mikemccand commented on code in PR #15979: URL: https://github.com/apache/lucene/pull/15979#discussion_r3664615151
########## lucene/core/src/java/org/apache/lucene/codecs/lucene106/dedup/Lucene106DedupHnswVectorsFormat.java: ########## @@ -0,0 +1,221 @@ +/* + * 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.codecs.lucene106.dedup; + +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.DEFAULT_BEAM_WIDTH; +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.DEFAULT_MAX_CONN; +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.DEFAULT_NUM_MERGE_WORKER; +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.HNSW_GRAPH_THRESHOLD; +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.MAXIMUM_BEAM_WIDTH; +import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat.MAXIMUM_MAX_CONN; + +import java.io.IOException; +import java.util.concurrent.ExecutorService; +import org.apache.lucene.codecs.KnnVectorsFormat; +import org.apache.lucene.codecs.KnnVectorsReader; +import org.apache.lucene.codecs.KnnVectorsWriter; +import org.apache.lucene.codecs.hnsw.FlatVectorsFormat; +import org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsReader; +import org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsWriter; +import org.apache.lucene.index.MergePolicy; +import org.apache.lucene.index.MergeScheduler; +import org.apache.lucene.index.SegmentReadState; +import org.apache.lucene.index.SegmentWriteState; +import org.apache.lucene.search.TaskExecutor; +import org.apache.lucene.util.hnsw.HnswGraph; + +/** + * An HNSW vector format that de-duplicates raw vectors. + * + * <p>Graph construction and search are identical to {@link + * org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsFormat}. A {@link DedupFlatVectorsFormat} is + * used for the flat vector storage, which stores each distinct vector exactly once, shared across + * all documents that reference it. This trades a small amount of indexing work for reduced storage + * when vectors repeat, e.g. multiple fields derived from the same embedding or heavily duplicated + * content. + * + * @lucene.experimental + */ +public final class Lucene106DedupHnswVectorsFormat extends KnnVectorsFormat { Review Comment: > So -- if two fields have _the same vectors_ we will collapse their storage, but build identical graphs twice? This would be an interesting way to measure HNSW noise -- index the same vector into N fields and then at search time test recall/CPU curve against them. And if one of them seems to always be better, use that one for your searching! -- 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]
