mayya-sharipova commented on a change in pull request #656:
URL: https://github.com/apache/lucene/pull/656#discussion_r804028845
##########
File path: lucene/core/src/java/org/apache/lucene/search/KnnVectorQuery.java
##########
@@ -96,43 +107,98 @@ public Query rewrite(IndexReader reader) throws
IOException {
return createRewrittenQuery(reader, topK);
}
- private TopDocs searchLeaf(LeafReaderContext ctx, int kPerLeaf, Bits
bitsFilter)
+ private TopDocs searchLeaf(LeafReaderContext ctx, int kPerLeaf,
BitSetCollector filterCollector)
throws IOException {
- // If the filter is non-null, then it already handles live docs
- if (bitsFilter == null) {
- bitsFilter = ctx.reader().getLiveDocs();
+
+ if (filterCollector == null) {
+ Bits acceptDocs = ctx.reader().getLiveDocs();
+ return ctx.reader()
+ .searchNearestVectors(field, target, kPerLeaf, acceptDocs,
Integer.MAX_VALUE);
+ } else {
+ BitSetIterator filterIterator = filterCollector.getIterator(ctx.ord);
+ if (filterIterator == null || filterIterator.cost() == 0) {
+ return NO_RESULTS;
+ }
+
+ if (filterIterator.cost() <= k) {
+ // If there <= k possible matches, short-circuit and perform exact
search, since HNSW must
+ // always visit at least k documents
+ return exactSearch(ctx, target, k, filterIterator);
+ }
+
+ try {
+ // The filter iterator already incorporates live docs
+ Bits acceptDocs = filterIterator.getBitSet();
+ int visitedLimit = (int) filterIterator.cost();
+ return ctx.reader().searchNearestVectors(field, target, kPerLeaf,
acceptDocs, visitedLimit);
+ } catch (
+ @SuppressWarnings("unused")
+ CollectionTerminatedException e) {
Review comment:
I liked very much of "a special subtype of TopDocs instead, which has an
explicit "complete" flag"
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
