Skip to content

[KNN] Adding default value for oversampling in 9.1.0 #1290

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.

Sign up for GitHub

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

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open

[KNN] Adding default value for oversampling in 9.1.0 #1290

Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
865e5ee
adding default value for oversampling in the documentation
Samiul-TheSoccerFan Apr 28, 2025
251291c
updating the documentation for bbq and oversample
Samiul-TheSoccerFan Apr 29, 2025
85feae8
Update documentation wording and adding version
Samiul-TheSoccerFan May 5, 2025
5a34ea3
Merge branch 'main' into update_documentation_for_oversample
Samiul-TheSoccerFan May 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions solutions/search/vector/knn.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -901,7 +901,7 @@ Approximate kNN search always uses the [`dfs_query_then_fetch`](https://www.elas

When using [quantized vectors](elasticsearch://reference/elasticsearch/mapping-reference/dense-vector.md#dense-vector-quantization) for kNN search, you can optionally rescore results to balance performance and accuracy, by doing:

* **Oversampling**: Retrieve more candidates per shard.
* **Oversampling**: Retrieve more candidates per shard. The default is `3.0` in `bbq`.
* **Rescoring**: Use the original vector values for re-calculating the score on the oversampled candidates.

As the non-quantized, original vectors are used to calculate the final score on the top results, rescoring combines:
Expand All @@ -913,7 +913,7 @@ All forms of quantization will result in some accuracy loss and as the quantizat

* `int8` requires minimal if any rescoring
* `int4` requires some rescoring for higher accuracy and larger recall scenarios. Generally, oversampling by 1.5x-2x recovers most of the accuracy loss.
* `bbq` requires rescoring except on exceptionally large indices or models specifically designed for quantization. We have found that between 3x-5x oversampling is generally sufficient. But for fewer dimensions or vectors that do not quantize well, higher oversampling may be required.
* `bbq` requires rescoring except on exceptionally large indices or models specifically designed for quantization. We have found that between 3x-5x oversampling is generally sufficient. But for fewer dimensions or vectors that do not quantize well, higher oversampling may be required. As noted above, we default to an oversampling factor of `3.0`.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
* `bbq` requires rescoring except on exceptionally large indices or models specifically designed for quantization. We have found that between 3x-5x oversampling is generally sufficient. But for fewer dimensions or vectors that do not quantize well, higher oversampling may be required. As noted above, we default to an oversampling factor of `3.0`.
* `bbq` requires rescoring except on exceptionally large indices or models specifically designed for quantization. We have found that between 3x-5x oversampling is generally sufficient. But for fewer dimensions or vectors that do not quantize well, higher oversampling may be required. Starting in `9.1.0`, the default oversampling factor is 3.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, that works. Thank you :)


You can use the `rescore_vector` [preview] option to automatically perform reranking. When a rescore `oversample` parameter is specified, the approximate kNN search will:

Expand Down
Loading