pg_embedding
pg_embedding copied to clipboard
Missing features
Lists of features that would be useful
Correctness
- VACUUM support
Performance
- Deduplicate vectors (like GIN & btree)
- Implement IVF-flat, with clusters stored in btrees, ordered by distance from centroid. This allows for prefetching & improved prioritization of candidate tuples)
- Store the index on disk
The index shouldn't need time to rebuild every time the system restarts, and shouldn't consume memory if there is no-one to use the index.
I hear that this is WIP.
Features
- More similarity/distance functions: