Sampling-Based Estimation of the Number of Distinct Values of an Attribute

We provide several new sampling-based estimators of the number of distinct values of an attribute in a relation. We compare these new estimators to estimators from the database and statistical literature empirically, using a large number of attribute-value distributions drawn from a variety of real-world databases. This appears to be the first extensive comparison of distinct-value estimators in either the database or statistical literature, and is certainly the first to use highlyskewed data of the sort frequently encountered in database applications. Our experiments indicate that a new “hybrid” estimator yields the highest precision on average for a given sampling fraction. This estimator explicitly takes into account the degree of skew in the data and combines a new “smoothed jackknife” estimator with an estimator due to Shlosser. We investigate how the hybrid estimator behaves as we scale up the size of the database.