krzysztof/landing_page
1
0
Fork 0
mirror of https://github.com/qdrant/landing_page.git synced 2026-09-25 22:18:30 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix number

Browse Source
This commit is contained in:
Dylan Couzon
2026-08-07 21:44:33 -04:00
parent 0ee049853e
commit 9ed51561db
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
qdrant-landing/content/articles/filtered-vector-search-acorn.md
+1 -1
View File
@@ -131,7 +131,7 @@ Extra edges also make ACORN stronger. On the 4% intersection it reached 95.2% on
## What to Measure on Your Own Collection ## What to Measure on Your Own Collection
Measure recall for each filter shape you serve. Start with the ones most likely to break: values covering roughly a fifth of the collection or more, and `AND` combinations of them. [Facet counts](/documentation/manage-data/payload/#facet-counts) show which values are that broad. On the default configuration here, two filters fell short with ACORN off, at 39.7% and 90.8%, while the rest ranged 97.2% to 100%, and a single aggregate number would have hidden both. If those filters come back clean, test narrower values next. Measure recall for each filter shape you serve. Start with the ones most likely to break: values covering roughly a fifth of the collection or more, and `AND` combinations of them. [Facet counts](/documentation/manage-data/payload/#facet-counts) show which values are that broad. On the default configuration here, one filter returned 39.7% with ACORN off while every other filter stayed above 90%, and a single aggregate number would have hidden it. If those filters come back clean, test narrower values next.
Create a payload index on every field you filter on, and leave ACORN off to start, since that is Qdrant's default. Then sample a few hundred real queries per filter shape, 500 if you want to match this benchmark. Get exact results with `exact: true`, and score both recall and latency with ACORN off and then on. Create a payload index on every field you filter on, and leave ACORN off to start, since that is Qdrant's default. Then sample a few hundred real queries per filter shape, 500 if you want to match this benchmark. Get exact results with `exact: true`, and score both recall and latency with ACORN off and then on.