Principle:Sdv dev SDV Inequality Constraint

Knowledge Sources	SDV Documentation SDV
Domains	Data_Quality, Constraint_Satisfaction
Last Updated	2026-02-14 00:00 GMT

Overview

A column ordering constraint that ensures one column's values are always greater than (or equal to) another column's values in synthetic data.

Description

The inequality constraint enforces ordering relationships between two columns, such as start_date < end_date or min_price <= max_price. During data transformation, the constraint replaces the high column with the difference between high and low columns. This ensures the model learns the difference distribution (which is always positive) rather than two independent columns. During reverse transformation, the high column is reconstructed by adding the difference back to the low column.

Usage

Use this constraint when the dataset has paired columns with a known ordering relationship (e.g., temporal intervals, price ranges, min/max bounds).

Theoretical Basis

Transform: Replace high column with $\mathrm{\Delta }=high-low$ (guaranteed non-negative)

Reverse Transform: Reconstruct $high=low+\mathrm{\Delta }$

Validation: Check $high>low$ (strict) or $high\ge low$ (non-strict)

Related Pages

Implemented By

• Implementation:Sdv_dev_SDV_Inequality_Init