Performance: avoid sorting the hashes in MatchHashesAndScoreQuery (#724)

alexklibisz · Aug 29, 2024 · c22b52a · c22b52a
1 parent 9088ef5
commit c22b52a
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 13 deletions.
diff --git a/docs/pages/performance/fashion-mnist/plot.b64 b/docs/pages/performance/fashion-mnist/plot.b64
diff --git a/docs/pages/performance/fashion-mnist/plot.png b/docs/pages/performance/fashion-mnist/plot.png
diff --git a/docs/pages/performance/fashion-mnist/results.md b/docs/pages/performance/fashion-mnist/results.md
@@ -1,10 +1,10 @@
 |Model|Parameters|Recall|Queries per Second|
 |---|---|---|---|
-|eknn-l2lsh|L=100 k=4 w=1024 candidates=500 probes=0|0.379|380.371|
-|eknn-l2lsh|L=100 k=4 w=1024 candidates=1000 probes=0|0.447|305.831|
-|eknn-l2lsh|L=100 k=4 w=1024 candidates=500 probes=3|0.635|292.164|
-|eknn-l2lsh|L=100 k=4 w=1024 candidates=1000 probes=3|0.717|253.758|
-|eknn-l2lsh|L=100 k=4 w=2048 candidates=500 probes=0|0.767|330.408|
-|eknn-l2lsh|L=100 k=4 w=2048 candidates=1000 probes=0|0.847|278.242|
-|eknn-l2lsh|L=100 k=4 w=2048 candidates=500 probes=3|0.921|221.691|
-|eknn-l2lsh|L=100 k=4 w=2048 candidates=1000 probes=3|0.960|197.147|
+|eknn-l2lsh|L=100 k=4 w=1024 candidates=500 probes=0|0.378|373.943|
+|eknn-l2lsh|L=100 k=4 w=1024 candidates=1000 probes=0|0.447|322.600|
+|eknn-l2lsh|L=100 k=4 w=1024 candidates=500 probes=3|0.635|278.750|
+|eknn-l2lsh|L=100 k=4 w=1024 candidates=1000 probes=3|0.717|248.708|
+|eknn-l2lsh|L=100 k=4 w=2048 candidates=500 probes=0|0.767|328.214|
+|eknn-l2lsh|L=100 k=4 w=2048 candidates=1000 probes=0|0.847|291.762|
+|eknn-l2lsh|L=100 k=4 w=2048 candidates=500 probes=3|0.922|217.030|
+|eknn-l2lsh|L=100 k=4 w=2048 candidates=1000 probes=3|0.960|197.218|
diff --git a/elastiknn-lucene/src/main/java/org/apache/lucene/search/MatchHashesAndScoreQuery.java b/elastiknn-lucene/src/main/java/org/apache/lucene/search/MatchHashesAndScoreQuery.java
@@ -34,10 +34,6 @@ public MatchHashesAndScoreQuery(final String field,
                                     final int candidates,
                                     final IndexReader indexReader,
                                     final Function<LeafReaderContext, ScoreFunction> scoreFunctionBuilder) {
-        // `countMatches` expects hashes to be in sorted order.
-        // java's sort seems to be faster than lucene's ArrayUtil.
-        java.util.Arrays.sort(hashAndFrequencies, HashAndFreq::compareTo);
-
         this.field = field;
         this.hashAndFrequencies = hashAndFrequencies;
         this.candidates = candidates;