Evaluating Data Sampling Methods with a Synthetic Quality Score

The article discusses the evaluation of sampling procedures on the quality of synthetic tabular data using Gretel.ai's Synthetic Quality Score (SQS). It explains how to calculate and interpret the SQS, which measures inter-columnar correlations, variance directions via principal component analysis, and discrete mass distributions of each feature in a dataset. The article also explores different sampling methods and their impact on the quality of synthetically generated data using an ensemble of sampling methods that performs as well as direct sampling from the categorical distribution while reducing SQS performance variance.