Principle:Sdv dev SDV Fixed Combinations Constraint
| Knowledge Sources | |
|---|---|
| Domains | Data_Quality, Constraint_Satisfaction |
| Last Updated | 2026-02-14 00:00 GMT |
Overview
A referential integrity constraint that ensures only previously observed combinations of values across multiple columns appear in synthetic data.
Description
The FixedCombinations constraint preserves valid value tuples across two or more columns. For example, in a dataset with country and city columns, only country-city pairs observed in the training data will appear in synthetic data. The constraint works by joining the constrained columns into a single composite column during transformation, then splitting them back during reverse transformation.
Usage
Use this constraint when the dataset has columns with dependent categorical values where not all combinations are valid (e.g., country-city, department-job_title, make-model).
Theoretical Basis
Transform: Join columns into a single composite column:
Reverse Transform: Split the composite column back into original columns
Validation: Check that all output tuples exist in the training data combinations