Vector Search in ClickHouse
summary
Describes using ClickHouse for vector search to accelerate multidimensional queries such as those arising from text and image embeddings. Covers performance techniques and ready-made solutions for analytical workloads that involve high-dimensional data in ClickHouse.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:31What is vector search
- 3:00Vector search in ClickHouse
- 5:55Choosing index algorithms
- 9:27ClickHouse index structure
- 11:17Proxy neural network indexes
- 13:04Algorithm integration issues
- 14:51Data structure and optimization
- 16:40Index accuracy and recall
- 17:42How Annoy works
- 18:48Annoy test results
- 20:14Limit and granularity impact
- 22:03Embedding quality and inserts
- 23:51Vector search applications
- 24:14Index demo
- 26:23Q&A
related talks
33mBringing Vector Search to OLTP
45mClickHouse: what is behind the fastest columnar database
59mContextual search with vector search: exploring your options with open source tools
44mVector Search Beyond the Hype: Real-World Implementation and Performance Insights
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
42mClickHouse: what is behind the fastest columnar database
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
52mVector search is now boring, but PostgreSQL has ways to go