Recordvary

Recordvary is a data-analytic concept used to describe the degree of variability among the records in a dataset. It focuses on how much attribute values differ across records and is commonly used in data profiling, quality assessment, and feature engineering. The term can refer to two related notions: the variability of individual attributes across records and the overall diversity of records when considering multiple attributes.

For numeric attributes, recordvary is often quantified by dispersion measures such as variance or standard deviation; for categorical attributes, by distributional diversity measures such as entropy or the Gini impurity. Some implementations compute a composite recordvary score for a subset of attributes by aggregating the per-attribute variability, or by calculating a distance-based diversity metric across records (for example, average pairwise distance in attribute space).

Applications include identifying columns with high or low discriminative power, guiding data cleaning, detecting anomalies or duplicate patterns, and informing feature selection for machine learning. Recordvary can also aid in monitoring data drift over time when applied to time-sliced snapshots of a dataset.

Limitations include dependence on chosen attributes, handling of missing values, and sensitivity to data encoding. Interpretations of recordvary should consider the context and whether the goal is to maximize diversity, detect anomalies, or ensure consistency.

See also: data profiling, data quality, variance, entropy, distance metrics.