-
Notifications
You must be signed in to change notification settings - Fork 1.4k
Add detail to KNN no-match explanations #16271
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
Open
slow-J
wants to merge
4
commits into
apache:main
Choose a base branch
from
slow-J:KNNexplain
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
+186
−15
Open
Changes from all commits
Commits
Show all changes
4 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -149,9 +149,64 @@ public Query rewrite(IndexSearcher indexSearcher) throws IOException { | |
| topK = runSearchTasks(tasks, taskExecutor, perLeafResults, leafReaderContexts); | ||
| } | ||
| if (topK.scoreDocs.length == 0) { | ||
| return MatchNoDocsQuery.INSTANCE; | ||
| return new MatchNoDocsQuery("No documents matched the nearest-neighbor search"); | ||
| } | ||
| return DocAndScoreQuery.createDocAndScoreQuery(reader, topK, reentryCount); | ||
| return DocAndScoreQuery.createDocAndScoreQuery( | ||
| reader, topK, reentryCount, noMatchExplainer(topK, filterWeight)); | ||
| } | ||
|
|
||
| /** Builds the explainer for documents this query did not collect, capturing minTopKScore. */ | ||
| private DocAndScoreQuery.NoMatchExplainer noMatchExplainer(TopDocs topK, Weight filterWeight) { | ||
| // topK is score-descending, so the lowest collected score is the last entry. | ||
| final float minTopKScore = topK.scoreDocs[topK.scoreDocs.length - 1].score; | ||
| return (context, doc, topN) -> | ||
| explainNotCollected(context, doc, topN, filterWeight, minTopKScore); | ||
| } | ||
|
|
||
| /** Explains why a doc was not collected, by recomputing its score. null when no vectors. */ | ||
| private Explanation explainNotCollected( | ||
| LeafReaderContext context, int doc, int topN, Weight filterWeight, float minTopKScore) | ||
| throws IOException { | ||
| String prefix = "Not in top " + topN + " doc(s): "; | ||
| FieldInfo fi = context.reader().getFieldInfos().fieldInfo(field); | ||
| if (fi == null || fi.getVectorDimension() == 0) { | ||
| return null; | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. what happens when null is returned? |
||
| } | ||
| VectorScorer vectorScorer = createVectorScorer(context, fi); | ||
| if (vectorScorer == null) { | ||
| return null; | ||
| } | ||
| if (vectorScorer.iterator().advance(doc) != doc) { | ||
| return Explanation.noMatch(prefix + "no vector value in field \"" + field + "\""); | ||
| } | ||
| if (filterWeight != null && docPassesFilter(filterWeight, context, doc) == false) { | ||
| return Explanation.noMatch(prefix + "excluded by filter"); | ||
| } | ||
| float score = vectorScorer.score(); | ||
| if (score < minTopKScore) { | ||
| return Explanation.noMatch(prefix + "score " + score + " < minTopKScore " + minTopKScore); | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. can we expand minTopKScore in the explanation to something more human-readable like "minimum top-K score"? |
||
| } | ||
| // Score meets the cutoff but the doc was not collected (tie-break, recall miss, or rescoring). | ||
| return Explanation.noMatch( | ||
| prefix | ||
| + "score " | ||
| + score | ||
| + " >= minTopKScore " | ||
| + minTopKScore | ||
| + " (tie-break or approximate-search miss)"); | ||
| } | ||
|
|
||
| private static boolean docPassesFilter(Weight filterWeight, LeafReaderContext context, int doc) | ||
| throws IOException { | ||
| Scorer scorer = filterWeight.scorer(context); | ||
| if (scorer == null) { | ||
| return false; | ||
| } | ||
| TwoPhaseIterator twoPhase = scorer.twoPhaseIterator(); | ||
| if (twoPhase != null) { | ||
| return twoPhase.approximation().advance(doc) == doc && twoPhase.matches(); | ||
| } | ||
| return scorer.iterator().advance(doc) == doc; | ||
| } | ||
|
|
||
| private TopDocs runSearchTasks( | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
filterWeight here was created during rewrite() against a specific reader. But explain() can be called later with a different reader. when that happens, filterWeight.scorer(context) gets a LeafReaderContext it wasn't built for, which could blow up.
match side is fine since scores/docs are already materialized in the array. but this no-match path re-evaluates the filter live, so it has this stale-reader risk.
can you add a try/catch around the filter check.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks, good catch!
Looking into this a bit, I don't think it would always throw an exception when called with a different reader, but it would evaluate the filter against the other reader's segment, giving the wrong answer.
Instead of the try/catch, I think we should gate the explain in DocAndScoreQuery#createWeight on contextIdentity, like so:
This is similar to the existing check in createWeight.
What do you think?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added the guard as proposed above and confirmed that it works as intended.
Steps taken to confirmed it:
explain(Weight, int)withsearcherB.explain(weightA, 6);Without the guard, the IndexSearcher#explain(Weight, int) gives an incorrect explanation, with the check it gives the generic
Not in top X doc(s).message.