Extended PMML Support and New Options in Performance Monitoring
Support for More PMML Models & Versions
With the release of SAS Viya 2025.06, SAS Model Manager has extended our support for Predictive Model Markup Language, known as PMML. PMML offers a standard for describing and exchanging predictive models using XML. SAS Model Manager now supports converting more types of models across more versions of PMML into SAS Data Step code. The full list of supported models can be found in our documentation, but the versions of PMML supported now include 4.1, 4.2, 4.3, and 4.4
SAS Model Manager uses PROC PSCORE to generate SAS Data Step code that is functionality equivalent to the PMML model. The generated score code can be executed on all platforms that are supported by SAS to score the data sets. To take advantage of this code generation, simply import the supported PMML file into SAS Model Manager using the Import button. As the model is being imported, SAS Model Manager will automatically initiate PROC PSCORE to create the SAS Data Step code. Opening the model will show the score code, XML file, inputs, outputs, and any metadata gleaned from the file. The following demo walks through importing a PMML 4.1, 4.2, 4.3, and 4.4 file into SAS Model Manager:
When developing the PMML code, it is important to note that the PROC PSCORE restricts that all input, output, or generated variable names be less than 32 characters. Attempting to import a PMML file with a variable name that is too long will result in an error. To resolve the error, simply open the PMML file in a text editor and reduce the variable name to fewer than 32 characters, ensuring to correct the variable name in all locations it is referenced, before trying to import the model again into SAS Model Manager.
This extended PMML support expands the number of models in an organization that can be registered, managed, tested, deployed, and monitored in SAS Model Manager.
Better Monitor Rare Events
In the following SAS Viya release 2025.07, SAS Model Manager made more options for Performance Monitoring available. When running performance monitoring, SAS Model Manager uses the same calculations for assessing model performance as SAS Model Studio or PROC ASSESS. This ensures consistency across models assessed throughout SAS Viya, but it also means that we can surface new options for our users.
Users that are modeling against rare events asked to surface options for specifying the maximum number of iterations for the percentile algorithm, the number of bins to be used in the lift calculation, and the number of cuts to be used in the ROC calculation. With the latest release of SAS Viya, users can now change these values when creating a new performance monitoring definition.
Note that increasing these values can cause performance monitoring to run for longer as more calculations are being made. Additionally, changing the number of bins for the lift calculation may cause some of the lift-based KPIs to be unavailable. But for users tracking rare events, they’ll have a better chance to see performance changes by having more fine-grained calculations.
Did you know that both of these features were based on customer feature requests? You can share what features you would like to see next in SAS Model Manager in the comments below, but you can also use SAS Product Suggestions to make a suggestion for any product. If you do submit a suggestion, feel free to tag @SophiaRowland.
