?
For SAS Viya Copilot to suggest the next node, a few conditions must be met:
- The pipeline must be unlocked (some pipelines that are created by SAS automation can be locked).
- The pipeline must not be read-only.
- The selected parent node must be one of the applicable nodes.
- The parent node must have successfully run.
- The parent node must be selected in the pipeline or explicitly mentioned by name in your prompt.
When all these conditions are satisfied, node suggestion is enabled for that parent node. If any condition is not met, the feature is disabled.
How do you get node suggestions?
You have two simple options:
- Ask SAS Viya Copilot directly (for example, by prompting it to suggest the next node).
- Right-click the applicable parent node and choose Copilot Suggestions.
Select any image to see a larger version.
Mobile users: To view the images, select the "Full" version at the bottom of the page.
What kind of suggestions do you get?
By default, Copilot provides initial node suggestions based on node metadata. These suggested nodes use their default parameter values, and no model tuning is performed at this stage.
If you want more advanced guidance, you can ask Copilot to run an exploratory process. In this mode, Copilot trains multiple exploratory models to identify the most suitable next node—or even a sequence of connected nodes. During this process, parameter tuning may occur, so the suggested nodes might include parameter values that differ from their defaults for better performance.
How to get initial node suggestions?
Once you select a parent node in your pipeline and open the SAS Viya Copilot side pane, you’ll see a Suggest the next node button. Clicking this button is the quickest way to get initial recommendations for what node to add next.
You can also ask Copilot for suggestions using a natural‑language prompt. To do this, make sure either:
- the parent node is already selected in the pipeline, or
- you explicitly mention the parent node by name in your prompt.
After typing your prompt, click Submit prompt or press Enter.
Example prompts for initial suggestions
| When a node is already selected: | When identifying the parent node by name: |
|
|
|
|
|
|
|
|
What Copilot returns
In response, SAS Viya Copilot provides a list of candidate nodes that you can add next. These suggestions may include:
- data preprocessing nodes,
- modeling nodes, or
- a mix of both—depending on the current state of your pipeline.
Each suggested node appears with a button that you can click to add it directly to your pipeline.
You can add only one of the suggested nodes. Once you click a node:
- that button is disabled,
- the other suggestions are removed from the response, and
- the selected node is added to your pipeline.
Copilot then provides a Run node button, which you can use to immediately run the newly added node and continue building your pipeline.
How to get better suggestions by using the exploratory process?
If you want more than metadata‑based recommendations, SAS Viya Copilot’s exploratory process can help you get stronger, data‑driven suggestions. Instead of proposing nodes based only on their definitions, this process actively runs a series of trials, training different models to determine the best child node—or sequence of nodes—to add next.
During exploration, Copilot may tune model parameters to improve performance. Model quality is evaluated using the project’s selection criteria, which are defined in the Project settings. As each trial completes, the models are compared and ranked. When the exploration finishes, Copilot recommends the best‑performing models or pipelines as suggested child nodes.
When you first ask for initial node suggestions, Copilot might include a Start exploration button in its response. If the initial suggestions don’t meet your expectations, you can simply click this button to launch the exploratory process. Alternatively, you can skip the initial suggestions altogether and ask Copilot to run the exploratory process directly.
To start the exploratory process, you must either:
- select the parent node in the pipeline before submitting your prompt, or
- explicitly mention the parent node by name in your prompt.
Example prompts to start the exploratory process
| When a node is already selected: | When identifying the parent node by name: |
|
|
|
|
|
|
What happens during exploration?
Once initiated, Copilot runs several trial models. This can take a few minutes, depending on the data and modeling options being evaluated.
As the exploration runs, interim results appear in the SAS Viya Copilot side pane and are refreshed periodically. These results show:Note: While the exploratory process is running, you cannot interact with that pipeline. However, you are free to switch to another pipeline or create a new one and continue working there.
- the nodes or pipelines tried so far, and
- the model comparison statistic (for example, Average Squared Error) computed on the validation partition, if one exists.
If no validation partition is available, Copilot uses training results instead. If data sampling was applied, that information is also shown in the results.
The exploratory process may recommend:
- a single modeling node, or
- a combination of data preprocessing nodes followed by a modeling node.
Stopping the process and adding suggestions
While the process is running, a Stop Running button appears in the side pane. If you are satisfied with the current results, you can stop the process early. Otherwise, you can let it run to completion.
When the process stops or finishes, Copilot displays the final model comparison results along with buttons that let you add a single suggested node, or a series of connected nodes.
You can select only one option. Once you click a button:Note: To add these suggestions, you must be viewing the same pipeline for which the exploratory process was run.
- the remaining buttons are removed, and
- the selected node or nodes are added to your pipeline.
Because tuning may have occurred during exploration, some parameters might differ from their default values. When you add a suggested node, these tuned values are clearly shown in the side pane.
Finally, Copilot provides a Run nodes button so you can immediately execute the newly added node or nodes and continue building your pipeline.
Concluding Remarks
In summary, SAS Viya Copilot’s node suggestion capability streamlines pipeline development by guiding users to appropriate next steps based on context, node readiness, and best practices. It reduces guesswork by offering quick metadata‑driven recommendations or more advanced, performance‑optimized suggestions through exploratory modeling and tuning. Overall, this capability accelerates model building, improves decision quality, and lowers the expertise barrier, enabling users to build effective pipelines with greater confidence and efficiency.
Find more articles from SAS Global Enablement and Learning here.
