-
Notifications
You must be signed in to change notification settings - Fork 100
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Investigate caching PQ similarity subcalculations #87
Comments
Initial testing of a lazy thread-local float[][] cache in the context of each search as well as precomputation suggests there is mileage here. Lazy caching has a small overhead on small graphs and precomputation of all subvector comparisons at the start of a query has fixed overhead determined by PQ. As comparisons go up, we see benefit from both approaches, with a surprising improvement in the precalculation cases. |
|
|
I'm tabling this for now and will retest after #80 is available. There's potential here, but it may need some heuristic to gate precalculation. |
Hacked together branch (shoving the cache into dotProduct) is here https://github.com/jkni/jvector/tree/cache-pq-results. Most recent commit precalculates, previous commit is lazy cache. |
No description provided.
The text was updated successfully, but these errors were encountered: