SAS Data Maker • Product Update
Conditional generation: more control over the synthetic data you create
When we published our last product roadmap update, conditional generation was one of the most anticipated capabilities on the SAS Data Maker roadmap. The goal was simple: give users the ability to generate synthetic data for specific scenarios, segments, and outcomes, rather than always generating data that mirrors the overall source population.
We're moving closer to that goal. Conditional generation gives organizations greater control over the data they create, while continuing to preserve the statistical characteristics and relationships learned during training.
Why conditional generation matters
Organizations often need data for specific business scenarios. Conditional generation helps teams create targeted synthetic datasets for purposes such as:
- Testing an underwriting model using a particular risk segment
- Studying rare patient outcomes in clinical research
- Creating validation datasets for underrepresented populations
- Simulating targeted business conditions before deployment
What's new
Conditional generation is now available from the Generation tab for generators trained using supported algorithms.
After training a generator, users can specify conditions when generating data, allowing them to create synthetic records that align with specific needs without retraining the generator.
Categorical conditions
Generate synthetic records where a categorical variable matches a specified value. For example:
TARGET = 1
SAS Data Maker generates synthetic observations that satisfy the condition while preserving the relationships found within the relevant subset of the source data.
Numerical, date, and time ranges
Generate synthetic data within a range specified for interval variables. For example:
Income between $75,000 and $100,000
January 1, 2025 through March 31, 2025
This enables users to focus on specific segments or business scenarios while maintaining realistic relationships across variables.
Multiple conditions
Users can also combine conditions using logical operators. For example:
TARGET = 1 AND Gender = 'M'
This allows for more precise scenario creation and supports increasingly specific analytical requirements.
Preserving relationships across data
Data rarely exists in a single table. That's why SAS Data Maker is designed to maintain consistency across related tables connected through primary key and foreign key relationships.
This means conditions applied to one table can propagate appropriately across related tables, helping maintain referential integrity throughout generated datasets.
Designed for reuse
One of the key design principles is the separation of training and generation. A generator is trained once.
After that, users can generate multiple synthetic datasets using different conditions without retraining. This reduces processing overhead and makes experimentation both easier and faster, allowing teams to explore multiple scenarios from the same trained generator.
Use cases
|
Rare events
Create synthetic data focused on low-frequency occurrences to support modeling, testing, and analysis.
|
|
Underrepresented segments
Generate data for specific segments that can support more balanced testing, development, and analysis.
|
|
Scenario creation and testing
Create datasets to target scenarios without extensive post-processing.
|
Looking ahead
Conditional generation represents a crucial step for making synthetic data more flexible, targeted, and useful within real-world analytical workflows.
The first implementation focuses on generation-time conditions, allowing users to apply filters after training and reuse generators across many business scenarios. As customer feedback continues to shape the roadmap, additional conditional generation capabilities may be introduced in future releases.
