Time series patterns are pretty common in everyday life. Extracting patterns on these data sets and providing interactive search responses has immense value across many software systems. This talk presents an overview of how such a search system can be built for very large time series data points.

This talk focuses on the STAMP algorithm (Scalable Time Series Anytime Matrix profile) to showcase how time series patterns can be extracted for big data sets. The talk covers concepts of Matrix profile and how motifs/patterns are detected. Once the patterns are analyzed, the results are converted into a document format that is amenable for indexing via a search engine. The session then proceeds to extend the idea to detecting time series across multiple dimensions. A part of the session also covers the distributed approach to extract these time series patterns to tackle big data sets. Further, you will also learn about mechanisms to calculate time series segmentations, time series chains and discords or anomalies thereby enabling ideas for users to search for anomalies and segments of time series.

Dask, Stumpy and Elasticsearch stack is used to describe a reference implementation.