Support queries with unsorted sparse bool query vecs (#134)

changelog.md

@@ -1,3 +1,5 @@
+- Support sparse bool query vectors with unsorted true indices.
+---
 - Added new submodules which can be used without Elasticsearch:
   - `com.klibisz.elastiknn:models` contains exact and approximate similarity models, all in Java with minimal dependencies.
   - `com.klibisz.elastiknn:lucene` contains the custom Lucene queries and some Lucene-related utilities used by Elastiknn.

elastiknn-api4s/src/main/scala/com/klibisz/elastiknn/api/package.scala

@@ -1,5 +1,6 @@
 package com.klibisz.elastiknn
 
+import scala.annotation.tailrec
 import scala.util.Random
 
 package object api {
@@ -28,10 +29,21 @@ package object api {
 
     final case class SparseBool(trueIndices: Array[Int], totalIndices: Int) extends Vec with KnownDims {
       def sorted(): SparseBool = copy(trueIndices.sorted)
+
+      def isSorted: Boolean = {
+        @tailrec
+        def check(i: Int): Boolean =
+          if (i == trueIndices.length) true
+          else if (trueIndices(i) < trueIndices(i - 1)) false
+          else check(i + 1)
+        check(1)
+      }
+
       override def equals(other: Any): Boolean = other match {
         case other: SparseBool => trueIndices.deep == other.trueIndices.deep && totalIndices == other.totalIndices
         case _                 => false
       }
+
       override def toString: String = s"SparseBool(${trueIndices.take(3).mkString(",")},...,${trueIndices.length}/$totalIndices)"
 
       def dims: Int = totalIndices

elastiknn-plugin/src/main/scala/com/klibisz/elastiknn/query/KnnQueryBuilder.scala

@@ -35,7 +35,12 @@ object KnnQueryBuilder {
       val map = parser.map()
       val json: Json = javaMapEncoder(map)
       val query = ElasticsearchCodec.decodeJsonGet[NearestNeighborsQuery](json)
-      new KnnQueryBuilder(query)
+      // Account for sparse bool vecs which need to be sorted.
+      val sortedVec = query.vec match {
+        case v: Vec.SparseBool if !v.isSorted => v.sorted()
+        case _                                => query.vec
+      }
+      new KnnQueryBuilder(query.withVec(sortedVec))
     }
   }
 

elastiknn-testing/src/test/scala/com/klibisz/elastiknn/query/NearestNeighborsQuerySpec.scala

@@ -145,4 +145,55 @@ class NearestNeighborsQuerySpec extends AsyncFunSpec with Matchers with Inspecto
     }
   }
 
+  // https://gitter.im/elastiknn/community?at=5f3012df65e829425e70ee31
+  describe("Sparse bool vectors with unsorted indices") {
+    implicit val rng: Random = new Random(0)
+    val indexPrefix = "test-sbv-unsorted"
+
+    val dims = 20000
+    val corpus = Vec.SparseBool.randoms(dims, 100)
+
+    val queryVec = {
+      val sorted = corpus.head
+      val shuffled = rng.shuffle(sorted.trueIndices.toVector).toArray
+      sorted.copy(shuffled)
+    }
+
+    // Test with multiple mappings/queries.
+    val mappingsAndQueries = Seq(
+      Mapping.SparseBool(dims) -> Seq(
+        NearestNeighborsQuery.Exact("vec", Similarity.Jaccard, queryVec),
+        NearestNeighborsQuery.Exact("vec", Similarity.Hamming, queryVec),
+      ),
+      Mapping.JaccardLsh(dims, 40, 1) -> Seq(
+        NearestNeighborsQuery.Exact("vec", Similarity.Jaccard, queryVec),
+        NearestNeighborsQuery.Exact("vec", Similarity.Hamming, queryVec),
+        NearestNeighborsQuery.JaccardLsh("vec", 100, queryVec)
+      ),
+      Mapping.HammingLsh(dims, 40, 2) -> Seq(
+        NearestNeighborsQuery.Exact("vec", Similarity.Jaccard, queryVec),
+        NearestNeighborsQuery.Exact("vec", Similarity.Hamming, queryVec),
+        NearestNeighborsQuery.HammingLsh("vec", 100, queryVec)
+      )
+    )
+
+    for {
+      (mapping, queries) <- mappingsAndQueries
+      query <- queries
+    } it(s"finds unsorted sparse bool vecs with mapping [${mapping}] and query [${query}]") {
+      val index = s"$indexPrefix-${UUID.randomUUID.toString}"
+      for {
+        _ <- deleteIfExists(index)
+        _ <- eknn.execute(createIndex(index).shards(1).replicas(1))
+        _ <- eknn.putMapping(index, "vec", "id", mapping)
+        _ <- eknn.index(index, "vec", corpus, "id", corpus.indices.map(i => s"v$i"))
+        _ <- eknn.execute(refreshIndex(index))
+        res <- eknn.nearestNeighbors(index, query, 5, "id")
+      } yield {
+        res.result.maxScore shouldBe 1d
+        res.result.hits.hits.head.id shouldBe "v0"
+      }
+    }
+  }
+
 }

version

@@ -1 +1 @@
-0.1.0-PRE30
+0.1.0-PRE31