Skip to content

Bulk score hnsw neighbor array #15958

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
leng25 wants to merge 9 commits into
base: main
Choose a base branch
from
+21 −21
Open

Bulk score hnsw neighbor array #15958

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
72fddd8
Utilize bulk scoring interface during HNSW graph builder NeighborArra…
Apr 13, 2026
9b80a3c
Merge branch 'apache:main' into bulk-score-hnsw-neighbor-array
leng25 Apr 14, 2026
4dec1ee
Merge branch 'apache:main' into bulk-score-hnsw-neighbor-array
leng25 Apr 14, 2026
cc51f88
Merge branch 'main' into bulk-score-hnsw-neighbor-array
leng25 Apr 16, 2026
b12cce5
Merge branch 'main' into bulk-score-hnsw-neighbor-array
leng25 Apr 17, 2026
fc10f97
Merge branch 'main' into bulk-score-hnsw-neighbor-array
leng25 Apr 27, 2026
f7e3c22
Merge branch 'main' into bulk-score-hnsw-neighbor-array
leng25 May 3, 2026
0d90048
Merge branch 'main' into bulk-score-hnsw-neighbor-array
leng25 May 9, 2026
e3e5261
Refactor: populate uncheckedNodes upfront in findWorstNonDiverse
May 28, 2026
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
2 changes: 2 additions & 0 deletions lucene/CHANGES.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,8 @@ Optimizations

* GITHUB#15597, GITHUB#15777: Reduce memory usage of NeighborArray (Viliam Durina)

* GITHUB#15606: Utilize bulk scoring for NeighborArray#isWorstNonDiverse (Luis Negrin)

Bug Fixes
---------------------
* GITHUB#14049: Randomize KNN codec params in RandomCodec. Fixes scalar quantization div-by-zero
Expand Down
40 changes: 19 additions & 21 deletions lucene/core/src/java/org/apache/lucene/util/hnsw/NeighborArray.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -292,13 +292,19 @@ private int findWorstNonDiverse(UpdateableRandomVectorScorer scorer) throws IOEx
int[] uncheckedIndexes = sort(scorer);
assert uncheckedIndexes != null : "We will always have something unchecked";
int uncheckedCursor = uncheckedIndexes.length - 1;
int[] uncheckedNodes = new int[uncheckedIndexes.length];
for (int i = uncheckedCursor; i >= 0; i--) {
uncheckedNodes[i] = nodes[uncheckedIndexes[i]];
}
float[] bulkScores = new float[size];
for (int i = size - 1; i > 0; i--) {
if (uncheckedCursor < 0) {
// no unchecked node left
break;
}
scorer.setScoringOrdinal(nodes[i]);
if (isWorstNonDiverse(i, uncheckedIndexes, uncheckedCursor, scorer)) {
if (isWorstNonDiverse(
i, uncheckedIndexes, uncheckedCursor, scorer, uncheckedNodes, bulkScores)) {
return i;
}
if (i == uncheckedIndexes[uncheckedCursor]) {
Expand All @@ -309,31 +315,23 @@ private int findWorstNonDiverse(UpdateableRandomVectorScorer scorer) throws IOEx
}

private boolean isWorstNonDiverse(
int candidateIndex, int[] uncheckedIndexes, int uncheckedCursor, RandomVectorScorer scorer)
int candidateIndex,
int[] uncheckedIndexes,
int uncheckedCursor,
RandomVectorScorer scorer,
int[] uncheckedNodes,
float[] bulkScores)
throws IOException {
float minAcceptedSimilarity = scores[candidateIndex];
if (candidateIndex == uncheckedIndexes[uncheckedCursor]) {
// the candidate itself is unchecked
for (int i = candidateIndex - 1; i >= 0; i--) {
float neighborSimilarity = scorer.score(nodes[i]);
// candidate node is too similar to node i given its score relative to the base node
if (neighborSimilarity >= minAcceptedSimilarity) {
return true;
}
}
} else {
// else we just need to make sure candidate does not violate diversity with the (newly
// inserted) unchecked nodes
assert candidateIndex > uncheckedIndexes[uncheckedCursor];
for (int i = uncheckedCursor; i >= 0; i--) {
float neighborSimilarity = scorer.score(nodes[uncheckedIndexes[i]]);
// candidate node is too similar to node i given its score relative to the base node
if (neighborSimilarity >= minAcceptedSimilarity) {
return true;
}
}
return scorer.bulkScore(nodes, bulkScores, candidateIndex) >= minAcceptedSimilarity;
Copy link
Copy Markdown
Member

@benwtrent benwtrent May 26, 2026

Choose a reason for hiding this comment

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

It seems like we only want to score uncheckedIndexes right?

Copy link
Copy Markdown
Contributor Author

@leng25 leng25 May 28, 2026

Choose a reason for hiding this comment

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

Hi @benwtrent , thanks for the review!
Not exactly, it depends on whether the candidate itself is checked or unchecked.

Case 1 (candidate is unchecked): brand new node, never been diversity-checked against anyone, so we score against all better neighbors (0..candidateIndex-1) not just unchecked ones.
Case 2 (candidate is checked): already survived diversity checks against all checked nodes, so we only score against the newly added unchecked nodes.

This is the same logic as before my change, I only replaced the per-node scorer.score() loop with bulkScore.

That being said, looking at it again I made two small improvements: renamed bulkScoreNodes to uncheckedNodes to make it clearer what it actually contains, and moved the population of uncheckedNodes up into findWorstNonDiverse so it's built once and reused across calls to isWorstNonDiverse rather than being recreated every time.

}
return false;
// else we just need to make sure candidate does not violate diversity with the (newly
// inserted) unchecked nodes
assert candidateIndex > uncheckedIndexes[uncheckedCursor];
return scorer.bulkScore(uncheckedNodes, bulkScores, uncheckedCursor + 1)
>= minAcceptedSimilarity;
}

public int maxSize() {
Expand Down
Loading