Principle:Sdv dev SDV Single Table Data Sampling
| Knowledge Sources | |
|---|---|
| Domains | Synthetic_Data, Data_Generation |
| Last Updated | 2026-02-14 00:00 GMT |
Overview
A data generation process that draws synthetic rows from a fitted statistical or neural model and reverse-transforms them to match the original data schema.
Description
After fitting, the sampling process generates new data by drawing from the learned model, then reversing all preprocessing transformations to produce output in the same format as the original data. The process includes batch sampling, constraint enforcement via reject sampling, and optional progress reporting.
Usage
Use sampling after a synthesizer has been fitted. Specify the number of rows to generate. For constrained synthesizers, the sampling process automatically applies constraint satisfaction through either transformation reversal or reject sampling.
Theoretical Basis
- Raw generation: Model draws samples in the transformed space
- Reverse transformation: DataProcessor reverses encoding, formatting, and anonymization
- Constraint enforcement: Chained constraints are reverse-transformed; reject-sampling constraints filter invalid rows
- Batch processing: For large requests, sampling can be done in batches with retries per batch
Related Pages
Implemented By
Uses Heuristic
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment