msokolov commented on code in PR #12582: URL: https://github.com/apache/lucene/pull/12582#discussion_r1363796956
########## lucene/core/src/java/org/apache/lucene/util/ScalarQuantizer.java: ########## @@ -0,0 +1,267 @@ +/* + * 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.util; + +import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS; + +import java.io.IOException; +import java.util.Arrays; +import java.util.Random; +import java.util.stream.IntStream; +import org.apache.lucene.index.FloatVectorValues; +import org.apache.lucene.index.VectorSimilarityFunction; + +/** Will scalar quantize float vectors into `int8` byte values */ +public class ScalarQuantizer { + + public static final int SCALAR_QUANTIZATION_SAMPLE_SIZE = 25_000; + + private final float alpha; + private final float scale; + private final float minQuantile, maxQuantile, configuredQuantile; + + /** + * @param minQuantile the lower quantile of the distribution + * @param maxQuantile the upper quantile of the distribution + * @param configuredQuantile The configured quantile/confidence interval used to calculate the + * quantiles. + */ + public ScalarQuantizer(float minQuantile, float maxQuantile, float configuredQuantile) { + assert maxQuantile >= maxQuantile; + this.minQuantile = minQuantile; + this.maxQuantile = maxQuantile; + this.scale = 127f / (maxQuantile - minQuantile); + this.alpha = (maxQuantile - minQuantile) / 127f; + this.configuredQuantile = configuredQuantile; + } + + /** + * Quantize a float vector into a byte vector + * + * @param src the source vector + * @param dest the destination vector + * @param similarityFunction the similarity function used to calculate the quantile + * @return the corrective offset that needs to be applied to the score + */ + public float quantize(float[] src, byte[] dest, VectorSimilarityFunction similarityFunction) { + assert src.length == dest.length; + float correctiveOffset = 0f; + for (int i = 0; i < src.length; i++) { + float v = src[i]; + float dx = Math.max(minQuantile, Math.min(maxQuantile, src[i])) - minQuantile; + float dxs = scale * dx; + float dxq = Math.round(dxs) * alpha; + correctiveOffset += minQuantile * (v - minQuantile / 2.0F) + (dx - dxq) * dxq; + dest[i] = (byte) Math.round(dxs); + } + if (similarityFunction.equals(VectorSimilarityFunction.EUCLIDEAN)) { + return 0; + } + return correctiveOffset; + } + + /** + * Recalculate the old score corrective value given new current quantiles + * + * @param quantizedVector the old vector + * @param oldQuantizer the old quantizer + * @param similarityFunction the similarity function used to calculate the quantile + * @return the new offset + */ + public float recalculateCorrectiveOffset( + byte[] quantizedVector, + ScalarQuantizer oldQuantizer, + VectorSimilarityFunction similarityFunction) { + if (similarityFunction.equals(VectorSimilarityFunction.EUCLIDEAN)) { + return 0f; + } + // TODO this could probably be some simple algebra with the old offset + float correctiveOffset = 0f; + for (int i = 0; i < quantizedVector.length; i++) { + float v = (oldQuantizer.alpha * quantizedVector[i]) + oldQuantizer.minQuantile; + float dx = Math.max(minQuantile, Math.min(maxQuantile, v)) - minQuantile; + float dxs = scale * dx; + float dxq = Math.round(dxs) * alpha; + correctiveOffset += minQuantile * (v - minQuantile / 2.0F) + (dx - dxq) * dxq; + } + return correctiveOffset; + } + + /** + * Dequantize a byte vector into a float vector + * + * @param src the source vector + * @param dest the destination vector + */ + public void deQuantize(byte[] src, float[] dest) { + assert src.length == dest.length; + for (int i = 0; i < src.length; i++) { + dest[i] = (alpha * src[i]) + minQuantile; + } + } + + public float getLowerQuantile() { + return minQuantile; + } + + public float getUpperQuantile() { + return maxQuantile; + } + + public float getConfiguredQuantile() { + return configuredQuantile; + } + + public float getConstantMultiplier() { + return alpha * alpha; + } + + @Override + public String toString() { + return "ScalarQuantizer{" + + "minQuantile=" + + minQuantile + + ", maxQuantile=" + + maxQuantile + + ", configuredQuantile=" + + configuredQuantile + + '}'; + } + + private static final Random random = new Random(42); + + /** + * This will read the float vector values and calculate the quantiles. If the number of float + * vectors is less than {@link #SCALAR_QUANTIZATION_SAMPLE_SIZE} then all the values will be read + * and the quantiles calculated. If the number of float vectors is greater than {@link + * #SCALAR_QUANTIZATION_SAMPLE_SIZE} then a random sample of {@link + * #SCALAR_QUANTIZATION_SAMPLE_SIZE} will be read and the quantiles calculated. + * + * @param floatVectorValues the float vector values from which to calculate the quantiles + * @param quantile the quantile/confidence interval used to calculate the quantiles + * @return A new {@link ScalarQuantizer} instance + * @throws IOException if there is an error reading the float vector values + */ + public static ScalarQuantizer fromVectors(FloatVectorValues floatVectorValues, float quantile) + throws IOException { + assert 0.9f <= quantile && quantile <= 1f; + if (floatVectorValues.size() == 0) { + return new ScalarQuantizer(0f, 0f, quantile); + } + if (quantile == 1f) { + float min = Float.POSITIVE_INFINITY; + float max = Float.NEGATIVE_INFINITY; + while (floatVectorValues.nextDoc() != NO_MORE_DOCS) { + for (float v : floatVectorValues.vectorValue()) { + min = Math.min(min, v); + max = Math.max(max, v); + } + } + return new ScalarQuantizer(min, max, quantile); + } + int dim = floatVectorValues.dimension(); + if (floatVectorValues.size() < SCALAR_QUANTIZATION_SAMPLE_SIZE) { + int copyOffset = 0; + float[] values = new float[floatVectorValues.size() * dim]; + while (floatVectorValues.nextDoc() != NO_MORE_DOCS) { + float[] floatVector = floatVectorValues.vectorValue(); + System.arraycopy(floatVector, 0, values, copyOffset, floatVector.length); + copyOffset += dim; + } + float[] upperAndLower = getUpperAndLowerQuantile(values, quantile); + return new ScalarQuantizer(upperAndLower[0], upperAndLower[1], quantile); + } + int numFloatVecs = floatVectorValues.size(); + // Reservoir sample the vector ordinals we want to read + float[] values = new float[SCALAR_QUANTIZATION_SAMPLE_SIZE * dim]; + int[] vectorsToTake = IntStream.range(0, SCALAR_QUANTIZATION_SAMPLE_SIZE).toArray(); + for (int i = SCALAR_QUANTIZATION_SAMPLE_SIZE; i < numFloatVecs; i++) { + int j = random.nextInt(i + 1); + if (j < SCALAR_QUANTIZATION_SAMPLE_SIZE) { + vectorsToTake[j] = i; Review Comment: sorry I haven't been able to review in detail, but I am disturbed that we keep around full-precision data since rather than *reducing* storage that means this will actually *increase* storage. Given that, what is the purpose of the quantization? Is it that we get faster searches? Or is this intended to be a precursor to some more radical quantization like product space quantization? I feel I'm missing the purpose. -- 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
