CountMin

Count-Min Sketch, sometimes written as CountMin Sketch, is a probabilistic data structure used to estimate the frequencies of items in a data stream while using sublinear memory. It provides fast, approximate counts with formal error guarantees and is widely applied in streaming analytics, network monitoring, and large-scale databases.

The structure consists of a two-dimensional array of counters with d rows and w columns. Each row has an independent hash function that maps an item to a column index. To update an item, the corresponding counters in all d rows are incremented. To estimate an item's frequency, the values at the d mapped positions are read and the minimum of these values is returned. Because different items can hash to the same positions, estimates can overcount; taking the minimum across rows mitigates this effect.

Space and time complexity: the data structure uses O(w · d) counters. Update and query operations run in O(d) time. By choosing w and d appropriately—typically w ≈ e/ε and d ≈ ln(1/δ)—one can bound the overestimation of any item by ε times the total number of updates with probability at least 1 − δ.

Variants and extensions include conservative updates to reduce overestimation and the ability to support deletions in counting, as well as other sketch families that trade accuracy for different guarantees. Limitations include reliance on quality hash functions and the fact that exact counts are not recoverable; estimates may be biased upward in the presence of many collisions.

Applications include network traffic analysis, query frequency estimation in large databases, and real-time analytics where exact counting is impractical.

History: Count-Min Sketch was introduced by Graham Cormode and S. Muthukrishnan in 2005.