[GitHub] [lucene] msokolov commented on a change in pull request #238: LUCENE-10016: add support for Knnvector index and query to the demo

[GitBox <gi...@apache.org>]

msokolov commented on a change in pull request #238:
URL: https://github.com/apache/lucene/pull/238#discussion_r684771041



##########
File path: lucene/core/src/java/org/apache/lucene/search/KnnVectorQuery.java
##########
@@ -0,0 +1,318 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.lucene.search;
+
+import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS;
+
+import java.io.IOException;
+import java.util.Arrays;
+import java.util.Comparator;
+import java.util.Objects;
+import org.apache.lucene.codecs.KnnVectorsReader;
+import org.apache.lucene.document.KnnVectorField;
+import org.apache.lucene.index.IndexReader;
+import org.apache.lucene.index.LeafReaderContext;
+
+/** Uses {@link KnnVectorsReader#search} to perform nearest Neighbour search. */
+public class KnnVectorQuery extends Query {
+
+  private static final TopDocs NO_RESULTS =
+      new TopDocs(new TotalHits(0, TotalHits.Relation.EQUAL_TO), new ScoreDoc[0]);
+
+  private final String field;
+  private final float[] target;
+  private final int k;
+
+  /**
+   * Find the <code>k</code> nearest documents to the target vector according to the vectors in the
+   * given field. <code>target</code> vector.
+   *
+   * @param field a field that has been indexed as a {@link KnnVectorField}.
+   * @param target the target of the search
+   * @param k the number of documents to find
+   * @throws IllegalArgumentException if <code>k</code> is less than 1
+   */
+  public KnnVectorQuery(String field, float[] target, int k) {
+    this.field = field;
+    this.target = target;
+    this.k = k;
+    if (k < 1) {
+      throw new IllegalArgumentException("k must be at least 1, got: " + k);
+    }
+  }
+
+  @Override
+  public Query rewrite(IndexReader reader) throws IOException {
+    int boundedK = Math.min(k, reader.numDocs());
+    TopDocs[] perLeafResults = new TopDocs[reader.leaves().size()];
+    for (LeafReaderContext ctx : reader.leaves()) {
+      // Calculate kPerLeaf as an overestimate of the expected number of the closest k documents in
+      // this leaf
+      int expectedKPerLeaf = Math.max(1, boundedK * ctx.reader().numDocs() / reader.numDocs());
+      // Increase to include 3 std. deviations of a Binomial distribution.
+      int kPerLeaf = (int) (expectedKPerLeaf + 3 * Math.sqrt(expectedKPerLeaf));
+      perLeafResults[ctx.ord] = searchLeaf(ctx, kPerLeaf);
+    }
+    // Merge sort the results
+    TopDocs topK = TopDocs.merge(boundedK, perLeafResults);
+    // re-query any outlier segments (normally there should be none).
+    topK = checkForOutlierSegments(reader, topK, perLeafResults);
+    if (topK.scoreDocs.length == 0) {
+      return new MatchNoDocsQuery();
+    }
+    return createRewrittenQuery(reader, topK);
+  }
+
+  private TopDocs searchLeaf(LeafReaderContext ctx, int kPerLeaf) throws IOException {
+    TopDocs results = ctx.reader().searchNearestVectors(field, target, kPerLeaf);

Review comment:
       IIRC we need something on `LeafReader` so it can flow through filters and the like to the call to `CodecReader.getVectorReader()`. If we eliminate the `LeafReader` method then we'd have to conditionally cast to `FilterReader` unwrap and cast to `CodecReader`. This seems cleaner to me. Plus there can be uses for nearest vector search outside of IndexSearcher.search. 




-- 
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: issues-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org