Technical Perspective: k-Shape: Efficient and Accurate Clustering of Time Series

Abstract

Database research frequently cuts across many layers of abstraction (from formal foundations to algorithms to languages to systems) and the software stack (from data storage and distribution to runtime systems and query optimizers). It does this in a way that is specialized to a particular class of data and workloads. Over the decades, we have seen this pattern applied to enterprise data, persistent objects, Web data, sensor data, data streams, and so on. Each time, the community has developed extensions to algebraic query primitives, specialized implementation techniques (index structures, pattern detection algorithms, update and consistency mechanisms, etc.), benchmarks, and new optimization techniques.