jimczi commented on code in PR #12582: URL: https://github.com/apache/lucene/pull/12582#discussion_r1334758274
########## lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99ScalarQuantizedVectorsWriter.java: ########## @@ -0,0 +1,851 @@ +/* + * 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.lucene99; + +import static org.apache.lucene.codecs.lucene99.Lucene99ScalarQuantizedVectorsFormat.DIRECT_MONOTONIC_BLOCK_SHIFT; +import static org.apache.lucene.codecs.lucene99.Lucene99ScalarQuantizedVectorsFormat.calculateDefaultQuantile; +import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.List; +import org.apache.lucene.codecs.CodecUtil; +import org.apache.lucene.codecs.KnnFieldVectorsWriter; +import org.apache.lucene.codecs.KnnVectorsReader; +import org.apache.lucene.codecs.KnnVectorsWriter; +import org.apache.lucene.codecs.lucene90.IndexedDISI; +import org.apache.lucene.codecs.perfield.PerFieldKnnVectorsFormat; +import org.apache.lucene.index.DocIDMerger; +import org.apache.lucene.index.DocsWithFieldSet; +import org.apache.lucene.index.FieldInfo; +import org.apache.lucene.index.FloatVectorValues; +import org.apache.lucene.index.IndexFileNames; +import org.apache.lucene.index.MergeState; +import org.apache.lucene.index.SegmentWriteState; +import org.apache.lucene.index.Sorter; +import org.apache.lucene.index.VectorEncoding; +import org.apache.lucene.index.VectorSimilarityFunction; +import org.apache.lucene.search.DocIdSetIterator; +import org.apache.lucene.store.IndexInput; +import org.apache.lucene.store.IndexOutput; +import org.apache.lucene.util.ArrayUtil; +import org.apache.lucene.util.IOUtils; +import org.apache.lucene.util.RamUsageEstimator; +import org.apache.lucene.util.ScalarQuantizer; +import org.apache.lucene.util.VectorUtil; +import org.apache.lucene.util.packed.DirectMonotonicWriter; + +/** + * Writes quantized vector values and metadata to index segments. + * + * @lucene.experimental + */ +public final class Lucene99ScalarQuantizedVectorsWriter implements QuantizedVectorsWriter { + + private static final long BASE_RAM_BYTES_USED = + RamUsageEstimator.shallowSizeOfInstance(Lucene99ScalarQuantizedVectorsWriter.class); + + private static final float QUANTIZATION_RECOMPUTE_LIMIT = 32; + private final SegmentWriteState segmentWriteState; + private final IndexOutput meta, quantizedVectorData; + private final Float quantile; + private final List<QuantizationVectorWriter> fields = new ArrayList<>(); + + private boolean finished; + + Lucene99ScalarQuantizedVectorsWriter(SegmentWriteState state, Float quantile) throws IOException { + this.quantile = quantile; + segmentWriteState = state; + String metaFileName = + IndexFileNames.segmentFileName( + state.segmentInfo.name, + state.segmentSuffix, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_META_EXTENSION); + + String quantizedVectorDataFileName = + IndexFileNames.segmentFileName( + state.segmentInfo.name, + state.segmentSuffix, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_EXTENSION); + + boolean success = false; + try { + meta = state.directory.createOutput(metaFileName, state.context); + quantizedVectorData = + state.directory.createOutput(quantizedVectorDataFileName, state.context); + + CodecUtil.writeIndexHeader( + meta, + Lucene99ScalarQuantizedVectorsFormat.META_CODEC_NAME, + Lucene99ScalarQuantizedVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + CodecUtil.writeIndexHeader( + quantizedVectorData, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_CODEC_NAME, + Lucene99ScalarQuantizedVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + success = true; + } finally { + if (success == false) { + IOUtils.closeWhileHandlingException(this); + } + } + } + + @Override + public KnnFieldVectorsWriter<float[]> addField(FieldInfo fieldInfo) throws IOException { + if (fieldInfo.getVectorEncoding() != VectorEncoding.FLOAT32) { + throw new IllegalArgumentException( + "Only float32 vector fields are supported for quantization"); + } + float quantile = + this.quantile == null + ? calculateDefaultQuantile(fieldInfo.getVectorDimension()) + : this.quantile; + QuantizationVectorWriter newField = QuantizationVectorWriter.create(fieldInfo, quantile); + fields.add(newField); + return newField; + } + + @Override + public void flush(int maxDoc, Sorter.DocMap sortMap) throws IOException { Review Comment: The main downside of this approach is that we expose the float and byte vector values as doc values in the reader/writer. That prevents any transformation in the codec because the API forces us to expose the original values in a sense. We need to decide whether exposing these values on the reader side is the expected behaviour. The API was not meant for lossy compression. -- 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-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
