Searching for Time Series Patterns at Big Data Scale
summary
Presents a practical pipeline that extracts motifs, discords and chains from massive series using STAMP (matrix profile), accelerates distance computation with FFT and offers a reference implementation on Dask + Stumpy + Elasticsearch. Shows multi‑dimensional and streaming support, an anytime/approximate path (SCRIMP+), and GPU scaling to DGX1/DGX2, but warns real runs over 150 million points still take hours–days; indexing matrix profiles into Elasticsearch enables interactive nearest‑neighbor queries.
AI-written summary · about
chapters (9)
- 0:00Introduction and agenda
- 9:58Dask for distributed computing
- 18:26Elasticsearch for indexing
- 26:04Matrix profile algorithm basics
- 36:05Finding patterns and motifs
- 39:59Detecting outliers and discords
- 43:11Scaling with Stumpy and Dask
- 47:27Advanced use cases and optimizations
- 53:11Key takeaways and conclusion
related talks
42mMulti-GPU Accelerated Processing of Time-Series Data of Huge Academic
21mAnalyzing Massive Time Series Data with Spark
35mSignatures, patterns and trends: Timeseries data mining @ Etsy
01h 01mParallel Python Analyzing Large Data Sets
26mTime Series Analysis in Python: Easy Tools for Scientific Insight
10mScaling Time Series Analysis With Argo Workflows: Patterns and Practi...
44mMethods and Tools for Time Series Data Science Problems
19mBig Data