ML Functions Release Notes

2025-12-16

Automatic ML Training App Lifecycle Management

When ML Functions are enabled for an organization, SingleStore now automatically provisions an associated ML Training app. Disabling ML Functions will also terminate the ML Training app for the org. For eligible customers, an internal beta track is also available, where enabling the relevant beta flag switches the ML Training app to the latest internal version, and disabling it switches back to the stable customer version.

2025-11-18

Improved Data Validation and Guidance for ML Model Training

Training ML models now includes enhanced data validation with clearer guidance. The system checks for a minimum number of rows based on the number of labels and alerts you if your dataset is too small or too large, suggesting when to add more data or sample/filter the table. Additional preventive checks validate that the target column has at least two classes, that each class has sufficient samples, and that the dataset is not empty, providing actionable messages when these conditions are not met.

2025-11-07

Stronger Validation and New Field for ML Model Creation

The CreateMLModel API now enforces clearer constraints on model metadata and schema fields, helping you catch configuration issues earlier, where:

• ProjectID and workspaceID must be valid UUIDs

• Name allows only @, -, and _, with trimmed length between 4 and 200 characters

• Runtime string length 1 to 100 characters

• ModelType restricted to Classification and Anomaly

• PipelineVersion length 1 to 20 characters

• Description either null or 1 to 1000 characters after trimming

• SelectedFeature must be valid JSON (mandatory for Classification)

• TableName and ColumnName allow leading and internal spaces but forbid / and trailing spaces, with a 256-character limit

• DBname must be alphanumeric/underscore and cannot start with a digit

In addition, a new optional TargetLabelColumn field is available for supervised anomaly detection workflows, complementing existing target series settings and enabling richer anomaly use cases.